feat(coding-agent): replace model registry with model runtime

Move provider auth and OAuth flows onto pi-ai Models, compose models.json and extension overlays through ModelRuntime, and retain ModelRegistry as an extension compatibility facade.

packages/agent/docs/models.md

@@ -68,7 +68,7 @@ packages/ai/src/
     openrouter-images.ts      # image-generation provider factory
     faux.ts                   # test provider factory
     all.ts                    # explicit aggregate: builtinModels(), builtinImagesModels(), getBuiltin*()
-  utils/oauth/                # OAuth flow implementations (node), lazy-loaded
+  auth/oauth/                 # Canonical OAuth implementations (node), lazy-loaded
 ```
 
 `src/index.ts` must stay core-only. It must not import:
@@ -407,7 +407,7 @@ export interface ApiKeyAuth {
   name: string; // "Anthropic API key"
 
   /** Interactive setup (prompt for key/provider env). Absent = ambient-only (env, ADC, IAM). */
-  login?(callbacks: AuthLoginCallbacks): Promise<ApiKeyCredential>;
+  login?(interaction: AuthInteraction): Promise<ApiKeyCredential>;
 
   /**
    * Resolve auth from the stored credential and/or ambient sources, merging
@@ -424,7 +424,7 @@ export interface ApiKeyAuth {
 export interface OAuthAuth {
   name: string; // "Anthropic (Claude Pro/Max)"
 
-  login(callbacks: AuthLoginCallbacks): Promise<OAuthCredential>;
+  login(interaction: AuthInteraction): Promise<OAuthCredential>;
 
   /** Exchange the refresh token. Network call; throws on failure (invalid_grant etc.). Runs under the store lock. */
   refresh(credential: OAuthCredential): Promise<OAuthCredential>;
@@ -445,9 +445,9 @@ export interface AuthContext {
 }
 ```
 
-The OAuth split (`refresh` + `toAuth` instead of one `resolve`) matches the old `OAuthProviderInterface` (`refreshToken` + `getApiKey`) and lets `Models` own the locking pattern without closure gymnastics: refresh produces a credential, `toAuth` derives request auth from whatever credential ends up stored.
+The `refresh`/`toAuth` split lets `Models` own the locked refresh pattern without closure gymnastics: refresh produces a credential, while `toAuth` derives request auth from whatever credential ends up stored.
 
-There is no `usesCallbackServer` flag. With `prompt()/notify()` callbacks the flow self-describes at runtime: a flow that runs a callback server issues a `manual_code` prompt racing the server and aborts the prompt when the callback wins. The UI needs no static foreknowledge.
+OAuth implementations use the provider-neutral `AuthInteraction` protocol directly. A callback-server flow issues a `manual_code` prompt racing the server and aborts the prompt when the callback wins, so the UI needs no provider-specific callback or static callback-server flag.
 
 ### Credentials
 
@@ -564,7 +564,7 @@ FileCredentialStore        ports AuthStorage's lock backend: read = memory snaps
    └─ withRuntimeOverrides --api-key
       └─ createModels({ credentials: store })
 
-login/logout UI            provider.auth.{oauth,apiKey}.login(callbacks) + store.modify/delete
+login/logout UI            provider.auth.{oauth,apiKey}.login(interaction) + store.modify/delete
 status UI                  store.read(pid) + getAuth try/catch ("needs /login" on rejection)
 getOAuthProviders          presence of provider.auth.oauth across registered providers
 ```
@@ -574,7 +574,7 @@ getOAuthProviders          presence of provider.auth.oauth across registered pro
 One interface serves api-key and OAuth login:
 
 ```ts
-export interface AuthLoginCallbacks {
+export interface AuthInteraction {
   /** Aborts the whole login flow. Per-prompt cancellation uses AuthPrompt.signal. */
   signal?: AbortSignal;
 
@@ -612,7 +612,7 @@ export function anthropicProvider(): Provider {
       apiKey: envApiKeyAuth("Anthropic API key", ["ANTHROPIC_API_KEY"]),
       oauth: lazyOAuth({
         name: "Anthropic (Claude Pro/Max)",
-        load: () => import("../utils/oauth/anthropic.ts").then((m) => m.anthropicOAuth),
+        load: () => import("../auth/oauth/anthropic.ts").then((m) => m.anthropicOAuth),
       }),
     },
     models: ANTHROPIC_MODELS,
@@ -632,7 +632,7 @@ export function lazyOAuth(input: {
 
 OAuth must not force Node-only code (`node:http`, `node:crypto`) into browser bundles: the dynamic import inside `lazyOAuth()` uses the same bundler-opaque variable-specifier trick as the bedrock lazy wrapper. Browser hosts never trigger the load (no stored node OAuth credentials, no login flow). If web OAuth lands later (sitegeist proved feasibility: Web Crypto PKCE, auth tab, fetch token exchange, device-code polling), it is just a different `OAuthAuth` implementation — no reserved option values.
 
-The existing flows in `src/utils/oauth/` (anthropic, openai-codex, github-copilot) are adapted to `OAuthAuth` (`login`/`refresh`/`toAuth`, replacing `login`/`refreshToken`/`getApiKey`/`modifyModels`) with the new callbacks, staying Node-targeted and lazy-loaded. Copilot's `modifyModels` baseUrl rewriting becomes `toAuth` returning `ModelAuth.baseUrl`.
+The built-in flows in `src/auth/oauth/` implement `OAuthAuth` and `AuthInteraction` directly while remaining Node-targeted and lazy-loaded. Copilot derives its credential-specific request endpoint through `toAuth().baseUrl`.
 
 ## Provider wrappers and models.json
 
@@ -817,7 +817,7 @@ Check items off as they land. Keep this list current; it is the working state fo
 
 ### Phase 3 — provider factories + catalogs
 
-- [x] Auth helpers in `src/auth/helpers.ts`: `envApiKeyAuth()` (with secret-prompt `login`), `lazyOAuth()`. OAuth flow loads go through `utils/oauth/load.ts` (bundler-opaque dynamic import); the `OAuthAuth` exports it references land in Phase 4.
+- [x] Auth helpers in `src/auth/helpers.ts`: `envApiKeyAuth()` (with secret-prompt `login`), `lazyOAuth()`. OAuth flow loads go through `auth/oauth/load.ts` (bundler-opaque dynamic import); the `OAuthAuth` exports it references land in Phase 4.
 - [x] `createProvider()` in `models.ts` (single + mixed `api` map, dispatch on `model.api`, unknown api -> stream error).
 - [x] Per-provider factories under `src/providers/` for all built-in catalog providers; OAuth attached via `lazyOAuth()` (anthropic, openai-codex, github-copilot); ambient `ApiKeyAuth` for amazon-bedrock (AWS env/profile) and google-vertex (key or ADC+project+location).
 - [x] `providers/all.ts`: `builtinProviders()`, `builtinModels()`, `getBuiltinModel/getBuiltinModels/getBuiltinProviders` re-exports.
@@ -826,8 +826,8 @@ Check items off as they land. Keep this list current; it is the working state fo
 
 ### Phase 4 — OAuth adaptation
 
-- [x] Adapt `utils/oauth/anthropic.ts`, `openai-codex.ts`, `github-copilot.ts` to `OAuthAuth` (`login`/`refresh`/`toAuth`) + `prompt()/notify()`; `modifyModels` baseUrl rewriting becomes `toAuth().baseUrl`. New exports (`anthropicOAuth`, `openaiCodexOAuth`, `githubCopilotOAuth`) sit next to the old `OAuthProviderInterface` objects, which survive until Phase 7.
-- [x] No `usesCallbackServer` on `OAuthAuth`: callback-server flows race a `manual_code` prompt (aborted via `AuthPrompt.signal` once the flow settles). The old interface keeps its flag until it dies with compat.
+- [x] Built-in implementations live under `auth/oauth/` and implement `OAuthAuth` directly through `AuthInteraction.prompt()`/`notify()`. They are private provider implementations loaded lazily by provider factories.
+- [x] Callback-server flows race a `manual_code` prompt, aborted through `AuthPrompt.signal` once the flow settles. The public `oauth` subpath retains only coding-agent extension compatibility types.
 
 ### Phase 5 — packaging
 
@@ -876,7 +876,7 @@ Decisions:
 - Runtime `--api-key` overrides are an explicit store overlay (an override reads as an ephemeral stored api-key credential, masking stored OAuth — matches today's priority). Every registered provider is guaranteed an `apiKey` auth slot so overrides apply to OAuth-only providers too.
 - `ModelRegistry.getAll`/`find`/`getAvailable` stay sync for SDK and extension compatibility, delegating to the collection's last-known sync model lists and fast configured-looking status checks. Dynamic providers update through explicit async `refresh()`, and request auth remains async through `getApiKeyAndHeaders()`/`Models.getAuth()`. Extensions also get the collection itself as the forward API.
 - models.json keeps FULL feature parity, implemented as provider decoration: builtin factories wrapped so `getModels()` applies provider `baseUrl`/`compat` overlays, `modelOverrides`, and custom-model merges (async-safe); provider `apiKey`/`headers`/`authHeader` configs become that provider's `ApiKeyAuth` (config first, factory auth fallback); parse errors keep `getError()` semantics.
-- Extension `ProviderConfig` parity: provider-keyed `streamSimple`, old-style `oauth` adapted to `OAuthAuth` (`modifyModels` -> `getModels` wrap + `toAuth`), full model replacement per provider. Legacy `registerApiProvider` writes stay compat-local for consumers that call global `complete()`; they die with compat.
+- Extension `ProviderConfig` parity: provider-keyed `streamSimple`, legacy extension OAuth callbacks adapted to `OAuthAuth`, and full model replacement per provider. Legacy `registerApiProvider` writes stay compat-local for consumers that call global `complete()`; they die with compat.
 - Copilot: stored-credential baseUrl applied in the wrapped `getModels()` (extension-visible models stay correct) plus per-request `toAuth().baseUrl`.
 - Cloudflare: provider-auth substitution (key + `CLOUDFLARE_ACCOUNT_ID`/`CLOUDFLARE_GATEWAY_ID` from credential `env` or ambient `AuthContext.env()` -> `ModelAuth.baseUrl`). Built-in compat calls route through `Models`, so they use the same provider auth path.
 
@@ -909,13 +909,11 @@ Ordering for new sessions:
    - Wrap Copilot's provider `getModels()` when an OAuth credential is present so extension/UI-visible model metadata also carries the authenticated account base URL.
    - Keep API-key/env-token Copilot behavior unchanged.
    - Add tests for model metadata before login, after OAuth credential, after refresh/baseUrl change, and logout.
-7. [ ] Extension OAuth adapter.
-   - Adapt old extension `OAuthProviderInterface` configs to pi-ai `OAuthAuth`.
-   - `login` maps old callbacks/events to `prompt()/notify()`.
-   - `refreshToken` maps to `refresh`.
-   - `getApiKey` maps to `toAuth`.
-   - `modifyModels` becomes a provider `getModels()` wrapper plus `toAuth().baseUrl` where applicable.
-   - Preserve existing extension runtime compatibility through the `/compat` alias until Phase 10.
+7. [x] Extension OAuth adapter.
+   - Keep only the legacy callback/credential declarations required by coding-agent `ProviderConfig.oauth`.
+   - `login` maps legacy callbacks/events to `AuthInteraction.prompt()`/`notify()`.
+   - `refreshToken` maps to `refresh`; `getApiKey` maps to `toAuth`.
+   - Preserve the type-only pi-ai `oauth` barrel and extension-loader aliases.
 8. [ ] Rebuild coding-agent `ModelRegistry` over `MutableModels`.
    - It owns a `MutableModels` instance built from decorated built-ins + models.json custom providers + extension providers.
    - `getAll()`, `find()`, and `getAvailable()` remain sync compatibility methods over last-known model lists and fast configured-looking auth status. Do not break the extension-facing `modelRegistry` surface for these reads.
@@ -939,7 +937,7 @@ Ordering for new sessions:
 
 - [ ] AgentSession -> AgentHarness; the registry facade dies in favor of harness `Models`.
 - [ ] Move ALL internal `/compat` imports to the new API: every package's src, all tests, and the example extensions (examples then demonstrate the new API). Nothing inside the repo may import `/compat` at that point.
-- [ ] Delete `/compat`, `env-api-keys.ts`, the extension-loader root-to-compat alias, the old `pi-ai/oauth` registry and `OAuthProviderInterface` (incl. `usesCallbackServer`), and the compat-local legacy API registry. This is the extension-author breaking release; changelog carries the migration guide.
+- [ ] Delete `/compat`, `env-api-keys.ts`, the extension-loader root-to-compat alias, and the compat-local legacy API registry. The old OAuth registry/provider interface is already gone; the type-only `oauth` barrel remains for extension compatibility.
 
 ### Deferred / follow-ups
 

packages/ai/CHANGELOG.md

@@ -2,13 +2,60 @@
 
 ## [Unreleased]
 
+### Breaking Changes
+
+- Changed runtime authentication to provider-scoped `Models.checkAuth()`, `getAuth()`, `login()`, and `logout()` APIs. `checkAuth()` now returns `AuthCheck | undefined`, and API-key auth resolvers no longer receive a model.
+- Removed the legacy built-in OAuth provider objects, global OAuth registry APIs, and public low-level built-in login/refresh functions. Use canonical `Provider.auth.oauth` methods instead; the `oauth` subpath now retains only extension compatibility types.
+- Renamed the canonical login interaction interface from `AuthLoginCallbacks` to `AuthInteraction`; it exposes the provider-neutral `prompt()`/`notify()` protocol used by API-key and OAuth flows.
+- Changed the `Models` request contract: `getAuth(model)` now includes model headers, while `getAuth(providerId)` remains provider-scoped, and Models stream options may include `transformHeaders`. Custom `Models` implementations must execute the transform after merging auth/model and explicit headers, then remove it before provider dispatch.
+
+#### Models migration
+
+Use the `getAuth()` argument to choose the desired scope:
+
+```typescript
+// Provider auth only
+const providerAuth = await models.getAuth(model.provider);
+
+// Provider auth plus model.headers
+const modelAuth = await models.getAuth(model);
+```
+
+Use the Models-only `transformHeaders` option instead of resolving auth before streaming. It runs once on assembled headers and is not passed to `Provider.stream*()`:
+
+```typescript
+models.streamSimple(model, context, {
+  transformHeaders: async (headers) => ({
+    ...headers,
+    "X-Request-ID": requestId,
+  }),
+});
+```
+
+Custom `Models` implementations must apply the same order:
+
+```text
+getAuth(model) -> explicit options.headers -> transformHeaders -> Provider.stream*()
+```
+
+`Provider.stream*()` continues to accept ordinary `ApiStreamOptions`/`SimpleStreamOptions`; providers do not handle `transformHeaders`.
+
 ### Added
 
+- Added provider-owned authentication and availability resolution to `Models`, including stored OAuth refresh and interactive login support through `CredentialStore`.
+- Added async non-secret credential enumeration through `CredentialStore.list()` and credential-aware `Provider.filterModels()` availability policy.
+- Added neutral auth-flow information/link events and provider-owned Amazon Bedrock and Google Vertex AI credential selection flows.
 - Added a separate opt-in `max` thinking level, including native `xhigh` and `max` support for GPT-5.6 and Anthropic adaptive-thinking effort metadata matching Anthropic's documentation: `max` on all adaptive Claude models, native `xhigh` on Opus 4.7/4.8, Sonnet 5, and Fable 5 only.
 - Added request-wide input-token pricing tiers to model cost metadata and usage cost calculation.
 
+### Changed
+
+- Changed `Models.getAuth(model)` to include model headers and added a Models-only `transformHeaders` stream option that runs after auth and explicit header assembly but is not forwarded to providers.
+
 ### Fixed
 
+- Fixed Cloudflare Workers AI and AI Gateway streams to materialize account and gateway endpoint placeholders after auth resolution, including compat streaming with custom model objects.
+- Fixed lazy provider streams to preserve their final assistant message when forwarding an inner stream.
 - Fixed post-compaction output-token budgeting to ignore stale assistant usage from before the compaction boundary ([#6464](https://github.com/earendil-works/pi/issues/6464)).
 - Fixed GPT-5.4 and GPT-5.5 long-context cost accounting while retaining the intentional 272K default context limit for models that require an explicit override.
 - Fixed GPT-5.6 metadata to keep direct OpenAI requests in the 272K short-context tier while exposing the Codex backend's 372K context window with long-context pricing.

packages/ai/README.md

@@ -17,6 +17,7 @@ Unified LLM API with provider collections, automatic auth resolution, token and
   - [Dynamic Providers](#dynamic-providers)
 - [Auth](#auth)
   - [How Auth Resolves](#how-auth-resolves)
+  - [Transforming Request Headers](#transforming-request-headers)
   - [Credential Store](#credential-store)
   - [Environment Variables](#environment-variables)
 - [Tools](#tools)
@@ -334,18 +335,45 @@ await models.complete(model, context);
 await models.complete(model, context, { apiKey: 'sk-explicit' });
 ```
 
-You can inspect resolution without making a request — useful for status UIs:
+You can inspect resolution without making a request. Pass a provider ID for provider-scoped auth, or a model to include its static `model.headers`:
 
 ```typescript
-const auth = await models.getAuth(model);
-if (auth) {
-  console.log(`configured via ${auth.source}`); // e.g. "ANTHROPIC_API_KEY", "OAuth", "stored credential"
+const providerAuth = await models.getAuth(model.provider);
+const modelAuth = await models.getAuth(model);
+
+if (modelAuth) {
+  console.log(`configured via ${modelAuth.source}`); // e.g. "ANTHROPIC_API_KEY", "OAuth", "stored credential"
+  console.log(modelAuth.auth.headers);              // Provider auth headers + model.headers
 } else {
   console.log('not configured');
 }
 ```
 
-`getAuth()` resolves `undefined` for unconfigured providers and rejects with `ModelsError` when something is actually broken (`"oauth"`: token refresh failed, credential preserved for re-login; `"auth"`: key resolution or credential store failure). Request paths surface the same failures as stream errors.
+Both overloads resolve credentials, refresh expired OAuth when necessary, and may return an auth-derived `apiKey`, `headers`, or `baseUrl`. `getAuth()` resolves `undefined` for unconfigured providers and rejects with `ModelsError` when something is actually broken (`"oauth"`: token refresh failed, credential preserved for re-login; `"auth"`: key resolution or credential store failure). Request paths surface the same failures as stream errors.
+
+### Transforming Request Headers
+
+`Models.stream()`, `complete()`, `streamSimple()`, and `completeSimple()` accept a Models-only `transformHeaders` option. It runs once after provider auth, `model.headers`, and explicit `options.headers` have been merged, but before provider dispatch:
+
+```typescript
+const response = await models.completeSimple(model, context, {
+  headers: { "X-Client": "my-app" },
+  transformHeaders: async (headers) => ({
+    ...headers,
+    "X-Request-ID": crypto.randomUUID(),
+  }),
+});
+```
+
+The ordering is:
+
+```text
+provider auth headers -> model.headers -> explicit options.headers -> transformHeaders -> Provider.stream*()
+```
+
+Header names are merged case-insensitively. Explicit headers override auth/model headers, and the transform has final control; returning `null` for a header suppresses lower-level defaults that support deletion.
+
+`transformHeaders` belongs to `Models`, not `Provider`. A `Models` implementation must consume it and remove it before calling `Provider.stream*()`. Provider implementations continue receiving ordinary `ApiStreamOptions` or `SimpleStreamOptions` and never handle the transform themselves. Use this option instead of calling `getAuth(model)` before `stream*()`, which would resolve request auth twice.
 
 ### Credential Store
 
@@ -359,7 +387,7 @@ const models = createModels({ credentials: myFileBackedStore });
 // const models = builtinModels({ credentials: myFileBackedStore });
 ```
 
-The contract is small: `read(providerId)`, `modify(providerId, fn)` (the only write path — a serialized read-modify-write), and `delete(providerId)`. OAuth token refresh runs inside `modify`, so concurrent requests and processes cannot double-refresh a rotated token. A stored credential *owns* its provider: environment variables are only consulted when nothing is stored, and a failed refresh never silently falls back to an env key.
+The contract is small: `read(providerId)`, `list()` for non-secret `{ providerId, type }` metadata, `modify(providerId, fn)` (the only write path — a serialized read-modify-write), and `delete(providerId)`. Enumeration must not resolve secrets or execute configured key commands. OAuth token refresh runs inside `modify`, so concurrent requests and processes cannot double-refresh a rotated token. A stored credential *owns* its provider: environment variables are only consulted when nothing is stored, and a failed refresh never silently falls back to an env key.
 
 API-key credentials use the same discriminator as pi's `auth.json` and can carry provider-scoped env/config values:
 
@@ -412,7 +440,7 @@ Built-in providers resolve these env vars (Node.js; in browsers pass `apiKey` ex
 | Xiaomi MiMo Token Plan (Singapore) | `XIAOMI_TOKEN_PLAN_SGP_API_KEY` |
 | GitHub Copilot | `COPILOT_GITHUB_TOKEN` |
 
-Amazon Bedrock resolves ambient AWS credentials (`AWS_PROFILE`, access key pairs, `AWS_BEARER_TOKEN_BEDROCK`, ECS task roles, web identity tokens). Vertex AI resolves either an explicit key or gcloud Application Default Credentials plus project/location.
+Amazon Bedrock resolves ambient AWS credentials (`AWS_PROFILE`, access key pairs, `AWS_BEARER_TOKEN_BEDROCK`, ECS task roles, web identity tokens); its provider-owned login flow supports bearer tokens, AWS profiles, and the existing credential chain. Vertex AI resolves either an explicit key or gcloud Application Default Credentials plus project/location, with a provider-owned login flow for API keys, ADC, and service-account files.
 
 ## Tools
 
@@ -660,7 +688,7 @@ for (const block of result.output) {
 }
 ```
 
-Like the chat side, you can build the collection from parts: `createImagesModels({ credentials?, authContext? })`, the `openrouterImagesProvider()` factory from `@earendil-works/pi-ai/providers/openrouter-images`, and `createImagesProvider({ id, auth, models, refreshModels?, api })` for custom image providers (with `imagesModels.refresh(provider?)` for dynamic lists). Failures never reject — they return an `AssistantImages` with `stopReason: "error"`. The collection's `getAuth(model)` works exactly like the chat-side one.
+Like the chat side, you can build the collection from parts: `createImagesModels({ credentials?, authContext? })`, the `openrouterImagesProvider()` factory from `@earendil-works/pi-ai/providers/openrouter-images`, and `createImagesProvider({ id, auth, models, refreshModels?, api })` for custom image providers (with `imagesModels.refresh(provider?)` for dynamic lists). Failures never reject — they return an `AssistantImages` with `stopReason: "error"`. The collection's provider-scoped `getAuth(providerId)` works exactly like the chat-side one.
 
 The old global API (`getImageModel()` / `getImageModels()` / `getImageProviders()` / `generateImages()`) remains available on the [compat entrypoint](#migrating-from-the-old-global-api):
 
@@ -982,6 +1010,25 @@ const gateway = createProvider({
 });
 ```
 
+Provider-wide endpoint or request transformations belong in the provider's API implementation: wrap the `ProviderStreams` you pass as `api` so every request goes through the transformation before dispatch. The Cloudflare providers do this to materialize account/gateway endpoint placeholders from the resolved provider env:
+
+```typescript
+function tenantStreams(streams: ProviderStreams): ProviderStreams {
+  const withTenant = (model: Model<Api>) => ({ ...model, baseUrl: model.baseUrl.replace('{tenant}', tenantId) });
+  return {
+    stream: (model, context, options) => streams.stream(withTenant(model), context, options),
+    streamSimple: (model, context, options) => streams.streamSimple(withTenant(model), context, options),
+  };
+}
+
+const tenantGateway = createProvider({
+  id: 'tenant-gateway',
+  auth: { apiKey: envApiKeyAuth('Gateway key', ['GATEWAY_API_KEY']) },
+  models: [/* ... */],
+  api: tenantStreams(openAICompletionsApi()),
+});
+```
+
 Dynamic model lists use `refreshModels`; the provider lists empty until the first `models.refresh()`:
 
 ```typescript
@@ -997,7 +1044,7 @@ models.setProvider(llamacpp);
 await models.refresh('llamacpp');
 ```
 
-Custom models can carry `headers` (e.g. proxies behind bot detection) and `compat` flags — see [OpenAI Compatibility Settings](#openai-compatibility-settings).
+Custom models can carry `headers` (e.g. proxies behind bot detection) and `compat` flags. `Models.getAuth(model)` includes those model headers, and stream methods merge them before explicit request headers and `transformHeaders`. See [OpenAI Compatibility Settings](#openai-compatibility-settings).
 
 Some OpenAI-compatible servers do not understand the `developer` role used for reasoning-capable models. For those providers, set `compat.supportsDeveloperRole` to `false` so the system prompt is sent as a `system` message instead. If the server also does not support `reasoning_effort`, set `compat.supportsReasoningEffort` to `false` too. This commonly applies to Ollama, vLLM, SGLang, and similar OpenAI-compatible servers.
 
@@ -1368,7 +1415,7 @@ Several providers support OAuth authentication instead of static API keys:
 - **OpenAI Codex** (ChatGPT Plus/Pro subscription, access to GPT-5.x Codex models)
 - **GitHub Copilot** (Copilot subscription)
 
-Each of these providers carries an `OAuthAuth` on `provider.auth.oauth` with three operations: `login(callbacks)` runs the interactive flow and returns a credential, `refresh(credential)` exchanges the refresh token, and `toAuth(credential)` derives request auth (GitHub Copilot's per-account base URL comes from here). Refresh is automatic: `models.getAuth()` and the request paths refresh expired tokens under a credential-store lock, so concurrent requests and processes cannot double-refresh.
+Each of these providers carries an `OAuthAuth` on `provider.auth.oauth` with three operations: `login(interaction)` uses the provider-neutral `AuthInteraction.prompt()`/`notify()` protocol and returns a credential, `refresh(credential)` exchanges the refresh token, and `toAuth(credential)` derives request auth (GitHub Copilot's per-account base URL comes from here). Refresh is automatic: `models.getAuth(providerId)` and request paths refresh expired tokens under a credential-store lock, so concurrent requests and processes cannot double-refresh.
 
 ```typescript
 import { createModels } from '@earendil-works/pi-ai';
@@ -1377,34 +1424,36 @@ import { anthropicProvider } from '@earendil-works/pi-ai/providers/anthropic';
 const models = createModels({ credentials: myStore }); // persistent CredentialStore
 models.setProvider(anthropicProvider());
 
-// Login: drive the flow with prompt()/notify() callbacks, persist the credential
-const provider = models.getProvider('anthropic')!;
-const credential = await provider.auth.oauth!.login({
+// Login: Models drives the flow and persists the credential
+await models.login('anthropic', 'oauth', {
   prompt: async (p) => {
     // p.type: 'text' | 'secret' | 'select' | 'manual_code'
     // manual_code prompts race a local callback server; p.signal aborts them when the server wins
     return await askUser(p.message);
   },
   notify: (event) => {
-    // event.type: 'auth_url' | 'device_code' | 'progress'
+    // event.type: 'info' | 'auth_url' | 'device_code' | 'progress'
+    if (event.type === 'info') {
+      console.log(event.message);
+      for (const link of event.links ?? []) console.log(`${link.label ?? 'More information'}: ${link.url}`);
+    }
     if (event.type === 'auth_url') console.log(`Open: ${event.url}`);
     if (event.type === 'device_code') console.log(`Code: ${event.userCode} at ${event.verificationUri}`);
     if (event.type === 'progress') console.log(event.message);
   },
 });
-await myStore.modify('anthropic', async () => credential);
 
 // From here on, requests resolve and refresh the token automatically
 const model = models.getModel('anthropic', 'claude-sonnet-4-5')!;
 await models.complete(model, context);
 
 // Logout
-await myStore.delete('anthropic');
+await models.logout('anthropic');
 ```
 
 ### Vertex AI
 
-Vertex AI models support either a Google Cloud API key or Application Default Credentials (ADC):
+Vertex AI models support either a Google Cloud API key or Application Default Credentials (ADC). Its provider-owned API-key login flow can configure either method:
 
 - **API key**: Set `GOOGLE_CLOUD_API_KEY` or pass `apiKey` in the call options.
 - **Local development (ADC)**: Run `gcloud auth application-default login`
@@ -1438,7 +1487,7 @@ Credentials are saved to `auth.json` in the current directory.
 
 ### Programmatic OAuth
 
-The legacy flow functions remain available via the `@earendil-works/pi-ai/oauth` entry point (`loginAnthropic`, `loginOpenAICodex`, `loginGitHubCopilot`, `refreshOAuthToken`, `getOAuthApiKey`); credential storage is the caller's responsibility there. New code should prefer the provider-owned `OAuthAuth` shown above — it composes with the credential store and gets locked auto-refresh for free.
+Built-in login and refresh flows are private provider implementations. Use provider-owned `OAuthAuth`, which composes with `CredentialStore` and gets locked auto-refresh through `Models`. The `@earendil-works/pi-ai/oauth` entry point retains only type declarations required by coding-agent extension OAuth compatibility.
 
 Provider notes:
 
@@ -1468,7 +1517,7 @@ Compat is a strict superset of the root entrypoint, so a file can switch its imp
 | `getModels('anthropic')` / `getProviders()` | `models.getModels('anthropic')` / `models.getProviders()` or `getBuiltin*` |
 | `stream(model, ctx, opts)` (env-key injection) | `models.stream(model, ctx, opts)` (provider auth resolution) |
 | `registerApiProvider({ api, stream, streamSimple })` | `createProvider({ id, auth, models, api })` + `models.setProvider()` |
-| `getEnvApiKey('openai')` | `await models.getAuth(model)` |
+| `getEnvApiKey('openai')` | `await models.getAuth(model.provider)` |
 | `streamAnthropic(model, ctx, opts)` | `stream` from `@earendil-works/pi-ai/api/anthropic-messages`, or a provider in a collection |
 | `registerFauxProvider()` | `fauxProvider()` + `models.setProvider()` |
 

packages/ai/src/api/lazy.ts

@@ -22,13 +22,20 @@ function createSetupErrorMessage(model: Model<Api>, error: unknown): AssistantMe
 	};
 }
 
-function forwardStream(target: AssistantMessageEventStream, source: AsyncIterable<AssistantMessageEvent>): void {
-	(async () => {
-		for await (const event of source) {
-			target.push(event);
-		}
-		target.end();
-	})();
+function hasResult(
+	source: AsyncIterable<AssistantMessageEvent>,
+): source is AsyncIterable<AssistantMessageEvent> & { result(): Promise<AssistantMessage> } {
+	return typeof (source as { result?: unknown }).result === "function";
+}
+
+async function forwardStream(
+	target: AssistantMessageEventStream,
+	source: AsyncIterable<AssistantMessageEvent>,
+): Promise<void> {
+	for await (const event of source) {
+		target.push(event);
+	}
+	target.end(hasResult(source) ? await source.result() : undefined);
 }
 
 /**
@@ -43,9 +50,7 @@ export function lazyStream(
 	const outer = new AssistantMessageEventStream();
 
 	setup()
-		.then((inner) => {
-			forwardStream(outer, inner);
-		})
+		.then((inner) => forwardStream(outer, inner))
 		.catch((error) => {
 			const message = createSetupErrorMessage(model, error);
 			outer.push({ type: "error", reason: "error", error: message });

packages/ai/src/auth/credential-store.ts

@@ -1,4 +1,4 @@
-import type { Credential, CredentialStore } from "./types.ts";
+import type { Credential, CredentialInfo, CredentialStore } from "./types.ts";
 
 /**
  * Default in-memory credential store. Apps inject persistent stores.
@@ -27,6 +27,10 @@ export class InMemoryCredentialStore implements CredentialStore {
 		return this.credentials.get(providerId);
 	}
 
+	async list(): Promise<readonly CredentialInfo[]> {
+		return [...this.credentials].map(([providerId, credential]) => ({ providerId, type: credential.type }));
+	}
+
 	modify(
 		providerId: string,
 		fn: (current: Credential | undefined) => Promise<Credential | undefined>,

packages/ai/src/auth/helpers.ts

@@ -9,8 +9,8 @@ import type { ApiKeyAuth, OAuthAuth } from "./types.ts";
 export function envApiKeyAuth(name: string, envVars: readonly string[]): ApiKeyAuth {
 	return {
 		name,
-		login: async (callbacks) => {
-			const key = await callbacks.prompt({ type: "secret", message: `Enter ${name}` });
+		login: async (interaction) => {
+			const key = await interaction.prompt({ type: "secret", message: `Enter ${name}` });
 			return { type: "api_key", key };
 		},
 		resolve: async ({ ctx, credential }) => {
@@ -39,7 +39,7 @@ export function lazyOAuth(input: { name: string; load: () => Promise<OAuthAuth>
 	};
 	return {
 		name: input.name,
-		login: async (callbacks) => (await loaded()).login(callbacks),
+		login: async (interaction) => (await loaded()).login(interaction),
 		refresh: async (credential) => (await loaded()).refresh(credential),
 		toAuth: async (credential) => (await loaded()).toAuth(credential),
 	};

packages/ai/src/auth/oauth/anthropic.ts

@@ -6,11 +6,10 @@
  */
 
 import type { Server } from "node:http";
-import type { OAuthAuth } from "../../auth/types.ts";
-import { getProviderEnvValue } from "../provider-env.ts";
+import { getProviderEnvValue } from "../../utils/provider-env.ts";
+import type { AuthInteraction, OAuthAuth, OAuthCredential } from "../types.ts";
 import { oauthErrorHtml, oauthSuccessHtml } from "./oauth-page.ts";
 import { generatePKCE } from "./pkce.ts";
-import type { OAuthCredentials, OAuthLoginCallbacks, OAuthPrompt, OAuthProviderInterface } from "./types.ts";
 
 type CallbackServerInfo = {
 	server: Server;
@@ -193,7 +192,7 @@ async function exchangeAuthorizationCode(
 	state: string,
 	verifier: string,
 	redirectUri: string,
-): Promise<OAuthCredentials> {
+): Promise<OAuthCredential> {
 	let responseBody: string;
 	try {
 		responseBody = await postJson(TOKEN_URL, {
@@ -220,27 +219,21 @@ async function exchangeAuthorizationCode(
 	}
 
 	return {
+		type: "oauth",
 		refresh: tokenData.refresh_token,
 		access: tokenData.access_token,
 		expires: Date.now() + tokenData.expires_in * 1000 - 5 * 60 * 1000,
 	};
 }
 
-/**
- * Login with Anthropic OAuth (authorization code + PKCE)
- */
-export async function loginAnthropic(options: {
-	onAuth: (info: { url: string; instructions?: string }) => void;
-	onPrompt: (prompt: OAuthPrompt) => Promise<string>;
-	onProgress?: (message: string) => void;
-	onManualCodeInput?: () => Promise<string>;
-}): Promise<OAuthCredentials> {
+async function loginAnthropic(interaction: AuthInteraction): Promise<OAuthCredential> {
 	const { verifier, challenge } = await generatePKCE();
 	const server = await startCallbackServer(verifier);
-
+	const manualAbort = new AbortController();
 	let code: string | undefined;
 	let state: string | undefined;
-	let redirectUriForExchange = REDIRECT_URI;
+	let manualInput: string | undefined;
+	let manualError: Error | undefined;
 
 	try {
 		const authParams = new URLSearchParams({
@@ -253,93 +246,58 @@ export async function loginAnthropic(options: {
 			code_challenge_method: "S256",
 			state: verifier,
 		});
-
-		options.onAuth({
+		interaction.notify({
+			type: "auth_url",
 			url: `${AUTHORIZE_URL}?${authParams.toString()}`,
 			instructions:
 				"Complete login in your browser. If the browser is on another machine, paste the final redirect URL here.",
 		});
 
-		if (options.onManualCodeInput) {
-			let manualInput: string | undefined;
-			let manualError: Error | undefined;
-			const manualPromise = options
-				.onManualCodeInput()
-				.then((input) => {
-					manualInput = input;
-					server.cancelWait();
-				})
-				.catch((err) => {
-					manualError = err instanceof Error ? err : new Error(String(err));
-					server.cancelWait();
-				});
-
-			const result = await server.waitForCode();
-
-			if (manualError) {
-				throw manualError;
-			}
-
-			if (result?.code) {
-				code = result.code;
-				state = result.state;
-				redirectUriForExchange = REDIRECT_URI;
-			} else if (manualInput) {
-				const parsed = parseAuthorizationInput(manualInput);
-				if (parsed.state && parsed.state !== verifier) {
-					throw new Error("OAuth state mismatch");
-				}
-				code = parsed.code;
-				state = parsed.state ?? verifier;
-			}
-
-			if (!code) {
-				await manualPromise;
-				if (manualError) {
-					throw manualError;
-				}
-				if (manualInput) {
-					const parsed = parseAuthorizationInput(manualInput);
-					if (parsed.state && parsed.state !== verifier) {
-						throw new Error("OAuth state mismatch");
-					}
-					code = parsed.code;
-					state = parsed.state ?? verifier;
-				}
-			}
-		} else {
-			const result = await server.waitForCode();
-			if (result?.code) {
-				code = result.code;
-				state = result.state;
-				redirectUriForExchange = REDIRECT_URI;
-			}
-		}
-
-		if (!code) {
-			const input = await options.onPrompt({
-				message: "Paste the authorization code or full redirect URL:",
+		const manualPromise = interaction
+			.prompt({
+				type: "manual_code",
+				message: "Complete login in your browser, or paste the authorization code / redirect URL here:",
 				placeholder: REDIRECT_URI,
+				signal: manualAbort.signal,
+			})
+			.then((input) => {
+				manualInput = input;
+				server.cancelWait();
+			})
+			.catch((error) => {
+				manualError = error instanceof Error ? error : new Error(String(error));
+				server.cancelWait();
 			});
-			const parsed = parseAuthorizationInput(input);
-			if (parsed.state && parsed.state !== verifier) {
-				throw new Error("OAuth state mismatch");
-			}
+
+		const result = await server.waitForCode();
+		if (manualError) throw manualError;
+		if (result?.code) {
+			code = result.code;
+			state = result.state;
+		} else if (manualInput) {
+			const parsed = parseAuthorizationInput(manualInput);
+			if (parsed.state && parsed.state !== verifier) throw new Error("OAuth state mismatch");
 			code = parsed.code;
 			state = parsed.state ?? verifier;
 		}
 
 		if (!code) {
-			throw new Error("Missing authorization code");
+			await manualPromise;
+			if (manualError) throw manualError;
+			if (manualInput) {
+				const parsed = parseAuthorizationInput(manualInput);
+				if (parsed.state && parsed.state !== verifier) throw new Error("OAuth state mismatch");
+				code = parsed.code;
+				state = parsed.state ?? verifier;
+			}
 		}
 
-		if (!state) {
-			throw new Error("Missing OAuth state");
-		}
-
-		options.onProgress?.("Exchanging authorization code for tokens...");
-		return exchangeAuthorizationCode(code, state, verifier, redirectUriForExchange);
+		if (!code) throw new Error("Missing authorization code");
+		if (!state) throw new Error("Missing OAuth state");
+		interaction.notify({ type: "progress", message: "Exchanging authorization code for tokens..." });
+		return exchangeAuthorizationCode(code, state, verifier, REDIRECT_URI);
 	} finally {
+		manualAbort.abort();
 		server.server.close();
 	}
 }
@@ -347,7 +305,7 @@ export async function loginAnthropic(options: {
 /**
  * Refresh Anthropic OAuth token
  */
-export async function refreshAnthropicToken(refreshToken: string): Promise<OAuthCredentials> {
+async function refreshAnthropicToken(refreshToken: string): Promise<OAuthCredential> {
 	let responseBody: string;
 	try {
 		responseBody = await postJson(TOKEN_URL, {
@@ -374,6 +332,7 @@ export async function refreshAnthropicToken(refreshToken: string): Promise<OAuth
 	}
 
 	return {
+		type: "oauth",
 		refresh: data.refresh_token,
 		access: data.access_token,
 		expires: Date.now() + data.expires_in * 1000 - 5 * 60 * 1000,
@@ -382,59 +341,10 @@ export async function refreshAnthropicToken(refreshToken: string): Promise<OAuth
 
 export const anthropicOAuth: OAuthAuth = {
 	name: "Anthropic (Claude Pro/Max)",
-
-	async login(callbacks) {
-		// The manual_code prompt races the local callback server; abort it once
-		// the flow settles so the UI can dismiss the pending input.
-		const manualAbort = new AbortController();
-		try {
-			const credentials = await loginAnthropic({
-				onAuth: (info) => callbacks.notify({ type: "auth_url", url: info.url, instructions: info.instructions }),
-				onProgress: (message) => callbacks.notify({ type: "progress", message }),
-				onPrompt: (prompt) =>
-					callbacks.prompt({ type: "text", message: prompt.message, placeholder: prompt.placeholder }),
-				onManualCodeInput: () =>
-					callbacks.prompt({
-						type: "manual_code",
-						message: "Complete login in your browser, or paste the authorization code / redirect URL here:",
-						placeholder: REDIRECT_URI,
-						signal: manualAbort.signal,
-					}),
-			});
-			return { ...credentials, type: "oauth" };
-		} finally {
-			manualAbort.abort();
-		}
-	},
-
-	async refresh(credential) {
-		return { ...(await refreshAnthropicToken(credential.refresh)), type: "oauth" };
-	},
+	login: loginAnthropic,
+	refresh: (credential) => refreshAnthropicToken(credential.refresh),
 
 	async toAuth(credential) {
 		return { apiKey: credential.access };
 	},
 };
-
-export const anthropicOAuthProvider: OAuthProviderInterface = {
-	id: "anthropic",
-	name: "Anthropic (Claude Pro/Max)",
-	usesCallbackServer: true,
-
-	async login(callbacks: OAuthLoginCallbacks): Promise<OAuthCredentials> {
-		return loginAnthropic({
-			onAuth: callbacks.onAuth,
-			onPrompt: callbacks.onPrompt,
-			onProgress: callbacks.onProgress,
-			onManualCodeInput: callbacks.onManualCodeInput,
-		});
-	},
-
-	async refreshToken(credentials: OAuthCredentials): Promise<OAuthCredentials> {
-		return refreshAnthropicToken(credentials.refresh);
-	},
-
-	getApiKey(credentials: OAuthCredentials): string {
-		return credentials.access;
-	},
-};

packages/ai/src/auth/oauth/github-copilot.ts

@@ -2,16 +2,9 @@
  * GitHub Copilot OAuth flow
  */
 
-import type { OAuthAuth, OAuthCredential } from "../../auth/types.ts";
 import { GITHUB_COPILOT_MODELS } from "../../providers/github-copilot.models.ts";
-import type { Api, Model } from "../../types.ts";
+import type { AuthInteraction, OAuthAuth, OAuthCredential } from "../types.ts";
 import { pollOAuthDeviceCodeFlow } from "./device-code.ts";
-import type { OAuthCredentials, OAuthDeviceCodeInfo, OAuthLoginCallbacks, OAuthProviderInterface } from "./types.ts";
-
-type CopilotCredentials = OAuthCredentials & {
-	enterpriseUrl?: string;
-	availableModelIds: string[];
-};
 
 const decode = (s: string) => atob(s);
 const CLIENT_ID = decode("SXYxLmI1MDdhMDhjODdlY2ZlOTg=");
@@ -44,7 +37,7 @@ type DeviceTokenErrorResponse = {
 	interval?: number;
 };
 
-export function normalizeDomain(input: string): string | null {
+function normalizeDomain(input: string): string | null {
 	const trimmed = input.trim();
 	if (!trimmed) return null;
 	try {
@@ -81,7 +74,7 @@ function getBaseUrlFromToken(token: string): string | null {
 	return `https://${apiHost}`;
 }
 
-export function getGitHubCopilotBaseUrl(token?: string, enterpriseDomain?: string): string {
+function getGitHubCopilotBaseUrl(token?: string, enterpriseDomain?: string): string {
 	// If we have a token, extract the base URL from proxy-ep
 	if (token) {
 		const urlFromToken = getBaseUrlFromToken(token);
@@ -251,7 +244,7 @@ async function pollForGitHubAccessToken(
 async function refreshGitHubCopilotAccessToken(
 	refreshToken: string,
 	enterpriseDomain?: string,
-): Promise<OAuthCredentials> {
+): Promise<OAuthCredential> {
 	const domain = enterpriseDomain || "github.com";
 	const urls = getUrls(domain);
 
@@ -275,6 +268,7 @@ async function refreshGitHubCopilotAccessToken(
 	}
 
 	return {
+		type: "oauth",
 		refresh: refreshToken,
 		access: token,
 		expires: expiresAt * 1000 - 5 * 60 * 1000,
@@ -285,10 +279,7 @@ async function refreshGitHubCopilotAccessToken(
 /**
  * Refresh GitHub Copilot token
  */
-export async function refreshGitHubCopilotToken(
-	refreshToken: string,
-	enterpriseDomain?: string,
-): Promise<OAuthCredentials> {
+async function refreshGitHubCopilotToken(refreshToken: string, enterpriseDomain?: string): Promise<OAuthCredential> {
 	const credentials = await refreshGitHubCopilotAccessToken(refreshToken, enterpriseDomain);
 	return {
 		...credentials,
@@ -326,68 +317,41 @@ async function enableGitHubCopilotModel(token: string, modelId: string, enterpri
  * Enable all known GitHub Copilot models that may require policy acceptance.
  * Called after successful login to ensure all models are available.
  */
-async function enableAllGitHubCopilotModels(
-	token: string,
-	enterpriseDomain?: string,
-	onProgress?: (model: string, success: boolean) => void,
-): Promise<void> {
+async function enableAllGitHubCopilotModels(token: string, enterpriseDomain?: string): Promise<void> {
 	const models = Object.values(GITHUB_COPILOT_MODELS);
 	await Promise.all(
 		models.map(async (model) => {
-			const success = await enableGitHubCopilotModel(token, model.id, enterpriseDomain);
-			onProgress?.(model.id, success);
+			await enableGitHubCopilotModel(token, model.id, enterpriseDomain);
 		}),
 	);
 }
 
-/**
- * Login with GitHub Copilot OAuth (device code flow)
- *
- * @param options.onDeviceCode - Callback with URL and user code
- * @param options.onPrompt - Callback to prompt user for input
- * @param options.onProgress - Optional progress callback
- * @param options.signal - Optional AbortSignal for cancellation
- */
-export async function loginGitHubCopilot(options: {
-	onDeviceCode: (info: OAuthDeviceCodeInfo) => void;
-	onPrompt: (prompt: { message: string; placeholder?: string; allowEmpty?: boolean }) => Promise<string>;
-	onProgress?: (message: string) => void;
-	signal?: AbortSignal;
-}): Promise<OAuthCredentials> {
-	const input = await options.onPrompt({
+async function loginGitHubCopilot(interaction: AuthInteraction): Promise<OAuthCredential> {
+	const input = await interaction.prompt({
+		type: "text",
 		message: "GitHub Enterprise URL/domain (blank for github.com)",
 		placeholder: "company.ghe.com",
-		allowEmpty: true,
 	});
-
-	if (options.signal?.aborted) {
-		throw new Error("Login cancelled");
-	}
+	if (interaction.signal?.aborted) throw new Error("Login cancelled");
 
 	const trimmed = input.trim();
 	const enterpriseDomain = normalizeDomain(input);
-	if (trimmed && !enterpriseDomain) {
-		throw new Error("Invalid GitHub Enterprise URL/domain");
-	}
+	if (trimmed && !enterpriseDomain) throw new Error("Invalid GitHub Enterprise URL/domain");
 	const domain = enterpriseDomain || "github.com";
 
 	const device = await startDeviceFlow(domain);
-	options.onDeviceCode({
+	interaction.notify({
+		type: "device_code",
 		userCode: device.user_code,
 		verificationUri: device.verification_uri,
 		intervalSeconds: device.interval,
 		expiresInSeconds: device.expires_in,
 	});
 
-	const githubAccessToken = await pollForGitHubAccessToken(domain, device, options.signal);
+	const githubAccessToken = await pollForGitHubAccessToken(domain, device, interaction.signal);
 	const credentials = await refreshGitHubCopilotAccessToken(githubAccessToken, enterpriseDomain ?? undefined);
-
-	// Enable all models after successful login
-	options.onProgress?.("Enabling models...");
+	interaction.notify({ type: "progress", message: "Enabling models..." });
 	await enableAllGitHubCopilotModels(credentials.access, enterpriseDomain ?? undefined);
-
-	// Fetch availability after policy enable so newly enabled models are included,
-	// while unavailable models are still filtered out.
 	return {
 		...credentials,
 		availableModelIds: await fetchAvailableGitHubCopilotModelIds(credentials.access, enterpriseDomain ?? undefined),
@@ -402,26 +366,10 @@ function copilotEnterpriseDomain(credential: OAuthCredential): string | undefine
 
 export const githubCopilotOAuth: OAuthAuth = {
 	name: "GitHub Copilot",
+	login: loginGitHubCopilot,
+	refresh: (credential) => refreshGitHubCopilotToken(credential.refresh, copilotEnterpriseDomain(credential)),
 
-	async login(callbacks) {
-		const credentials = await loginGitHubCopilot({
-			onDeviceCode: (info) => callbacks.notify({ type: "device_code", ...info }),
-			onPrompt: (prompt) =>
-				callbacks.prompt({ type: "text", message: prompt.message, placeholder: prompt.placeholder }),
-			onProgress: (message) => callbacks.notify({ type: "progress", message }),
-			signal: callbacks.signal,
-		});
-		return { ...credentials, type: "oauth" };
-	},
-
-	async refresh(credential) {
-		return {
-			...(await refreshGitHubCopilotToken(credential.refresh, copilotEnterpriseDomain(credential))),
-			type: "oauth",
-		};
-	},
-
-	/** Per-credential baseUrl from the token's proxy endpoint replaces the old `modifyModels` rewriting. */
+	/** Derive the credential-specific proxy endpoint for each request. */
 	async toAuth(credential) {
 		return {
 			apiKey: credential.access,
@@ -429,41 +377,3 @@ export const githubCopilotOAuth: OAuthAuth = {
 		};
 	},
 };
-
-export const githubCopilotOAuthProvider: OAuthProviderInterface = {
-	id: "github-copilot",
-	name: "GitHub Copilot",
-
-	async login(callbacks: OAuthLoginCallbacks): Promise<OAuthCredentials> {
-		return loginGitHubCopilot({
-			onDeviceCode: callbacks.onDeviceCode,
-			onPrompt: callbacks.onPrompt,
-			onProgress: callbacks.onProgress,
-			signal: callbacks.signal,
-		});
-	},
-
-	async refreshToken(credentials: OAuthCredentials): Promise<OAuthCredentials> {
-		const creds = credentials as CopilotCredentials;
-		return refreshGitHubCopilotToken(creds.refresh, creds.enterpriseUrl);
-	},
-
-	getApiKey(credentials: OAuthCredentials): string {
-		return credentials.access;
-	},
-
-	modifyModels(models: Model<Api>[], credentials: OAuthCredentials): Model<Api>[] {
-		const creds = credentials as CopilotCredentials;
-		const domain = creds.enterpriseUrl ? (normalizeDomain(creds.enterpriseUrl) ?? undefined) : undefined;
-		const baseUrl = getGitHubCopilotBaseUrl(creds.access, domain);
-		// Older stored Pi auth entries do not have account-specific model IDs yet;
-		// keep their existing generated-catalog behavior until the next refresh/login.
-		const availableModelIds = "availableModelIds" in creds ? new Set(creds.availableModelIds) : undefined;
-
-		return models.flatMap((m) => {
-			if (m.provider !== "github-copilot") return [m];
-			if (availableModelIds && !availableModelIds.has(m.id)) return [];
-			return [{ ...m, baseUrl }];
-		});
-	},
-};

packages/ai/src/auth/oauth/load.ts

@@ -1,4 +1,4 @@
-import type { OAuthAuth } from "../../auth/types.ts";
+import type { OAuthAuth } from "../types.ts";
 
 /**
  * Loads an OAuth flow module through a variable specifier so bundlers cannot

packages/ai/src/auth/oauth/openai-codex.ts

@@ -17,18 +17,11 @@ if (typeof process !== "undefined" && (process.versions?.node || process.version
 	});
 }
 
-import type { OAuthAuth } from "../../auth/types.ts";
-import { getProviderEnvValue } from "../provider-env.ts";
+import { getProviderEnvValue } from "../../utils/provider-env.ts";
+import type { AuthInteraction, OAuthAuth, OAuthCredential } from "../types.ts";
 import { pollOAuthDeviceCodeFlow } from "./device-code.ts";
 import { oauthErrorHtml, oauthSuccessHtml } from "./oauth-page.ts";
 import { generatePKCE } from "./pkce.ts";
-import type {
-	OAuthCredentials,
-	OAuthDeviceCodeInfo,
-	OAuthLoginCallbacks,
-	OAuthPrompt,
-	OAuthProviderInterface,
-} from "./types.ts";
 
 const CLIENT_ID = "app_EMoamEEZ73f0CkXaXp7hrann";
 const AUTH_BASE_URL = "https://auth.openai.com";
@@ -40,8 +33,8 @@ const DEVICE_TOKEN_URL = `${AUTH_BASE_URL}/api/accounts/deviceauth/token`;
 const DEVICE_VERIFICATION_URI = `${AUTH_BASE_URL}/codex/device`;
 const DEVICE_REDIRECT_URI = `${AUTH_BASE_URL}/deviceauth/callback`;
 const DEVICE_CODE_TIMEOUT_SECONDS = 15 * 60;
-export const OPENAI_CODEX_BROWSER_LOGIN_METHOD = "browser";
-export const OPENAI_CODEX_DEVICE_CODE_LOGIN_METHOD = "device_code";
+const OPENAI_CODEX_BROWSER_LOGIN_METHOD = "browser";
+const OPENAI_CODEX_DEVICE_CODE_LOGIN_METHOD = "device_code";
 const SCOPE = "openid profile email offline_access";
 const JWT_CLAIM_PATH = "https://api.openai.com/auth";
 
@@ -406,13 +399,14 @@ function getAccountId(accessToken: string): string | null {
 	return typeof accountId === "string" && accountId.length > 0 ? accountId : null;
 }
 
-function credentialsFromToken(token: OAuthToken): OAuthCredentials {
+function credentialsFromToken(token: OAuthToken): OAuthCredential {
 	const accountId = getAccountId(token.access);
 	if (!accountId) {
 		throw new Error("Failed to extract accountId from token");
 	}
 
 	return {
+		type: "oauth",
 		access: token.access,
 		refresh: token.refresh,
 		expires: token.expires,
@@ -425,132 +419,83 @@ async function exchangeAuthorizationCodeForCredentials(
 	verifier: string,
 	redirectUri: string,
 	signal?: AbortSignal,
-): Promise<OAuthCredentials> {
+): Promise<OAuthCredential> {
 	return credentialsFromToken(await exchangeAuthorizationCode(code, verifier, redirectUri, signal));
 }
 
-/**
- * Login with OpenAI Codex OAuth using the Codex device-code flow.
- */
-export async function loginOpenAICodexDeviceCode(options: {
-	onDeviceCode: (info: OAuthDeviceCodeInfo) => void;
-	signal?: AbortSignal;
-}): Promise<OAuthCredentials> {
-	const device = await startOpenAICodexDeviceAuth(options.signal);
-	options.onDeviceCode({
+async function loginOpenAICodexDeviceCode(interaction: AuthInteraction): Promise<OAuthCredential> {
+	const device = await startOpenAICodexDeviceAuth(interaction.signal);
+	interaction.notify({
+		type: "device_code",
 		userCode: device.userCode,
 		verificationUri: DEVICE_VERIFICATION_URI,
 		intervalSeconds: device.intervalSeconds,
 		expiresInSeconds: DEVICE_CODE_TIMEOUT_SECONDS,
 	});
-	const code = await pollOpenAICodexDeviceAuth(device, options.signal);
+	const code = await pollOpenAICodexDeviceAuth(device, interaction.signal);
 	return exchangeAuthorizationCodeForCredentials(
 		code.authorizationCode,
 		code.codeVerifier,
 		DEVICE_REDIRECT_URI,
-		options.signal,
+		interaction.signal,
 	);
 }
 
-/**
- * Login with OpenAI Codex OAuth
- *
- * @param options.onAuth - Called with URL and instructions when auth starts
- * @param options.onPrompt - Called to prompt user for manual code paste (fallback if no onManualCodeInput)
- * @param options.onProgress - Optional progress messages
- * @param options.onManualCodeInput - Optional promise that resolves with user-pasted code.
- *                                    Races with browser callback - whichever completes first wins.
- *                                    Useful for showing paste input immediately alongside browser flow.
- * @param options.originator - OAuth originator parameter (defaults to "pi")
- */
-export async function loginOpenAICodex(options: {
-	onAuth: (info: { url: string; instructions?: string }) => void;
-	onPrompt: (prompt: OAuthPrompt) => Promise<string>;
-	onProgress?: (message: string) => void;
-	onManualCodeInput?: () => Promise<string>;
-	originator?: string;
-}): Promise<OAuthCredentials> {
-	const { verifier, state, url } = await createAuthorizationFlow(options.originator);
+async function loginOpenAICodex(interaction: AuthInteraction): Promise<OAuthCredential> {
+	const { verifier, state, url } = await createAuthorizationFlow();
 	const server = await startLocalOAuthServer(state);
-
-	options.onAuth({ url, instructions: "A browser window should open. Complete login to finish." });
-
+	const manualAbort = new AbortController();
 	let code: string | undefined;
+	let manualCode: string | undefined;
+	let manualError: Error | undefined;
+
+	interaction.notify({
+		type: "auth_url",
+		url,
+		instructions: "A browser window should open. Complete login to finish.",
+	});
+
 	try {
-		if (options.onManualCodeInput) {
-			// Race between browser callback and manual input
-			let manualCode: string | undefined;
-			let manualError: Error | undefined;
-			const manualPromise = options
-				.onManualCodeInput()
-				.then((input) => {
-					manualCode = input;
-					server.cancelWait();
-				})
-				.catch((err) => {
-					manualError = err instanceof Error ? err : new Error(String(err));
-					server.cancelWait();
-				});
-
-			const result = await server.waitForCode();
-
-			// If manual input was cancelled, throw that error
-			if (manualError) {
-				throw manualError;
-			}
-
-			if (result?.code) {
-				// Browser callback won
-				code = result.code;
-			} else if (manualCode) {
-				// Manual input won (or callback timed out and user had entered code)
-				const parsed = parseAuthorizationInput(manualCode);
-				if (parsed.state && parsed.state !== state) {
-					throw new Error("State mismatch");
-				}
-				code = parsed.code;
-			}
-
-			// If still no code, wait for manual promise to complete and try that
-			if (!code) {
-				await manualPromise;
-				if (manualError) {
-					throw manualError;
-				}
-				if (manualCode) {
-					const parsed = parseAuthorizationInput(manualCode);
-					if (parsed.state && parsed.state !== state) {
-						throw new Error("State mismatch");
-					}
-					code = parsed.code;
-				}
-			}
-		} else {
-			// Original flow: wait for callback, then prompt if needed
-			const result = await server.waitForCode();
-			if (result?.code) {
-				code = result.code;
-			}
-		}
-
-		// Fallback to onPrompt if still no code
-		if (!code) {
-			const input = await options.onPrompt({
-				message: "Paste the authorization code (or full redirect URL):",
+		const manualPromise = interaction
+			.prompt({
+				type: "manual_code",
+				message: "Complete login in your browser, or paste the authorization code / redirect URL here:",
+				placeholder: REDIRECT_URI,
+				signal: manualAbort.signal,
+			})
+			.then((input) => {
+				manualCode = input;
+				server.cancelWait();
+			})
+			.catch((error) => {
+				manualError = error instanceof Error ? error : new Error(String(error));
+				server.cancelWait();
 			});
-			const parsed = parseAuthorizationInput(input);
-			if (parsed.state && parsed.state !== state) {
-				throw new Error("State mismatch");
-			}
+
+		const result = await server.waitForCode();
+		if (manualError) throw manualError;
+		if (result?.code) {
+			code = result.code;
+		} else if (manualCode) {
+			const parsed = parseAuthorizationInput(manualCode);
+			if (parsed.state && parsed.state !== state) throw new Error("State mismatch");
 			code = parsed.code;
 		}
 
 		if (!code) {
-			throw new Error("Missing authorization code");
+			await manualPromise;
+			if (manualError) throw manualError;
+			if (manualCode) {
+				const parsed = parseAuthorizationInput(manualCode);
+				if (parsed.state && parsed.state !== state) throw new Error("State mismatch");
+				code = parsed.code;
+			}
 		}
 
-		return exchangeAuthorizationCodeForCredentials(code, verifier, REDIRECT_URI);
+		if (!code) throw new Error("Missing authorization code");
+		return exchangeAuthorizationCodeForCredentials(code, verifier, REDIRECT_URI, interaction.signal);
 	} finally {
+		manualAbort.abort();
 		server.close();
 	}
 }
@@ -558,15 +503,15 @@ export async function loginOpenAICodex(options: {
 /**
  * Refresh OpenAI Codex OAuth token
  */
-export async function refreshOpenAICodexToken(refreshToken: string): Promise<OAuthCredentials> {
+async function refreshOpenAICodexToken(refreshToken: string): Promise<OAuthCredential> {
 	return credentialsFromToken(await refreshAccessToken(refreshToken));
 }
 
 export const openaiCodexOAuth: OAuthAuth = {
 	name: "OpenAI (ChatGPT Plus/Pro)",
 
-	async login(callbacks) {
-		const method = await callbacks.prompt({
+	async login(interaction) {
+		const method = await interaction.prompt({
 			type: "select",
 			message: "Select OpenAI Codex login method:",
 			options: [
@@ -576,89 +521,18 @@ export const openaiCodexOAuth: OAuthAuth = {
 		});
 
 		if (method === OPENAI_CODEX_DEVICE_CODE_LOGIN_METHOD) {
-			const credentials = await loginOpenAICodexDeviceCode({
-				onDeviceCode: (info) => callbacks.notify({ type: "device_code", ...info }),
-				signal: callbacks.signal,
-			});
-			return { ...credentials, type: "oauth" };
+			return loginOpenAICodexDeviceCode(interaction);
 		}
 		if (method !== OPENAI_CODEX_BROWSER_LOGIN_METHOD) {
 			throw new Error(`Unknown OpenAI Codex login method: ${method}`);
 		}
 
-		// The manual_code prompt races the local callback server; abort it once
-		// the flow settles so the UI can dismiss the pending input.
-		const manualAbort = new AbortController();
-		try {
-			const credentials = await loginOpenAICodex({
-				onAuth: (info) => callbacks.notify({ type: "auth_url", url: info.url, instructions: info.instructions }),
-				onProgress: (message) => callbacks.notify({ type: "progress", message }),
-				onPrompt: (prompt) =>
-					callbacks.prompt({ type: "text", message: prompt.message, placeholder: prompt.placeholder }),
-				onManualCodeInput: () =>
-					callbacks.prompt({
-						type: "manual_code",
-						message: "Complete login in your browser, or paste the authorization code / redirect URL here:",
-						placeholder: REDIRECT_URI,
-						signal: manualAbort.signal,
-					}),
-			});
-			return { ...credentials, type: "oauth" };
-		} finally {
-			manualAbort.abort();
-		}
+		return loginOpenAICodex(interaction);
 	},
 
-	async refresh(credential) {
-		return { ...(await refreshOpenAICodexToken(credential.refresh)), type: "oauth" };
-	},
+	refresh: (credential) => refreshOpenAICodexToken(credential.refresh),
 
 	async toAuth(credential) {
 		return { apiKey: credential.access };
 	},
 };
-
-export const openaiCodexOAuthProvider: OAuthProviderInterface = {
-	id: "openai-codex",
-	name: "ChatGPT Plus/Pro (Codex Subscription)",
-	usesCallbackServer: true,
-
-	async login(callbacks: OAuthLoginCallbacks): Promise<OAuthCredentials> {
-		const loginMethod = await callbacks.onSelect({
-			message: "Select OpenAI Codex login method:",
-			options: [
-				{ id: OPENAI_CODEX_BROWSER_LOGIN_METHOD, label: "Browser login (default)" },
-				{ id: OPENAI_CODEX_DEVICE_CODE_LOGIN_METHOD, label: "Device code login (headless)" },
-			],
-		});
-		if (!loginMethod) {
-			throw new Error("Login cancelled");
-		}
-
-		if (loginMethod === OPENAI_CODEX_DEVICE_CODE_LOGIN_METHOD) {
-			return loginOpenAICodexDeviceCode({
-				onDeviceCode: callbacks.onDeviceCode,
-				signal: callbacks.signal,
-			});
-		}
-
-		if (loginMethod !== OPENAI_CODEX_BROWSER_LOGIN_METHOD) {
-			throw new Error(`Unknown OpenAI Codex login method: ${loginMethod}`);
-		}
-
-		return loginOpenAICodex({
-			onAuth: callbacks.onAuth,
-			onPrompt: callbacks.onPrompt,
-			onProgress: callbacks.onProgress,
-			onManualCodeInput: callbacks.onManualCodeInput,
-		});
-	},
-
-	async refreshToken(credentials: OAuthCredentials): Promise<OAuthCredentials> {
-		return refreshOpenAICodexToken(credentials.refresh);
-	},
-
-	getApiKey(credentials: OAuthCredentials): string {
-		return credentials.access;
-	},
-};

packages/ai/src/auth/resolve.ts

@@ -1,4 +1,4 @@
-import type { Api, ImagesApi, ImagesModel, Model, ProviderEnv } from "../types.ts";
+import type { ProviderEnv } from "../types.ts";
 import type {
 	ApiKeyAuth,
 	ApiKeyCredential,
@@ -28,9 +28,6 @@ export class ModelsError extends Error {
 	}
 }
 
-/** Model shape auth resolution receives: chat or image-generation models. */
-export type AuthModel = Model<Api> | ImagesModel<ImagesApi>;
-
 /**
  * Auth resolution shared by the `Models` and `ImagesModels` collections.
  * A stored credential owns the provider: ambient/env is consulted only when
@@ -39,7 +36,6 @@ export type AuthModel = Model<Api> | ImagesModel<ImagesApi>;
  */
 export async function resolveProviderAuth(
 	provider: { id: string; auth: ProviderAuth },
-	model: AuthModel,
 	credentials: CredentialStore,
 	authContext: AuthContext,
 	overrides?: AuthResolutionOverrides,
@@ -47,7 +43,7 @@ export async function resolveProviderAuth(
 	const requestAuthContext = overrides?.env ? overlayEnvAuthContext(authContext, overrides.env) : authContext;
 
 	if (overrides?.apiKey !== undefined && provider.auth.apiKey) {
-		return resolveApiKey(requestAuthContext, provider.auth.apiKey, model, {
+		return resolveApiKey(requestAuthContext, provider.auth.apiKey, provider.id, {
 			type: "api_key",
 			key: overrides.apiKey,
 			env: overrides.env,
@@ -61,13 +57,15 @@ export async function resolveProviderAuth(
 		}
 		if (stored.type === "api_key" && provider.auth.apiKey) {
 			const credential = overrides?.env ? { ...stored, env: { ...stored.env, ...overrides.env } } : stored;
-			return resolveApiKey(requestAuthContext, provider.auth.apiKey, model, credential);
+			return resolveApiKey(requestAuthContext, provider.auth.apiKey, provider.id, credential);
 		}
 		return undefined;
 	}
 
 	// Ambient (env vars, AWS profiles, ADC files).
-	return provider.auth.apiKey ? resolveApiKey(requestAuthContext, provider.auth.apiKey, model, undefined) : undefined;
+	return provider.auth.apiKey
+		? resolveApiKey(requestAuthContext, provider.auth.apiKey, provider.id, undefined)
+		: undefined;
 }
 
 function overlayEnvAuthContext(base: AuthContext, env: ProviderEnv): AuthContext {
@@ -122,13 +120,13 @@ async function resolveStoredOAuth(
 async function resolveApiKey(
 	authContext: AuthContext,
 	apiKey: ApiKeyAuth,
-	model: AuthModel,
+	providerId: string,
 	credential: ApiKeyCredential | undefined,
 ): Promise<AuthResult | undefined> {
 	try {
-		return await apiKey.resolve({ model, ctx: authContext, credential });
+		return await apiKey.resolve({ ctx: authContext, credential });
 	} catch (error) {
-		throw new ModelsError("auth", `API key auth failed for provider ${model.provider}`, { cause: error });
+		throw new ModelsError("auth", `API key auth failed for provider ${providerId}`, { cause: error });
 	}
 }
 

packages/ai/src/auth/types.ts

@@ -1,5 +1,4 @@
-import type { Api, ImagesApi, ImagesModel, Model, ProviderEnv, ProviderHeaders } from "../types.ts";
-import type { OAuthCredentials } from "../utils/oauth/types.ts";
+import type { ProviderEnv, ProviderHeaders } from "../types.ts";
 
 /**
  * Request auth for a single model request. If a value cannot be expressed as
@@ -21,7 +20,15 @@ export interface ApiKeyCredential {
 	env?: ProviderEnv;
 }
 
-/** Stored OAuth credential (`access`, `refresh`, `expires` from OAuthCredentials). */
+/** OAuth token data returned by extension compatibility flows. */
+export interface OAuthCredentials {
+	refresh: string;
+	access: string;
+	expires: number;
+	[key: string]: unknown;
+}
+
+/** Stored canonical OAuth credential. */
 export interface OAuthCredential extends OAuthCredentials {
 	type: "oauth";
 }
@@ -29,6 +36,12 @@ export interface OAuthCredential extends OAuthCredentials {
 /** One type-tagged credential per provider — the shape of today's auth.json. */
 export type Credential = ApiKeyCredential | OAuthCredential;
 
+/** Non-secret credential metadata for account/status enumeration. */
+export interface CredentialInfo {
+	providerId: string;
+	type: Credential["type"];
+}
+
 /**
  * App-owned credential storage, keyed by `Provider.id`, one credential per
  * provider. `modify` is the only write path, so every mutation is a
@@ -51,6 +64,12 @@ export interface CredentialStore {
 	 */
 	read(providerId: string): Promise<Credential | undefined>;
 
+	/**
+	 * List stored credential metadata without resolving or exposing secrets.
+	 * Implementations must not execute configured API-key commands while listing.
+	 */
+	list(): Promise<readonly CredentialInfo[]>;
+
 	/**
 	 * Serialized write — the only write path. `fn` sees the current credential
 	 * because correct writes (refresh, login-during-refresh) depend on it;
@@ -84,6 +103,13 @@ export interface AuthResult {
 	source?: string;
 }
 
+export interface AuthCheck {
+	source?: string;
+	type: "api_key" | "oauth";
+}
+
+export type AuthType = "api_key" | "oauth";
+
 /**
  * Prompt shown to the user during login. `signal` lets the flow cancel a
  * pending prompt when an out-of-band event resolves the step, e.g. a
@@ -97,7 +123,13 @@ export type AuthPrompt = { signal?: AbortSignal } & (
 	| { type: "manual_code"; message: string; placeholder?: string }
 );
 
+export interface AuthInfoLink {
+	url: string;
+	label?: string;
+}
+
 export type AuthEvent =
+	| { type: "info"; message: string; links?: readonly AuthInfoLink[] }
 	| { type: "auth_url"; url: string; instructions?: string }
 	| {
 			type: "device_code";
@@ -115,7 +147,7 @@ export type AuthEvent =
  * id). Rejects on cancel/abort. `signal` aborts the whole login flow;
  * per-prompt cancellation uses `AuthPrompt.signal`.
  */
-export interface AuthLoginCallbacks {
+export interface AuthInteraction {
 	signal?: AbortSignal;
 
 	prompt(prompt: AuthPrompt): Promise<string>;
@@ -131,19 +163,22 @@ export interface ApiKeyAuth {
 	name: string;
 
 	/** Interactive setup (prompt for key/provider env). Absent = ambient-only. */
-	login?(callbacks: AuthLoginCallbacks): Promise<ApiKeyCredential>;
+	login?(interaction: AuthInteraction): Promise<ApiKeyCredential>;
+
+	/**
+	 * Optional side-effect-free availability check. Use this when `resolve()` may
+	 * execute commands or perform other request-time work. Missing means Models
+	 * checks availability by resolving auth.
+	 */
+	check?(input: { ctx: AuthContext; credential?: ApiKeyCredential }): Promise<AuthCheck | undefined>;
 
 	/**
 	 * Resolve auth from the stored credential and/or ambient sources, merging
 	 * per field (`credential.key ?? env("...")`, `credential.env?.NAME ?? env("...")`).
-	 * undefined = not configured. Receives the chat or image-generation model
-	 * the request is for (both carry `provider` and `baseUrl`).
+	 * undefined = not configured. Resolution is provider-scoped; model-specific
+	 * endpoint preparation happens after auth has been resolved.
 	 */
-	resolve(input: {
-		model: Model<Api> | ImagesModel<ImagesApi>;
-		ctx: AuthContext;
-		credential?: ApiKeyCredential;
-	}): Promise<AuthResult | undefined>;
+	resolve(input: { ctx: AuthContext; credential?: ApiKeyCredential }): Promise<AuthResult | undefined>;
 }
 
 /**
@@ -155,7 +190,7 @@ export interface OAuthAuth {
 	/** Display name, e.g. "Anthropic (Claude Pro/Max)". */
 	name: string;
 
-	login(callbacks: AuthLoginCallbacks): Promise<OAuthCredential>;
+	login(interaction: AuthInteraction): Promise<OAuthCredential>;
 
 	/**
 	 * Exchange the refresh token. Network call; throws on failure

packages/ai/src/cli.ts

@@ -1,71 +1,74 @@
 #!/usr/bin/env node
 
+import { existsSync, readFileSync, writeFileSync } from "node:fs";
 import { createInterface } from "node:readline";
-import { existsSync, readFileSync, writeFileSync } from "fs";
-import { getOAuthProvider, getOAuthProviders } from "./utils/oauth/index.ts";
-import type { OAuthCredentials, OAuthProviderId } from "./utils/oauth/types.ts";
+import type { AuthPrompt, OAuthCredential, Provider } from "./index.ts";
+import { builtinProviders } from "./providers/all.ts";
 
 const AUTH_FILE = "auth.json";
-const PROVIDERS = getOAuthProviders();
+const PROVIDERS = builtinProviders().filter(
+	(provider): provider is Provider & { auth: { oauth: NonNullable<Provider["auth"]["oauth"]> } } =>
+		provider.auth.oauth !== undefined,
+);
 
 function prompt(rl: ReturnType<typeof createInterface>, question: string): Promise<string> {
 	return new Promise((resolve) => rl.question(question, resolve));
 }
 
-function loadAuth(): Record<string, { type: "oauth" } & OAuthCredentials> {
+function loadAuth(): Record<string, OAuthCredential> {
 	if (!existsSync(AUTH_FILE)) return {};
 	try {
-		return JSON.parse(readFileSync(AUTH_FILE, "utf-8"));
+		return JSON.parse(readFileSync(AUTH_FILE, "utf-8")) as Record<string, OAuthCredential>;
 	} catch {
 		return {};
 	}
 }
 
-function saveAuth(auth: Record<string, { type: "oauth" } & OAuthCredentials>): void {
+function saveAuth(auth: Record<string, OAuthCredential>): void {
 	writeFileSync(AUTH_FILE, JSON.stringify(auth, null, 2), "utf-8");
 }
 
-async function login(providerId: OAuthProviderId): Promise<void> {
-	const provider = getOAuthProvider(providerId);
-	if (!provider) {
-		console.error(`Unknown provider: ${providerId}`);
-		process.exit(1);
+async function answerPrompt(rl: ReturnType<typeof createInterface>, authPrompt: AuthPrompt): Promise<string> {
+	if (authPrompt.type === "select") {
+		console.log(`\n${authPrompt.message}`);
+		for (let index = 0; index < authPrompt.options.length; index++) {
+			console.log(`  ${index + 1}. ${authPrompt.options[index].label}`);
+		}
+		const choice = Number.parseInt(await prompt(rl, `Enter number (1-${authPrompt.options.length}): `), 10) - 1;
+		const selected = authPrompt.options[choice];
+		if (!selected) throw new Error("Invalid selection");
+		return selected.id;
 	}
+	return prompt(rl, `${authPrompt.message}${authPrompt.placeholder ? ` (${authPrompt.placeholder})` : ""}: `);
+}
 
+async function login(providerId: string): Promise<void> {
+	const provider = PROVIDERS.find((entry) => entry.id === providerId);
+	if (!provider) throw new Error(`Unknown provider: ${providerId}`);
 	const rl = createInterface({ input: process.stdin, output: process.stdout });
-	const promptFn = (msg: string) => prompt(rl, `${msg} `);
-
 	try {
-		const credentials = await provider.login({
-			onAuth: (info) => {
-				console.log(`\nOpen this URL in your browser:\n${info.url}`);
-				if (info.instructions) console.log(info.instructions);
-				console.log();
-			},
-			onDeviceCode: (info) => {
-				console.log(`\nOpen this URL in your browser:\n${info.verificationUri}`);
-				console.log(`Enter code: ${info.userCode}`);
-				console.log();
-			},
-			onPrompt: async (p) => {
-				return await promptFn(`${p.message}${p.placeholder ? ` (${p.placeholder})` : ""}:`);
-			},
-			onSelect: async (p) => {
-				console.log(`\n${p.message}`);
-				for (let i = 0; i < p.options.length; i++) {
-					console.log(`  ${i + 1}. ${p.options[i].label}`);
+		const credential = await provider.auth.oauth.login({
+			prompt: (authPrompt) => answerPrompt(rl, authPrompt),
+			notify: (event) => {
+				switch (event.type) {
+					case "auth_url":
+						console.log(`\nOpen this URL in your browser:\n${event.url}`);
+						if (event.instructions) console.log(event.instructions);
+						break;
+					case "device_code":
+						console.log(`\nOpen this URL in your browser:\n${event.verificationUri}`);
+						console.log(`Enter code: ${event.userCode}`);
+						break;
+					case "info":
+					case "progress":
+						console.log(event.message);
+						break;
 				}
-				const choice = await promptFn(`Enter number (1-${p.options.length}):`);
-				const index = parseInt(choice, 10) - 1;
-				return p.options[index]?.id;
 			},
-			onProgress: (msg) => console.log(msg),
 		});
-
 		const auth = loadAuth();
-		auth[providerId] = { type: "oauth", ...credentials };
+		auth[providerId] = credential;
 		saveAuth(auth);
-
 		console.log(`\nCredentials saved to ${AUTH_FILE}`);
 	} finally {
 		rl.close();
@@ -75,73 +78,41 @@ async function login(providerId: OAuthProviderId): Promise<void> {
 async function main(): Promise<void> {
 	const args = process.argv.slice(2);
 	const command = args[0];
-
 	if (!command || command === "help" || command === "--help" || command === "-h") {
-		const providerList = PROVIDERS.map((p) => `  ${p.id.padEnd(20)} ${p.name}`).join("\n");
-		console.log(`Usage: npx @earendil-works/pi-ai <command> [provider]
-
-Commands:
-  login [provider]  Login to an OAuth provider
-  list              List available providers
-
-Providers:
-${providerList}
-
-Examples:
-  npx @earendil-works/pi-ai login              # interactive provider selection
-  npx @earendil-works/pi-ai login anthropic    # login to specific provider
-  npx @earendil-works/pi-ai list               # list providers
-`);
+		const providerList = PROVIDERS.map((provider) => `  ${provider.id.padEnd(20)} ${provider.name}`).join("\n");
+		console.log(
+			`Usage: npx @earendil-works/pi-ai <command> [provider]\n\nCommands:\n  login [provider]  Login to an OAuth provider\n  list              List available providers\n\nProviders:\n${providerList}`,
+		);
 		return;
 	}
-
 	if (command === "list") {
-		console.log("Available OAuth providers:\n");
-		for (const p of PROVIDERS) {
-			console.log(`  ${p.id.padEnd(20)} ${p.name}`);
-		}
+		for (const provider of PROVIDERS) console.log(`${provider.id.padEnd(20)} ${provider.name}`);
 		return;
 	}
-
 	if (command === "login") {
-		let provider = args[1] as OAuthProviderId | undefined;
-
-		if (!provider) {
+		let providerId = args[1];
+		if (!providerId) {
 			const rl = createInterface({ input: process.stdin, output: process.stdout });
-			console.log("Select a provider:\n");
-			for (let i = 0; i < PROVIDERS.length; i++) {
-				console.log(`  ${i + 1}. ${PROVIDERS[i].name}`);
+			try {
+				for (let index = 0; index < PROVIDERS.length; index++) {
+					console.log(`  ${index + 1}. ${PROVIDERS[index].name}`);
+				}
+				const index = Number.parseInt(await prompt(rl, `Enter number (1-${PROVIDERS.length}): `), 10) - 1;
+				providerId = PROVIDERS[index]?.id;
+			} finally {
+				rl.close();
 			}
-			console.log();
-
-			const choice = await prompt(rl, `Enter number (1-${PROVIDERS.length}): `);
-			rl.close();
-
-			const index = parseInt(choice, 10) - 1;
-			if (index < 0 || index >= PROVIDERS.length) {
-				console.error("Invalid selection");
-				process.exit(1);
-			}
-			provider = PROVIDERS[index].id;
 		}
-
-		if (!PROVIDERS.some((p) => p.id === provider)) {
-			console.error(`Unknown provider: ${provider}`);
-			console.error(`Use 'npx @earendil-works/pi-ai list' to see available providers`);
-			process.exit(1);
+		if (!providerId || !PROVIDERS.some((provider) => provider.id === providerId)) {
+			throw new Error(`Unknown provider: ${providerId ?? ""}`);
 		}
-
-		console.log(`Logging in to ${provider}...`);
-		await login(provider);
+		await login(providerId);
 		return;
 	}
-
-	console.error(`Unknown command: ${command}`);
-	console.error(`Use 'npx @earendil-works/pi-ai --help' for usage`);
-	process.exit(1);
+	throw new Error(`Unknown command: ${command}`);
 }
 
-main().catch((err) => {
-	console.error("Error:", err.message);
+main().catch((error: unknown) => {
+	console.error("Error:", error instanceof Error ? error.message : String(error));
 	process.exit(1);
 });

packages/ai/src/compat.ts

@@ -37,6 +37,7 @@ import { openAICodexResponsesApi } from "./api/openai-codex-responses.lazy.ts";
 import { openAICompletionsApi } from "./api/openai-completions.lazy.ts";
 import { openAIResponsesApi } from "./api/openai-responses.lazy.ts";
 import { getEnvApiKey } from "./env-api-keys.ts";
+import type { ModelsApiStreamOptions } from "./models.ts";
 import { builtinModels, getBuiltinModel, getBuiltinModels, getBuiltinProviders } from "./providers/all.ts";
 import { createFauxCore, type FauxProviderRegistration, type RegisterFauxProviderOptions } from "./providers/faux.ts";
 import type {
@@ -221,9 +222,14 @@ function withEnvApiKey<TOptions extends StreamOptions>(
 	return { ...options, apiKey } as TOptions;
 }
 
-function shouldUseBuiltinModels(model: Model<Api>): boolean {
-	const builtin = compatModels.getModel(model.provider, model.id);
-	return builtin?.api === model.api && getApiProvider(model.api) === builtinApiProviderInstances.get(model.api);
+function hasResolvedCloudflareAuth(options: StreamOptions | undefined): boolean {
+	return hasExplicitApiKey(options?.apiKey) || typeof options?.headers?.["cf-aig-authorization"] === "string";
+}
+
+function getBuiltinProviderForModel(model: Model<Api>) {
+	if (getApiProvider(model.api) !== builtinApiProviderInstances.get(model.api)) return undefined;
+	const provider = compatModels.getProvider(model.provider);
+	return provider?.getModels().some((candidate) => candidate.api === model.api) ? provider : undefined;
 }
 
 function resolveApiProvider(api: Api) {
@@ -239,8 +245,12 @@ export function stream<TApi extends Api>(
 	context: Context,
 	options?: ProviderStreamOptions,
 ): AssistantMessageEventStream {
-	if (shouldUseBuiltinModels(model)) {
-		return compatModels.stream(model, context, options as ApiStreamOptions<TApi> | undefined);
+	const builtinProvider = getBuiltinProviderForModel(model);
+	if (builtinProvider) {
+		if (model.provider.startsWith("cloudflare-") && !hasResolvedCloudflareAuth(options)) {
+			return compatModels.stream(model, context, options as ModelsApiStreamOptions<TApi> | undefined);
+		}
+		return builtinProvider.stream(model, context, withEnvApiKey(model, options) as ApiStreamOptions<TApi>);
 	}
 	const provider = resolveApiProvider(model.api);
 	return provider.stream(model, context, withEnvApiKey(model, options) as StreamOptions);
@@ -260,8 +270,12 @@ export function streamSimple<TApi extends Api>(
 	context: Context,
 	options?: SimpleStreamOptions,
 ): AssistantMessageEventStream {
-	if (shouldUseBuiltinModels(model)) {
-		return compatModels.streamSimple(model, context, options);
+	const builtinProvider = getBuiltinProviderForModel(model);
+	if (builtinProvider) {
+		if (model.provider.startsWith("cloudflare-") && !hasResolvedCloudflareAuth(options)) {
+			return compatModels.streamSimple(model, context, options);
+		}
+		return builtinProvider.streamSimple(model, context, withEnvApiKey(model, options));
 	}
 	const provider = resolveApiProvider(model.api);
 	return provider.streamSimple(model, context, withEnvApiKey(model, options));

packages/ai/src/compat/extension-oauth-types.ts

@@ -0,0 +1,45 @@
+import type { OAuthCredentials } from "../auth/types.ts";
+
+/** Legacy extension OAuth prompt. */
+export interface OAuthPrompt {
+	message: string;
+	placeholder?: string;
+	allowEmpty?: boolean;
+}
+
+/** Legacy extension OAuth authorization link. */
+export interface OAuthAuthInfo {
+	url: string;
+	instructions?: string;
+}
+
+/** Legacy extension OAuth device-code notification. */
+export interface OAuthDeviceCodeInfo {
+	userCode: string;
+	verificationUri: string;
+	intervalSeconds?: number;
+	expiresInSeconds?: number;
+}
+
+export interface OAuthSelectOption {
+	id: string;
+	label: string;
+}
+
+export interface OAuthSelectPrompt {
+	message: string;
+	options: OAuthSelectOption[];
+}
+
+/** Callback surface retained only for coding-agent extension compatibility. */
+export interface OAuthLoginCallbacks {
+	onAuth(info: OAuthAuthInfo): void;
+	onDeviceCode(info: OAuthDeviceCodeInfo): void;
+	onPrompt(prompt: OAuthPrompt): Promise<string>;
+	onProgress?(message: string): void;
+	onManualCodeInput?(): Promise<string>;
+	onSelect(prompt: OAuthSelectPrompt): Promise<string | undefined>;
+	signal?: AbortSignal;
+}
+
+export type { OAuthCredentials };

packages/ai/src/images-models.ts

@@ -1,6 +1,6 @@
 import { defaultProviderAuthContext as defaultAuthContext } from "./auth/context.ts";
 import { InMemoryCredentialStore } from "./auth/credential-store.ts";
-import { ModelsError, resolveProviderAuth } from "./auth/resolve.ts";
+import { type AuthResolutionOverrides, ModelsError, resolveProviderAuth } from "./auth/resolve.ts";
 import type { AuthContext, AuthResult, CredentialStore, ProviderAuth } from "./auth/types.ts";
 import type { CreateModelsOptions } from "./models.ts";
 import type { AssistantImages, ImagesApi, ImagesContext, ImagesModel, ImagesOptions, ProviderImages } from "./types.ts";
@@ -68,11 +68,12 @@ export interface ImagesModels {
 	refresh(provider?: string): Promise<void>;
 
 	/**
-	 * Resolve request auth for an image model. Same contract as
+	 * Resolve request auth by provider id or image model. Same contract as
 	 * `Models.getAuth()`: undefined when unknown/unconfigured, rejects with
 	 * `ModelsError` ("oauth"/"auth") on real failures.
 	 */
-	getAuth(model: ImagesModel<ImagesApi>): Promise<AuthResult | undefined>;
+	getAuth(providerId: string, overrides?: AuthResolutionOverrides): Promise<AuthResult | undefined>;
+	getAuth(model: ImagesModel<ImagesApi>, overrides?: AuthResolutionOverrides): Promise<AuthResult | undefined>;
 
 	/**
 	 * Generate images through the owning provider with auth resolved and
@@ -167,10 +168,16 @@ class ImagesModelsImpl implements MutableImagesModels {
 		await Promise.allSettled(Array.from(this.providers.values(), async (entry) => entry.refreshModels?.()));
 	}
 
-	async getAuth(model: ImagesModel<ImagesApi>): Promise<AuthResult | undefined> {
-		const provider = this.providers.get(model.provider);
+	getAuth(providerId: string, overrides?: AuthResolutionOverrides): Promise<AuthResult | undefined>;
+	getAuth(model: ImagesModel<ImagesApi>, overrides?: AuthResolutionOverrides): Promise<AuthResult | undefined>;
+	async getAuth(
+		providerOrModel: string | ImagesModel<ImagesApi>,
+		overrides?: AuthResolutionOverrides,
+	): Promise<AuthResult | undefined> {
+		const providerId = typeof providerOrModel === "string" ? providerOrModel : providerOrModel.provider;
+		const provider = this.providers.get(providerId);
 		if (!provider) return undefined;
-		return resolveProviderAuth(provider, model, this.credentials, this.authContext);
+		return resolveProviderAuth(provider, this.credentials, this.authContext, overrides);
 	}
 
 	async generateImages(
@@ -184,7 +191,7 @@ class ImagesModelsImpl implements MutableImagesModels {
 				throw new ModelsError("provider", `Unknown provider: ${model.provider}`);
 			}
 
-			const resolution = await resolveProviderAuth(provider, model, this.credentials, this.authContext, {
+			const resolution = await this.getAuth(model, {
 				apiKey: options?.apiKey,
 				env: options?.env,
 			});

packages/ai/src/index.ts

@@ -21,6 +21,14 @@ export * from "./auth/context.ts";
 export * from "./auth/credential-store.ts";
 export * from "./auth/helpers.ts";
 export * from "./auth/types.ts";
+export type {
+	OAuthAuthInfo,
+	OAuthDeviceCodeInfo,
+	OAuthLoginCallbacks,
+	OAuthPrompt,
+	OAuthSelectOption,
+	OAuthSelectPrompt,
+} from "./compat/extension-oauth-types.ts";
 export * from "./images-models.ts";
 export * from "./models.ts";
 export * from "./providers/faux.ts";
@@ -29,19 +37,6 @@ export * from "./types.ts";
 export * from "./utils/diagnostics.ts";
 export * from "./utils/event-stream.ts";
 export * from "./utils/json-parse.ts";
-export type {
-	OAuthAuthInfo,
-	OAuthCredentials,
-	OAuthDeviceCodeInfo,
-	OAuthLoginCallbacks,
-	OAuthPrompt,
-	OAuthProvider,
-	OAuthProviderId,
-	OAuthProviderInfo,
-	OAuthProviderInterface,
-	OAuthSelectOption,
-	OAuthSelectPrompt,
-} from "./utils/oauth/types.ts";
 export * from "./utils/overflow.ts";
 export * from "./utils/retry.ts";
 export * from "./utils/typebox-helpers.ts";

packages/ai/src/models.ts

@@ -1,8 +1,17 @@
 import { lazyStream } from "./api/lazy.ts";
 import { defaultProviderAuthContext as defaultAuthContext } from "./auth/context.ts";
 import { InMemoryCredentialStore } from "./auth/credential-store.ts";
-import { ModelsError, resolveProviderAuth } from "./auth/resolve.ts";
-import type { AuthContext, AuthResult, CredentialStore, ProviderAuth } from "./auth/types.ts";
+import { type AuthResolutionOverrides, ModelsError, resolveProviderAuth } from "./auth/resolve.ts";
+import type {
+	AuthCheck,
+	AuthContext,
+	AuthInteraction,
+	AuthResult,
+	AuthType,
+	Credential,
+	CredentialStore,
+	ProviderAuth,
+} from "./auth/types.ts";
 import type {
 	Api,
 	ApiStreamOptions,
@@ -19,7 +28,15 @@ import type {
 	Usage,
 } from "./types.ts";
 
-export { type AuthModel, ModelsError, type ModelsErrorCode } from "./auth/resolve.ts";
+export { ModelsError, type ModelsErrorCode } from "./auth/resolve.ts";
+
+export interface ModelsStreamTransforms {
+	/** Transform fully assembled model/auth/request headers before provider dispatch. */
+	transformHeaders?: (headers: ProviderHeaders) => ProviderHeaders | Promise<ProviderHeaders>;
+}
+
+export type ModelsApiStreamOptions<TApi extends Api> = ApiStreamOptions<TApi> & ModelsStreamTransforms;
+export type ModelsSimpleStreamOptions = SimpleStreamOptions & ModelsStreamTransforms;
 
 /**
  * A provider is the concrete runtime unit. It owns id/name/base metadata,
@@ -63,6 +80,13 @@ export interface Provider<TApi extends Api = Api> {
 	 */
 	refreshModels?(): Promise<void>;
 
+	/**
+	 * Optional provider policy for credential-specific model availability.
+	 * `getModels()` remains the complete synchronous catalog; `Models.getAvailable()`
+	 * applies this filter after confirming that provider auth is configured.
+	 */
+	filterModels?(models: readonly Model<TApi>[], credential: Credential | undefined): readonly Model<TApi>[];
+
 	stream<T extends TApi>(
 		model: Model<T>,
 		context: Context,
@@ -101,31 +125,44 @@ export interface Models {
 	 */
 	refresh(provider?: string): Promise<void>;
 
+	/** Check whether a provider has complete auth configuration without refreshing OAuth. */
+	checkAuth(providerId: string): Promise<AuthCheck | undefined>;
+
+	/** Return models whose providers have complete auth configuration. */
+	getAvailable(providerId?: string): Promise<readonly Model<Api>[]>;
+
 	/**
-	 * Resolve request auth for a model. Includes a source label for status UI.
+	 * Resolve provider-scoped auth by provider id, or provider auth plus static
+	 * model headers when passed a model. Includes a source label for status UI.
 	 * Resolves `undefined` when the provider is unknown or unconfigured.
 	 * Rejects with `ModelsError`: code "oauth" when a token refresh fails (the
 	 * stored credential is preserved for retry; re-login fixes it), code "auth"
 	 * when api-key resolution or the credential store fails. Request paths
-	 * surface rejections as stream errors; status/availability UIs catch them
-	 * and render "needs re-login" instead of treating them as unconfigured.
+	 * surface rejections as stream errors.
 	 */
-	getAuth(model: Model<Api>): Promise<AuthResult | undefined>;
+	getAuth(providerId: string, overrides?: AuthResolutionOverrides): Promise<AuthResult | undefined>;
+	getAuth(model: Model<Api>, overrides?: AuthResolutionOverrides): Promise<AuthResult | undefined>;
+
+	/** Run a provider-owned login flow and persist its returned credential. */
+	login(providerId: string, type: AuthType, interaction: AuthInteraction): Promise<Credential>;
+
+	/** Remove the stored credential for a provider. */
+	logout(providerId: string): Promise<void>;
 
 	stream<TApi extends Api>(
 		model: Model<TApi>,
 		context: Context,
-		options?: ApiStreamOptions<TApi>,
+		options?: ModelsApiStreamOptions<TApi>,
 	): AssistantMessageEventStream;
 
 	complete<TApi extends Api>(
 		model: Model<TApi>,
 		context: Context,
-		options?: ApiStreamOptions<TApi>,
+		options?: ModelsApiStreamOptions<TApi>,
 	): Promise<AssistantMessage>;
 
-	streamSimple(model: Model<Api>, context: Context, options?: SimpleStreamOptions): AssistantMessageEventStream;
-	completeSimple(model: Model<Api>, context: Context, options?: SimpleStreamOptions): Promise<AssistantMessage>;
+	streamSimple(model: Model<Api>, context: Context, options?: ModelsSimpleStreamOptions): AssistantMessageEventStream;
+	completeSimple(model: Model<Api>, context: Context, options?: ModelsSimpleStreamOptions): Promise<AssistantMessage>;
 }
 
 export interface MutableModels extends Models {
@@ -140,6 +177,22 @@ export interface CreateModelsOptions {
 	authContext?: AuthContext;
 }
 
+function mergeHeaders(
+	base: ProviderHeaders | undefined,
+	override: ProviderHeaders | undefined,
+): ProviderHeaders | undefined {
+	if (!base && !override) return undefined;
+	const merged = { ...base };
+	for (const [name, value] of Object.entries(override ?? {})) {
+		const lowerName = name.toLowerCase();
+		for (const existingName of Object.keys(merged)) {
+			if (existingName.toLowerCase() === lowerName) delete merged[existingName];
+		}
+		merged[name] = value;
+	}
+	return merged;
+}
+
 class ModelsImpl implements MutableModels {
 	private providers = new Map<string, Provider>();
 	private credentials: CredentialStore;
@@ -214,10 +267,103 @@ class ModelsImpl implements MutableModels {
 		await Promise.allSettled(Array.from(this.providers.values(), async (entry) => entry.refreshModels?.()));
 	}
 
-	async getAuth(model: Model<Api>): Promise<AuthResult | undefined> {
-		const provider = this.providers.get(model.provider);
+	private async readCredential(providerId: string): Promise<Credential | undefined> {
+		try {
+			return await this.credentials.read(providerId);
+		} catch (error) {
+			throw new ModelsError("auth", `Credential store read failed for ${providerId}`, { cause: error });
+		}
+	}
+
+	private async checkProviderAuth(
+		provider: Provider,
+		credential: Credential | undefined,
+	): Promise<AuthCheck | undefined> {
+		if (credential?.type === "oauth") {
+			return provider.auth.oauth ? { source: "OAuth", type: "oauth" } : undefined;
+		}
+		const apiKey = provider.auth.apiKey;
+		if (!apiKey) return undefined;
+		if (apiKey.check) {
+			try {
+				return await apiKey.check({
+					ctx: this.authContext,
+					credential: credential?.type === "api_key" ? credential : undefined,
+				});
+			} catch (error) {
+				throw new ModelsError("auth", `API key auth check failed for provider ${provider.id}`, { cause: error });
+			}
+		}
+
+		const resolution = await resolveProviderAuth(provider, this.credentials, this.authContext);
+		return resolution ? { source: resolution.source, type: "api_key" } : undefined;
+	}
+
+	async checkAuth(providerId: string): Promise<AuthCheck | undefined> {
+		const provider = this.providers.get(providerId);
 		if (!provider) return undefined;
-		return resolveProviderAuth(provider, model, this.credentials, this.authContext);
+		return this.checkProviderAuth(provider, await this.readCredential(providerId));
+	}
+
+	async getAvailable(providerId?: string): Promise<readonly Model<Api>[]> {
+		const providers = providerId
+			? [this.providers.get(providerId)].filter((entry) => entry !== undefined)
+			: this.getProviders();
+		const checks = await Promise.all(
+			providers.map(async (provider) => {
+				const credential = await this.readCredential(provider.id);
+				return { provider, credential, auth: await this.checkProviderAuth(provider, credential) };
+			}),
+		);
+		return checks.flatMap(({ provider, credential, auth }) => {
+			if (!auth) return [];
+			const models = provider.getModels();
+			return provider.filterModels?.(models, credential) ?? models;
+		});
+	}
+
+	getAuth(providerId: string, overrides?: AuthResolutionOverrides): Promise<AuthResult | undefined>;
+	getAuth(model: Model<Api>, overrides?: AuthResolutionOverrides): Promise<AuthResult | undefined>;
+	async getAuth(
+		providerOrModel: string | Model<Api>,
+		overrides?: AuthResolutionOverrides,
+	): Promise<AuthResult | undefined> {
+		const providerId = typeof providerOrModel === "string" ? providerOrModel : providerOrModel.provider;
+		const provider = this.providers.get(providerId);
+		if (!provider) return undefined;
+		const result = await resolveProviderAuth(provider, this.credentials, this.authContext, overrides);
+		if (!result || typeof providerOrModel === "string" || !providerOrModel.headers) return result;
+		return {
+			...result,
+			auth: {
+				...result.auth,
+				headers: mergeHeaders(result.auth.headers, providerOrModel.headers),
+			},
+		};
+	}
+
+	async login(providerId: string, type: AuthType, interaction: AuthInteraction): Promise<Credential> {
+		const provider = this.providers.get(providerId);
+		if (!provider) throw new ModelsError("provider", `Unknown provider: ${providerId}`);
+		const method = type === "oauth" ? provider.auth.oauth : provider.auth.apiKey;
+		if (!method?.login) {
+			throw new ModelsError("auth", `${provider.name} does not support ${type} login`);
+		}
+		const credential = await method.login(interaction);
+		try {
+			await this.credentials.modify(providerId, async () => credential);
+		} catch (error) {
+			throw new ModelsError("auth", `Credential store modify failed for ${providerId}`, { cause: error });
+		}
+		return credential;
+	}
+
+	async logout(providerId: string): Promise<void> {
+		try {
+			await this.credentials.delete(providerId);
+		} catch (error) {
+			throw new ModelsError("auth", `Credential store delete failed for ${providerId}`, { cause: error });
+		}
 	}
 
 	private requireProvider(model: Model<Api>): Provider {
@@ -228,30 +374,28 @@ class ModelsImpl implements MutableModels {
 		return provider;
 	}
 
-	private async applyAuth<TOptions extends StreamOptions>(
+	private async applyAuth<TOptions extends StreamOptions & ModelsStreamTransforms>(
 		model: Model<Api>,
 		options: TOptions | undefined,
-	): Promise<{ requestModel: Model<Api>; requestOptions: TOptions | undefined }> {
-		const resolution = await resolveProviderAuth(
-			this.requireProvider(model),
-			model,
-			this.credentials,
-			this.authContext,
-			{
-				apiKey: options?.apiKey,
-				env: options?.env,
-			},
-		);
-		const auth = resolution?.auth;
-		if (!auth) return { requestModel: model, requestOptions: options };
+	): Promise<{ requestModel: Model<Api>; requestOptions: StreamOptions | undefined }> {
+		this.requireProvider(model);
+		const resolution = await this.getAuth(model, {
+			apiKey: options?.apiKey,
+			env: options?.env,
+		});
+		if (!resolution) {
+			throw new ModelsError("auth", `Provider is not configured: ${model.provider}`);
+		}
+		const auth = resolution.auth;
 
-		const requestModel = auth.baseUrl ? { ...model, baseUrl: auth.baseUrl } : model;
-
-		// Explicit request options win per-field; headers/env merge per key.
+		// Explicit request options win per-field; the Models-only transform runs last.
 		const apiKey = options?.apiKey ?? auth.apiKey;
-		const headers = auth.headers || options?.headers ? { ...auth.headers, ...options?.headers } : undefined;
+		let headers = mergeHeaders(auth.headers, options?.headers);
+		if (options?.transformHeaders) headers = await options.transformHeaders(headers ?? {});
 		const env = resolution.env || options?.env ? { ...(resolution.env ?? {}), ...(options?.env ?? {}) } : undefined;
-		const requestOptions = { ...options, apiKey, headers, env } as TOptions;
+		const requestModel = auth.baseUrl ? { ...model, baseUrl: auth.baseUrl } : model;
+		const { transformHeaders: _transformHeaders, ...providerOptions } = options ?? {};
+		const requestOptions = { ...providerOptions, apiKey, headers, env } as StreamOptions;
 
 		return { requestModel, requestOptions };
 	}
@@ -259,11 +403,14 @@ class ModelsImpl implements MutableModels {
 	stream<TApi extends Api>(
 		model: Model<TApi>,
 		context: Context,
-		options?: ApiStreamOptions<TApi>,
+		options?: ModelsApiStreamOptions<TApi>,
 	): AssistantMessageEventStream {
 		return lazyStream(model, async () => {
 			const provider = this.requireProvider(model);
-			const { requestModel, requestOptions } = await this.applyAuth(model, options as StreamOptions | undefined);
+			const { requestModel, requestOptions } = await this.applyAuth(
+				model,
+				options as ModelsApiStreamOptions<Api> | undefined,
+			);
 			return provider.stream(requestModel as Model<TApi>, context, requestOptions as ApiStreamOptions<TApi>);
 		});
 	}
@@ -271,20 +418,24 @@ class ModelsImpl implements MutableModels {
 	async complete<TApi extends Api>(
 		model: Model<TApi>,
 		context: Context,
-		options?: ApiStreamOptions<TApi>,
+		options?: ModelsApiStreamOptions<TApi>,
 	): Promise<AssistantMessage> {
 		return this.stream(model, context, options).result();
 	}
 
-	streamSimple(model: Model<Api>, context: Context, options?: SimpleStreamOptions): AssistantMessageEventStream {
+	streamSimple(model: Model<Api>, context: Context, options?: ModelsSimpleStreamOptions): AssistantMessageEventStream {
 		return lazyStream(model, async () => {
 			const provider = this.requireProvider(model);
 			const { requestModel, requestOptions } = await this.applyAuth(model, options);
-			return provider.streamSimple(requestModel, context, requestOptions);
+			return provider.streamSimple(requestModel, context, requestOptions as SimpleStreamOptions);
 		});
 	}
 
-	async completeSimple(model: Model<Api>, context: Context, options?: SimpleStreamOptions): Promise<AssistantMessage> {
+	async completeSimple(
+		model: Model<Api>,
+		context: Context,
+		options?: ModelsSimpleStreamOptions,
+	): Promise<AssistantMessage> {
 		return this.streamSimple(model, context, options).result();
 	}
 }
@@ -311,6 +462,7 @@ export interface CreateProviderOptions<TApi extends Api = Api> {
 	 * `Models.refresh(provider)`), and a later call retries.
 	 */
 	refreshModels?: () => Promise<readonly Model<TApi>[]>;
+	filterModels?: (models: readonly Model<TApi>[], credential: Credential | undefined) => readonly Model<TApi>[];
 	/** Single implementation, or map keyed by `model.api` for mixed-API providers. */
 	api: ProviderStreams | Partial<Record<TApi, ProviderStreams>>;
 }
@@ -363,6 +515,7 @@ export function createProvider<TApi extends Api = Api>(input: CreateProviderOpti
 					return inflightRefresh;
 				}
 			: undefined,
+		filterModels: input.filterModels,
 		stream: (model, context, options) => dispatch(model, (streams) => streams.stream(model, context, options)),
 		streamSimple: (model, context, options) =>
 			dispatch(model, (streams) => streams.streamSimple(model, context, options)),

packages/ai/src/oauth.ts

@@ -1 +1,10 @@
-export * from "./utils/oauth/index.ts";
+/** Type-only compatibility entry point for coding-agent extension OAuth declarations. */
+export type {
+	OAuthAuthInfo,
+	OAuthCredentials,
+	OAuthDeviceCodeInfo,
+	OAuthLoginCallbacks,
+	OAuthPrompt,
+	OAuthSelectOption,
+	OAuthSelectPrompt,
+} from "./compat/extension-oauth-types.ts";

packages/ai/src/providers/amazon-bedrock.models.ts

@@ -1290,6 +1290,60 @@ export const AMAZON_BEDROCK_MODELS = {
 		contextWindow: 272000,
 		maxTokens: 128000,
 	} satisfies Model<"bedrock-converse-stream">,
+	"openai.gpt-5.6-luna": {
+		id: "openai.gpt-5.6-luna",
+		name: "GPT-5.6 Luna",
+		api: "bedrock-converse-stream",
+		provider: "amazon-bedrock",
+		baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
+		reasoning: true,
+		thinkingLevelMap: {"xhigh":"xhigh"},
+		input: ["text", "image"],
+		cost: {
+			input: 1,
+			output: 6,
+			cacheRead: 0.1,
+			cacheWrite: 1.25,
+		},
+		contextWindow: 272000,
+		maxTokens: 128000,
+	} satisfies Model<"bedrock-converse-stream">,
+	"openai.gpt-5.6-sol": {
+		id: "openai.gpt-5.6-sol",
+		name: "GPT-5.6 Sol",
+		api: "bedrock-converse-stream",
+		provider: "amazon-bedrock",
+		baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
+		reasoning: true,
+		thinkingLevelMap: {"xhigh":"xhigh"},
+		input: ["text", "image"],
+		cost: {
+			input: 5,
+			output: 30,
+			cacheRead: 0.5,
+			cacheWrite: 6.25,
+		},
+		contextWindow: 272000,
+		maxTokens: 128000,
+	} satisfies Model<"bedrock-converse-stream">,
+	"openai.gpt-5.6-terra": {
+		id: "openai.gpt-5.6-terra",
+		name: "GPT-5.6 Terra",
+		api: "bedrock-converse-stream",
+		provider: "amazon-bedrock",
+		baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
+		reasoning: true,
+		thinkingLevelMap: {"xhigh":"xhigh"},
+		input: ["text", "image"],
+		cost: {
+			input: 2.5,
+			output: 15,
+			cacheRead: 0.25,
+			cacheWrite: 3.125,
+		},
+		contextWindow: 272000,
+		maxTokens: 128000,
+	} satisfies Model<"bedrock-converse-stream">,
 	"openai.gpt-oss-120b": {
 		id: "openai.gpt-oss-120b",
 		name: "gpt-oss-120b",

packages/ai/src/providers/amazon-bedrock.ts

@@ -4,16 +4,61 @@ import { createProvider, type Provider } from "../models.ts";
 import { AMAZON_BEDROCK_MODELS } from "./amazon-bedrock.models.ts";
 
 /**
- * Bedrock auth is ambient: the AWS SDK's default credential chain handles the
- * actual signing, so `resolve` only reports whether the provider is
- * configured. A stored credential key is surfaced as the bearer token.
+ * Bedrock accepts a bearer token or the AWS SDK's default credential chain.
+ * The login flow can store a token/profile choice; resolve also detects ambient
+ * AWS credentials without copying them into pi's credential store.
  */
 const bedrockAuth: ApiKeyAuth = {
-	name: "AWS credentials",
+	name: "AWS credentials or bearer token",
+	login: async (interaction) => {
+		const method = await interaction.prompt({
+			type: "select",
+			message: "Select Amazon Bedrock authentication method:",
+			options: [
+				{ id: "bearer-token", label: "Bearer token" },
+				{ id: "aws-profile", label: "AWS profile" },
+				{ id: "credential-chain", label: "Existing AWS credential chain" },
+			],
+		});
+		if (method === "bearer-token") {
+			return {
+				type: "api_key",
+				key: await interaction.prompt({ type: "secret", message: "Enter Amazon Bedrock bearer token" }),
+			};
+		}
+		interaction.notify({
+			type: "info",
+			message: "Amazon Bedrock supports AWS profiles, IAM credentials, and role-based credentials.",
+			links: [
+				{
+					label: "AWS credential provider chain",
+					url: "https://docs.aws.amazon.com/sdkref/latest/guide/standardized-credentials.html",
+				},
+			],
+		});
+		if (method === "aws-profile") {
+			return {
+				type: "api_key",
+				env: { AWS_PROFILE: await interaction.prompt({ type: "text", message: "Enter AWS profile name" }) },
+			};
+		}
+		if (method !== "credential-chain") throw new Error(`Unknown Amazon Bedrock auth method: ${method}`);
+		await interaction.prompt({
+			type: "text",
+			message: "Configure AWS credentials, then press Enter to continue",
+		});
+		return { type: "api_key" };
+	},
 	resolve: async ({ ctx, credential }) => {
 		if (credential?.key) return { auth: { apiKey: credential.key }, source: "stored credential" };
 		if (await ctx.env("AWS_BEARER_TOKEN_BEDROCK")) return { auth: {}, source: "AWS_BEARER_TOKEN_BEDROCK" };
-		if (await ctx.env("AWS_PROFILE")) return { auth: {}, source: "AWS_PROFILE" };
+		if (credential?.env?.AWS_PROFILE ?? (await ctx.env("AWS_PROFILE"))) {
+			return {
+				auth: {},
+				env: credential?.env,
+				source: credential?.env?.AWS_PROFILE ? "stored credential" : "AWS_PROFILE",
+			};
+		}
 		if ((await ctx.env("AWS_ACCESS_KEY_ID")) && (await ctx.env("AWS_SECRET_ACCESS_KEY"))) {
 			return { auth: {}, source: "AWS access keys" };
 		}

packages/ai/src/providers/anthropic.ts

@@ -1,7 +1,7 @@
 import { anthropicMessagesApi } from "../api/anthropic-messages.lazy.ts";
 import { envApiKeyAuth, lazyOAuth } from "../auth/helpers.ts";
+import { loadAnthropicOAuth } from "../auth/oauth/load.ts";
 import { createProvider, type Provider } from "../models.ts";
-import { loadAnthropicOAuth } from "../utils/oauth/load.ts";
 import { ANTHROPIC_MODELS } from "./anthropic.models.ts";
 
 export function anthropicProvider(): Provider<"anthropic-messages"> {

packages/ai/src/providers/azure-openai-responses.models.ts

@@ -660,6 +660,23 @@ export const AZURE_OPENAI_RESPONSES_MODELS = {
 		contextWindow: 1050000,
 		maxTokens: 128000,
 	} satisfies Model<"azure-openai-responses">,
+	"gpt-realtime-2.1": {
+		id: "gpt-realtime-2.1",
+		name: "GPT-Realtime-2.1",
+		api: "azure-openai-responses",
+		provider: "azure-openai-responses",
+		baseUrl: "",
+		reasoning: true,
+		input: ["text", "image"],
+		cost: {
+			input: 4,
+			output: 24,
+			cacheRead: 0.4,
+			cacheWrite: 0,
+		},
+		contextWindow: 128000,
+		maxTokens: 32000,
+	} satisfies Model<"azure-openai-responses">,
 	"o1": {
 		id: "o1",
 		name: "o1",

packages/ai/src/providers/cerebras.models.ts

@@ -52,7 +52,7 @@ export const CEREBRAS_MODELS = {
 		cost: {
 			input: 2.25,
 			output: 2.75,
-			cacheRead: 0,
+			cacheRead: 2.25,
 			cacheWrite: 0,
 		},
 		contextWindow: 131072,

packages/ai/src/providers/cloudflare-ai-gateway.models.ts

@@ -528,6 +528,60 @@ export const CLOUDFLARE_AI_GATEWAY_MODELS = {
 		contextWindow: 1050000,
 		maxTokens: 128000,
 	} satisfies Model<"openai-responses">,
+	"gpt-5.6-luna": {
+		id: "gpt-5.6-luna",
+		name: "GPT-5.6 Luna",
+		api: "openai-responses",
+		provider: "cloudflare-ai-gateway",
+		baseUrl: "https://gateway.ai.cloudflare.com/v1/{CLOUDFLARE_ACCOUNT_ID}/{CLOUDFLARE_GATEWAY_ID}/openai",
+		reasoning: true,
+		thinkingLevelMap: {"off":null,"xhigh":"xhigh","max":"max"},
+		input: ["text", "image"],
+		cost: {
+			input: 1,
+			output: 6,
+			cacheRead: 0.1,
+			cacheWrite: 0,
+		},
+		contextWindow: 1050000,
+		maxTokens: 128000,
+	} satisfies Model<"openai-responses">,
+	"gpt-5.6-sol": {
+		id: "gpt-5.6-sol",
+		name: "GPT-5.6 Sol",
+		api: "openai-responses",
+		provider: "cloudflare-ai-gateway",
+		baseUrl: "https://gateway.ai.cloudflare.com/v1/{CLOUDFLARE_ACCOUNT_ID}/{CLOUDFLARE_GATEWAY_ID}/openai",
+		reasoning: true,
+		thinkingLevelMap: {"off":null,"xhigh":"xhigh","max":"max"},
+		input: ["text", "image"],
+		cost: {
+			input: 5,
+			output: 30,
+			cacheRead: 0.5,
+			cacheWrite: 0,
+		},
+		contextWindow: 1050000,
+		maxTokens: 128000,
+	} satisfies Model<"openai-responses">,
+	"gpt-5.6-terra": {
+		id: "gpt-5.6-terra",
+		name: "GPT-5.6 Terra",
+		api: "openai-responses",
+		provider: "cloudflare-ai-gateway",
+		baseUrl: "https://gateway.ai.cloudflare.com/v1/{CLOUDFLARE_ACCOUNT_ID}/{CLOUDFLARE_GATEWAY_ID}/openai",
+		reasoning: true,
+		thinkingLevelMap: {"off":null,"xhigh":"xhigh","max":"max"},
+		input: ["text", "image"],
+		cost: {
+			input: 2.5,
+			output: 15,
+			cacheRead: 0.25,
+			cacheWrite: 0,
+		},
+		contextWindow: 1050000,
+		maxTokens: 128000,
+	} satisfies Model<"openai-responses">,
 	"o1": {
 		id: "o1",
 		name: "o1",
@@ -685,4 +739,22 @@ export const CLOUDFLARE_AI_GATEWAY_MODELS = {
 		contextWindow: 131072,
 		maxTokens: 131072,
 	} satisfies Model<"openai-completions">,
+	"workers-ai/@cf/zai-org/glm-5.2": {
+		id: "workers-ai/@cf/zai-org/glm-5.2",
+		name: "Glm 5.2",
+		api: "openai-completions",
+		provider: "cloudflare-ai-gateway",
+		baseUrl: "https://gateway.ai.cloudflare.com/v1/{CLOUDFLARE_ACCOUNT_ID}/{CLOUDFLARE_GATEWAY_ID}/compat",
+		compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false,"sendSessionAffinityHeaders":true},
+		reasoning: true,
+		input: ["text"],
+		cost: {
+			input: 1.4,
+			output: 4.4,
+			cacheRead: 0.26,
+			cacheWrite: 0,
+		},
+		contextWindow: 262144,
+		maxTokens: 262144,
+	} satisfies Model<"openai-completions">,
 } as const;

packages/ai/src/providers/cloudflare-ai-gateway.ts

@@ -4,6 +4,7 @@ import { openAIResponsesApi } from "../api/openai-responses.lazy.ts";
 import { createProvider, type Provider } from "../models.ts";
 import { CLOUDFLARE_AI_GATEWAY_MODELS } from "./cloudflare-ai-gateway.models.ts";
 import { cloudflareAIGatewayAuth } from "./cloudflare-auth.ts";
+import { cloudflareStreams } from "./cloudflare-stream.ts";
 
 export function cloudflareAIGatewayProvider(): Provider<
 	"anthropic-messages" | "openai-completions" | "openai-responses"
@@ -14,9 +15,9 @@ export function cloudflareAIGatewayProvider(): Provider<
 		auth: { apiKey: cloudflareAIGatewayAuth() },
 		models: Object.values(CLOUDFLARE_AI_GATEWAY_MODELS),
 		api: {
-			"anthropic-messages": anthropicMessagesApi(),
-			"openai-completions": openAICompletionsApi(),
-			"openai-responses": openAIResponsesApi(),
+			"anthropic-messages": cloudflareStreams(anthropicMessagesApi()),
+			"openai-completions": cloudflareStreams(openAICompletionsApi()),
+			"openai-responses": cloudflareStreams(openAIResponsesApi()),
 		},
 	});
 }

packages/ai/src/providers/cloudflare-auth.ts

@@ -1,5 +1,5 @@
 import type { ApiKeyAuth, ApiKeyCredential, AuthContext } from "../auth/types.ts";
-import type { Api, ImagesApi, ImagesModel, Model, ProviderEnv } from "../types.ts";
+import type { ProviderEnv } from "../types.ts";
 
 const CLOUDFLARE_API_KEY = "CLOUDFLARE_API_KEY";
 const CLOUDFLARE_ACCOUNT_ID = "CLOUDFLARE_ACCOUNT_ID";
@@ -19,22 +19,11 @@ async function resolveValue(
 	return ctx.env(name);
 }
 
-function resolveCloudflareBaseUrl(
-	model: Model<Api> | ImagesModel<ImagesApi>,
-	accountId: string,
-	gatewayId: string | undefined,
-): string {
-	return model.baseUrl
-		.replaceAll(`{${CLOUDFLARE_ACCOUNT_ID}}`, accountId)
-		.replaceAll(`{${CLOUDFLARE_GATEWAY_ID}}`, gatewayId ?? "");
-}
-
 async function resolveCloudflareEnv(
 	kind: CloudflareAuthKind,
-	model: Model<Api> | ImagesModel<ImagesApi>,
 	ctx: AuthContext,
 	credential: ApiKeyCredential | undefined,
-): Promise<{ apiKey: string; env: ProviderEnv; baseUrl: string; source: string } | undefined> {
+): Promise<{ apiKey: string; env: ProviderEnv; source: string } | undefined> {
 	const apiKey = await resolveValue(CLOUDFLARE_API_KEY, ctx, credential);
 	const accountId = await resolveValue(CLOUDFLARE_ACCOUNT_ID, ctx, credential);
 	const gatewayId = kind === "ai-gateway" ? await resolveValue(CLOUDFLARE_GATEWAY_ID, ctx, credential) : undefined;
@@ -47,7 +36,6 @@ async function resolveCloudflareEnv(
 			CLOUDFLARE_ACCOUNT_ID: accountId,
 			...(gatewayId ? { CLOUDFLARE_GATEWAY_ID: gatewayId } : {}),
 		},
-		baseUrl: resolveCloudflareBaseUrl(model, accountId, gatewayId),
 		source: credential ? "stored credential" : CLOUDFLARE_API_KEY,
 	};
 }
@@ -55,16 +43,16 @@ async function resolveCloudflareEnv(
 export function cloudflareWorkersAIAuth(): ApiKeyAuth {
 	return {
 		name: "Cloudflare API key",
-		login: async (callbacks) => {
-			const key = await callbacks.prompt({ type: "secret", message: "Enter Cloudflare API key" });
-			const accountId = await callbacks.prompt({ type: "text", message: "Enter Cloudflare account ID" });
+		login: async (interaction) => {
+			const key = await interaction.prompt({ type: "secret", message: "Enter Cloudflare API key" });
+			const accountId = await interaction.prompt({ type: "text", message: "Enter Cloudflare account ID" });
 			return { type: "api_key", key, env: { CLOUDFLARE_ACCOUNT_ID: accountId } };
 		},
-		resolve: async ({ model, ctx, credential }) => {
-			const resolved = await resolveCloudflareEnv("workers-ai", model, ctx, credential);
+		resolve: async ({ ctx, credential }) => {
+			const resolved = await resolveCloudflareEnv("workers-ai", ctx, credential);
 			if (!resolved) return undefined;
 			return {
-				auth: { apiKey: resolved.apiKey, baseUrl: resolved.baseUrl },
+				auth: { apiKey: resolved.apiKey },
 				env: resolved.env,
 				source: resolved.source,
 			};
@@ -75,18 +63,18 @@ export function cloudflareWorkersAIAuth(): ApiKeyAuth {
 export function cloudflareAIGatewayAuth(): ApiKeyAuth {
 	return {
 		name: "Cloudflare API key",
-		login: async (callbacks) => {
-			const key = await callbacks.prompt({ type: "secret", message: "Enter Cloudflare API key" });
-			const accountId = await callbacks.prompt({ type: "text", message: "Enter Cloudflare account ID" });
-			const gatewayId = await callbacks.prompt({ type: "text", message: "Enter Cloudflare AI Gateway ID" });
+		login: async (interaction) => {
+			const key = await interaction.prompt({ type: "secret", message: "Enter Cloudflare API key" });
+			const accountId = await interaction.prompt({ type: "text", message: "Enter Cloudflare account ID" });
+			const gatewayId = await interaction.prompt({ type: "text", message: "Enter Cloudflare AI Gateway ID" });
 			return {
 				type: "api_key",
 				key,
 				env: { CLOUDFLARE_ACCOUNT_ID: accountId, CLOUDFLARE_GATEWAY_ID: gatewayId },
 			};
 		},
-		resolve: async ({ model, ctx, credential }) => {
-			const resolved = await resolveCloudflareEnv("ai-gateway", model, ctx, credential);
+		resolve: async ({ ctx, credential }) => {
+			const resolved = await resolveCloudflareEnv("ai-gateway", ctx, credential);
 			if (!resolved) return undefined;
 			return {
 				auth: {
@@ -95,7 +83,6 @@ export function cloudflareAIGatewayAuth(): ApiKeyAuth {
 						Authorization: null,
 						"x-api-key": null,
 					},
-					baseUrl: resolved.baseUrl,
 				},
 				env: resolved.env,
 				source: resolved.source,

packages/ai/src/providers/cloudflare-stream.ts

@@ -0,0 +1,28 @@
+import type { Api, Model, ProviderEnv, ProviderStreams } from "../types.ts";
+
+const CLOUDFLARE_ACCOUNT_ID = "CLOUDFLARE_ACCOUNT_ID";
+const CLOUDFLARE_GATEWAY_ID = "CLOUDFLARE_GATEWAY_ID";
+
+export function resolveCloudflareModel<TApi extends Api>(
+	model: Model<TApi>,
+	env: ProviderEnv | undefined,
+): Model<TApi> {
+	if (!env) return model;
+	const baseUrl = model.baseUrl
+		.replaceAll(`{${CLOUDFLARE_ACCOUNT_ID}}`, env[CLOUDFLARE_ACCOUNT_ID] ?? `{${CLOUDFLARE_ACCOUNT_ID}}`)
+		.replaceAll(`{${CLOUDFLARE_GATEWAY_ID}}`, env[CLOUDFLARE_GATEWAY_ID] ?? `{${CLOUDFLARE_GATEWAY_ID}}`);
+	return baseUrl === model.baseUrl ? model : { ...model, baseUrl };
+}
+
+/**
+ * Wrap an API implementation so Cloudflare account/gateway endpoint
+ * placeholders materialize from the resolved provider env before dispatch.
+ */
+export function cloudflareStreams(streams: ProviderStreams): ProviderStreams {
+	return {
+		stream: (model, context, options) =>
+			streams.stream(resolveCloudflareModel(model, options?.env), context, options),
+		streamSimple: (model, context, options) =>
+			streams.streamSimple(resolveCloudflareModel(model, options?.env), context, options),
+	};
+}

packages/ai/src/providers/cloudflare-workers-ai.ts

@@ -1,6 +1,7 @@
 import { openAICompletionsApi } from "../api/openai-completions.lazy.ts";
 import { createProvider, type Provider } from "../models.ts";
 import { cloudflareWorkersAIAuth } from "./cloudflare-auth.ts";
+import { cloudflareStreams } from "./cloudflare-stream.ts";
 import { CLOUDFLARE_WORKERS_AI_MODELS } from "./cloudflare-workers-ai.models.ts";
 
 export function cloudflareWorkersAIProvider(): Provider<"openai-completions"> {
@@ -9,6 +10,6 @@ export function cloudflareWorkersAIProvider(): Provider<"openai-completions"> {
 		name: "Cloudflare Workers AI",
 		auth: { apiKey: cloudflareWorkersAIAuth() },
 		models: Object.values(CLOUDFLARE_WORKERS_AI_MODELS),
-		api: openAICompletionsApi(),
+		api: cloudflareStreams(openAICompletionsApi()),
 	});
 }

packages/ai/src/providers/github-copilot.models.ts

@@ -252,7 +252,7 @@ export const GITHUB_COPILOT_MODELS = {
 			cacheRead: 0.2,
 			cacheWrite: 0,
 		},
-		contextWindow: 200000,
+		contextWindow: 1000000,
 		maxTokens: 64000,
 	} satisfies Model<"openai-completions">,
 	"gemini-3.5-flash": {
@@ -445,6 +445,63 @@ export const GITHUB_COPILOT_MODELS = {
 		contextWindow: 1000000,
 		maxTokens: 128000,
 	} satisfies Model<"openai-responses">,
+	"gpt-5.6-luna": {
+		id: "gpt-5.6-luna",
+		name: "GPT-5.6 Luna",
+		api: "openai-responses",
+		provider: "github-copilot",
+		baseUrl: "https://api.individual.githubcopilot.com",
+		headers: {"User-Agent":"GitHubCopilotChat/0.35.0","Editor-Version":"vscode/1.107.0","Editor-Plugin-Version":"copilot-chat/0.35.0","Copilot-Integration-Id":"vscode-chat"},
+		reasoning: true,
+		thinkingLevelMap: {"off":null,"minimal":"low","xhigh":"xhigh","max":"max"},
+		input: ["text", "image"],
+		cost: {
+			input: 1,
+			output: 6,
+			cacheRead: 0.1,
+			cacheWrite: 0,
+		},
+		contextWindow: 1050000,
+		maxTokens: 128000,
+	} satisfies Model<"openai-responses">,
+	"gpt-5.6-sol": {
+		id: "gpt-5.6-sol",
+		name: "GPT-5.6 Sol",
+		api: "openai-responses",
+		provider: "github-copilot",
+		baseUrl: "https://api.individual.githubcopilot.com",
+		headers: {"User-Agent":"GitHubCopilotChat/0.35.0","Editor-Version":"vscode/1.107.0","Editor-Plugin-Version":"copilot-chat/0.35.0","Copilot-Integration-Id":"vscode-chat"},
+		reasoning: true,
+		thinkingLevelMap: {"off":null,"minimal":"low","xhigh":"xhigh","max":"max"},
+		input: ["text", "image"],
+		cost: {
+			input: 5,
+			output: 30,
+			cacheRead: 0.5,
+			cacheWrite: 0,
+		},
+		contextWindow: 1050000,
+		maxTokens: 128000,
+	} satisfies Model<"openai-responses">,
+	"gpt-5.6-terra": {
+		id: "gpt-5.6-terra",
+		name: "GPT-5.6 Terra",
+		api: "openai-responses",
+		provider: "github-copilot",
+		baseUrl: "https://api.individual.githubcopilot.com",
+		headers: {"User-Agent":"GitHubCopilotChat/0.35.0","Editor-Version":"vscode/1.107.0","Editor-Plugin-Version":"copilot-chat/0.35.0","Copilot-Integration-Id":"vscode-chat"},
+		reasoning: true,
+		thinkingLevelMap: {"off":null,"minimal":"low","xhigh":"xhigh","max":"max"},
+		input: ["text", "image"],
+		cost: {
+			input: 2.5,
+			output: 15,
+			cacheRead: 0.25,
+			cacheWrite: 0,
+		},
+		contextWindow: 1050000,
+		maxTokens: 128000,
+	} satisfies Model<"openai-responses">,
 	"kimi-k2.7-code": {
 		id: "kimi-k2.7-code",
 		name: "Kimi K2.7 Code",

packages/ai/src/providers/github-copilot.ts

@@ -2,8 +2,8 @@ import { anthropicMessagesApi } from "../api/anthropic-messages.lazy.ts";
 import { openAICompletionsApi } from "../api/openai-completions.lazy.ts";
 import { openAIResponsesApi } from "../api/openai-responses.lazy.ts";
 import { envApiKeyAuth, lazyOAuth } from "../auth/helpers.ts";
+import { loadGitHubCopilotOAuth } from "../auth/oauth/load.ts";
 import { createProvider, type Provider } from "../models.ts";
-import { loadGitHubCopilotOAuth } from "../utils/oauth/load.ts";
 import { GITHUB_COPILOT_MODELS } from "./github-copilot.models.ts";
 
 export function githubCopilotProvider(): Provider<"anthropic-messages" | "openai-completions" | "openai-responses"> {
@@ -16,6 +16,15 @@ export function githubCopilotProvider(): Provider<"anthropic-messages" | "openai
 			oauth: lazyOAuth({ name: "GitHub Copilot", load: loadGitHubCopilotOAuth }),
 		},
 		models: Object.values(GITHUB_COPILOT_MODELS),
+		filterModels: (models, credential) => {
+			if (credential?.type !== "oauth") return models;
+			const availableModelIds = credential.availableModelIds;
+			if (!Array.isArray(availableModelIds) || !availableModelIds.every((id) => typeof id === "string")) {
+				return models;
+			}
+			const available = new Set(availableModelIds);
+			return models.filter((model) => available.has(model.id));
+		},
 		api: {
 			"anthropic-messages": anthropicMessagesApi(),
 			"openai-completions": openAICompletionsApi(),

packages/ai/src/providers/google-vertex.ts

@@ -12,16 +12,71 @@ const VERTEX_ADC_PATH = "~/.config/gcloud/application_default_credentials.json";
  */
 const vertexAuth: ApiKeyAuth = {
 	name: "Google Cloud credentials",
+	login: async (interaction) => {
+		const method = await interaction.prompt({
+			type: "select",
+			message: "Select Google Vertex AI authentication method:",
+			options: [
+				{ id: "api-key", label: "Google Cloud API key" },
+				{ id: "adc", label: "Application Default Credentials" },
+				{ id: "service-account", label: "Service account credentials file" },
+			],
+		});
+		if (method === "api-key") {
+			return {
+				type: "api_key",
+				key: await interaction.prompt({ type: "secret", message: "Enter Google Cloud API key" }),
+			};
+		}
+		if (method !== "adc" && method !== "service-account") {
+			throw new Error(`Unknown Google Vertex AI auth method: ${method}`);
+		}
+		interaction.notify({
+			type: "info",
+			message:
+				method === "adc"
+					? "Run `gcloud auth application-default login`, then provide the project and location."
+					: "Provide a service account credentials file, project, and location.",
+			links: [
+				{
+					label: "Application Default Credentials",
+					url: "https://cloud.google.com/docs/authentication/provide-credentials-adc",
+				},
+			],
+		});
+		const project = await interaction.prompt({ type: "text", message: "Enter Google Cloud project ID" });
+		const location = await interaction.prompt({ type: "text", message: "Enter Google Cloud location" });
+		const credentialsPath =
+			method === "service-account"
+				? await interaction.prompt({ type: "text", message: "Enter service account credentials file path" })
+				: undefined;
+		return {
+			type: "api_key",
+			env: {
+				GOOGLE_CLOUD_PROJECT: project,
+				GOOGLE_CLOUD_LOCATION: location,
+				...(credentialsPath ? { GOOGLE_APPLICATION_CREDENTIALS: credentialsPath } : {}),
+			},
+		};
+	},
 	resolve: async ({ ctx, credential }) => {
 		const key = credential?.key ?? (await ctx.env("GOOGLE_CLOUD_API_KEY"));
 		if (key) return { auth: { apiKey: key }, source: credential?.key ? "stored credential" : "GOOGLE_CLOUD_API_KEY" };
 
-		const adcPath = await ctx.env("GOOGLE_APPLICATION_CREDENTIALS");
+		const adcPath =
+			credential?.env?.GOOGLE_APPLICATION_CREDENTIALS ?? (await ctx.env("GOOGLE_APPLICATION_CREDENTIALS"));
 		const hasCredentials = await ctx.fileExists(adcPath ?? VERTEX_ADC_PATH);
-		const hasProject = Boolean((await ctx.env("GOOGLE_CLOUD_PROJECT")) ?? (await ctx.env("GCLOUD_PROJECT")));
-		const hasLocation = Boolean(await ctx.env("GOOGLE_CLOUD_LOCATION"));
-		if (hasCredentials && hasProject && hasLocation) {
-			return { auth: {}, source: "gcloud application default credentials" };
+		const project =
+			credential?.env?.GOOGLE_CLOUD_PROJECT ??
+			(await ctx.env("GOOGLE_CLOUD_PROJECT")) ??
+			(await ctx.env("GCLOUD_PROJECT"));
+		const location = credential?.env?.GOOGLE_CLOUD_LOCATION ?? (await ctx.env("GOOGLE_CLOUD_LOCATION"));
+		if (hasCredentials && project && location) {
+			return {
+				auth: {},
+				env: credential?.env,
+				source: credential ? "stored credential" : "gcloud application default credentials",
+			};
 		}
 		return undefined;
 	},

packages/ai/src/providers/openai-codex.ts

@@ -1,7 +1,7 @@
 import { openAICodexResponsesApi } from "../api/openai-codex-responses.lazy.ts";
 import { lazyOAuth } from "../auth/helpers.ts";
+import { loadOpenAICodexOAuth } from "../auth/oauth/load.ts";
 import { createProvider, type Provider } from "../models.ts";
-import { loadOpenAICodexOAuth } from "../utils/oauth/load.ts";
 import { OPENAI_CODEX_MODELS } from "./openai-codex.models.ts";
 
 export function openaiCodexProvider(): Provider<"openai-codex-responses"> {

packages/ai/src/providers/openai.models.ts

@@ -667,6 +667,23 @@ export const OPENAI_MODELS = {
 		contextWindow: 272000,
 		maxTokens: 128000,
 	} satisfies Model<"openai-responses">,
+	"gpt-realtime-2.1": {
+		id: "gpt-realtime-2.1",
+		name: "GPT-Realtime-2.1",
+		api: "openai-responses",
+		provider: "openai",
+		baseUrl: "https://api.openai.com/v1",
+		reasoning: true,
+		input: ["text", "image"],
+		cost: {
+			input: 4,
+			output: 24,
+			cacheRead: 0.4,
+			cacheWrite: 0,
+		},
+		contextWindow: 128000,
+		maxTokens: 32000,
+	} satisfies Model<"openai-responses">,
 	"o1": {
 		id: "o1",
 		name: "o1",

packages/ai/src/providers/opencode.models.ts

@@ -674,6 +674,60 @@ export const OPENCODE_MODELS = {
 		contextWindow: 1050000,
 		maxTokens: 128000,
 	} satisfies Model<"openai-responses">,
+	"gpt-5.6-luna": {
+		id: "gpt-5.6-luna",
+		name: "GPT-5.6 Luna",
+		api: "openai-responses",
+		provider: "opencode",
+		baseUrl: "https://opencode.ai/zen/v1",
+		reasoning: true,
+		thinkingLevelMap: {"off":null,"xhigh":"xhigh","max":"max"},
+		input: ["text", "image"],
+		cost: {
+			input: 1,
+			output: 6,
+			cacheRead: 0.1,
+			cacheWrite: 1.25,
+		},
+		contextWindow: 1050000,
+		maxTokens: 128000,
+	} satisfies Model<"openai-responses">,
+	"gpt-5.6-sol": {
+		id: "gpt-5.6-sol",
+		name: "GPT-5.6 Sol",
+		api: "openai-responses",
+		provider: "opencode",
+		baseUrl: "https://opencode.ai/zen/v1",
+		reasoning: true,
+		thinkingLevelMap: {"off":null,"xhigh":"xhigh","max":"max"},
+		input: ["text", "image"],
+		cost: {
+			input: 5,
+			output: 30,
+			cacheRead: 0.5,
+			cacheWrite: 6.25,
+		},
+		contextWindow: 1050000,
+		maxTokens: 128000,
+	} satisfies Model<"openai-responses">,
+	"gpt-5.6-terra": {
+		id: "gpt-5.6-terra",
+		name: "GPT-5.6 Terra",
+		api: "openai-responses",
+		provider: "opencode",
+		baseUrl: "https://opencode.ai/zen/v1",
+		reasoning: true,
+		thinkingLevelMap: {"off":null,"xhigh":"xhigh","max":"max"},
+		input: ["text", "image"],
+		cost: {
+			input: 2.5,
+			output: 15,
+			cacheRead: 0.25,
+			cacheWrite: 3.125,
+		},
+		contextWindow: 1050000,
+		maxTokens: 128000,
+	} satisfies Model<"openai-responses">,
 	"grok-4.5": {
 		id: "grok-4.5",
 		name: "Grok 4.5",
@@ -726,7 +780,7 @@ export const OPENCODE_MODELS = {
 			cacheRead: 0,
 			cacheWrite: 0,
 		},
-		contextWindow: 256000,
+		contextWindow: 190000,
 		maxTokens: 64000,
 	} satisfies Model<"openai-completions">,
 	"kimi-k2.5": {

packages/ai/src/providers/openrouter.models.ts

@@ -461,24 +461,6 @@ export const OPENROUTER_MODELS = {
 		contextWindow: 262144,
 		maxTokens: 80000,
 	} satisfies Model<"openai-completions">,
-	"arcee-ai/trinity-mini": {
-		id: "arcee-ai/trinity-mini",
-		name: "Arcee AI: Trinity Mini",
-		api: "openai-completions",
-		provider: "openrouter",
-		baseUrl: "https://openrouter.ai/api/v1",
-		compat: {"supportsDeveloperRole":false,"thinkingFormat":"openrouter"},
-		reasoning: true,
-		input: ["text"],
-		cost: {
-			input: 0.045,
-			output: 0.15,
-			cacheRead: 0,
-			cacheWrite: 0,
-		},
-		contextWindow: 131072,
-		maxTokens: 131072,
-	} satisfies Model<"openai-completions">,
 	"arcee-ai/virtuoso-large": {
 		id: "arcee-ai/virtuoso-large",
 		name: "Arcee AI: Virtuoso Large",
@@ -687,8 +669,8 @@ export const OPENROUTER_MODELS = {
 		reasoning: true,
 		input: ["text"],
 		cost: {
-			input: 0.21,
-			output: 0.79,
+			input: 0.25,
+			output: 0.95,
 			cacheRead: 0.13,
 			cacheWrite: 0,
 		},
@@ -759,9 +741,9 @@ export const OPENROUTER_MODELS = {
 		reasoning: true,
 		input: ["text"],
 		cost: {
-			input: 0.2288,
-			output: 0.3432,
-			cacheRead: 0.02288,
+			input: 0.2145,
+			output: 0.32175,
+			cacheRead: 0.02145,
 			cacheWrite: 0,
 		},
 		contextWindow: 131072,
@@ -1121,13 +1103,13 @@ export const OPENROUTER_MODELS = {
 		reasoning: true,
 		input: ["text", "image"],
 		cost: {
-			input: 0.12,
+			input: 0.06,
 			output: 0.35,
-			cacheRead: 0.09,
+			cacheRead: 0,
 			cacheWrite: 0,
 		},
 		contextWindow: 262144,
-		maxTokens: 262144,
+		maxTokens: 8192,
 	} satisfies Model<"openai-completions">,
 	"google/gemma-4-31b-it:free": {
 		id: "google/gemma-4-31b-it:free",
@@ -1145,7 +1127,7 @@ export const OPENROUTER_MODELS = {
 			cacheWrite: 0,
 		},
 		contextWindow: 262144,
-		maxTokens: 8192,
+		maxTokens: 32768,
 	} satisfies Model<"openai-completions">,
 	"ibm-granite/granite-4.1-8b": {
 		id: "ibm-granite/granite-4.1-8b",
@@ -1238,6 +1220,24 @@ export const OPENROUTER_MODELS = {
 		contextWindow: 262144,
 		maxTokens: 65536,
 	} satisfies Model<"openai-completions">,
+	"kwaipilot/kat-coder-air-v2.5": {
+		id: "kwaipilot/kat-coder-air-v2.5",
+		name: "Kwaipilot: KAT-Coder-Air V2.5",
+		api: "openai-completions",
+		provider: "openrouter",
+		baseUrl: "https://openrouter.ai/api/v1",
+		compat: {"supportsDeveloperRole":false,"thinkingFormat":"openrouter"},
+		reasoning: false,
+		input: ["text"],
+		cost: {
+			input: 0.15,
+			output: 0.6,
+			cacheRead: 0.03,
+			cacheWrite: 0,
+		},
+		contextWindow: 256000,
+		maxTokens: 80000,
+	} satisfies Model<"openai-completions">,
 	"kwaipilot/kat-coder-pro-v2": {
 		id: "kwaipilot/kat-coder-pro-v2",
 		name: "Kwaipilot: KAT-Coder-Pro V2",
@@ -1256,23 +1256,23 @@ export const OPENROUTER_MODELS = {
 		contextWindow: 256000,
 		maxTokens: 80000,
 	} satisfies Model<"openai-completions">,
-	"liquid/lfm-2.5-1.2b-thinking:free": {
-		id: "liquid/lfm-2.5-1.2b-thinking:free",
-		name: "LiquidAI: LFM2.5-1.2B-Thinking (free)",
+	"kwaipilot/kat-coder-pro-v2.5": {
+		id: "kwaipilot/kat-coder-pro-v2.5",
+		name: "Kwaipilot: KAT-Coder-Pro V2.5",
 		api: "openai-completions",
 		provider: "openrouter",
 		baseUrl: "https://openrouter.ai/api/v1",
 		compat: {"supportsDeveloperRole":false,"thinkingFormat":"openrouter"},
-		reasoning: true,
+		reasoning: false,
 		input: ["text"],
 		cost: {
-			input: 0,
-			output: 0,
-			cacheRead: 0,
+			input: 0.74,
+			output: 2.96,
+			cacheRead: 0.15,
 			cacheWrite: 0,
 		},
-		contextWindow: 32768,
-		maxTokens: 4096,
+		contextWindow: 256000,
+		maxTokens: 80000,
 	} satisfies Model<"openai-completions">,
 	"meta-llama/llama-3.1-70b-instruct": {
 		id: "meta-llama/llama-3.1-70b-instruct",
@@ -1356,8 +1356,8 @@ export const OPENROUTER_MODELS = {
 		reasoning: false,
 		input: ["text", "image"],
 		cost: {
-			input: 0.15,
-			output: 0.6,
+			input: 0.2,
+			output: 0.8,
 			cacheRead: 0,
 			cacheWrite: 0,
 		},
@@ -1878,9 +1878,9 @@ export const OPENROUTER_MODELS = {
 		reasoning: true,
 		input: ["text", "image"],
 		cost: {
-			input: 0.65,
+			input: 0.66,
 			output: 3.41,
-			cacheRead: 0.14,
+			cacheRead: 0.15,
 			cacheWrite: 0,
 		},
 		contextWindow: 262144,
@@ -1896,9 +1896,9 @@ export const OPENROUTER_MODELS = {
 		reasoning: true,
 		input: ["text", "image"],
 		cost: {
-			input: 0.72,
+			input: 0.719,
 			output: 3.49,
-			cacheRead: 0.159,
+			cacheRead: 0.149,
 			cacheWrite: 0,
 		},
 		contextWindow: 262144,
@@ -2456,11 +2456,11 @@ export const OPENROUTER_MODELS = {
 		cost: {
 			input: 0.05,
 			output: 0.4,
-			cacheRead: 0.01,
+			cacheRead: 0.005,
 			cacheWrite: 0,
 		},
 		contextWindow: 400000,
-		maxTokens: 4096,
+		maxTokens: 128000,
 	} satisfies Model<"openai-completions">,
 	"openai/gpt-5-pro": {
 		id: "openai/gpt-5-pro",
@@ -2492,7 +2492,7 @@ export const OPENROUTER_MODELS = {
 		cost: {
 			input: 1.25,
 			output: 10,
-			cacheRead: 0.13,
+			cacheRead: 0.125,
 			cacheWrite: 0,
 		},
 		contextWindow: 400000,
@@ -2976,26 +2976,8 @@ export const OPENROUTER_MODELS = {
 		reasoning: true,
 		input: ["text"],
 		cost: {
-			input: 0.036,
-			output: 0.18,
-			cacheRead: 0,
-			cacheWrite: 0,
-		},
-		contextWindow: 131072,
-		maxTokens: 4096,
-	} satisfies Model<"openai-completions">,
-	"openai/gpt-oss-120b:free": {
-		id: "openai/gpt-oss-120b:free",
-		name: "OpenAI: gpt-oss-120b (free)",
-		api: "openai-completions",
-		provider: "openrouter",
-		baseUrl: "https://openrouter.ai/api/v1",
-		compat: {"thinkingFormat":"openrouter"},
-		reasoning: true,
-		input: ["text"],
-		cost: {
-			input: 0,
-			output: 0,
+			input: 0.03,
+			output: 0.15,
 			cacheRead: 0,
 			cacheWrite: 0,
 		},
@@ -3481,7 +3463,7 @@ export const OPENROUTER_MODELS = {
 		input: ["text"],
 		cost: {
 			input: 0.09,
-			output: 0.1,
+			output: 0.55,
 			cacheRead: 0,
 			cacheWrite: 0,
 		},
@@ -4074,13 +4056,13 @@ export const OPENROUTER_MODELS = {
 		reasoning: true,
 		input: ["text", "image"],
 		cost: {
-			input: 0.285,
+			input: 0.289,
 			output: 2.4,
-			cacheRead: 0.15,
+			cacheRead: 0,
 			cacheWrite: 0,
 		},
 		contextWindow: 262144,
-		maxTokens: 262140,
+		maxTokens: 131072,
 	} satisfies Model<"openai-completions">,
 	"qwen/qwen3.6-35b-a3b": {
 		id: "qwen/qwen3.6-35b-a3b",
@@ -4561,12 +4543,12 @@ export const OPENROUTER_MODELS = {
 		input: ["text"],
 		cost: {
 			input: 0.43,
-			output: 1.74,
+			output: 1.75,
 			cacheRead: 0.08,
 			cacheWrite: 0,
 		},
-		contextWindow: 202752,
-		maxTokens: 131072,
+		contextWindow: 200000,
+		maxTokens: 16384,
 	} satisfies Model<"openai-completions">,
 	"z-ai/glm-4.6v": {
 		id: "z-ai/glm-4.6v",
@@ -4638,7 +4620,7 @@ export const OPENROUTER_MODELS = {
 			cacheWrite: 0,
 		},
 		contextWindow: 202752,
-		maxTokens: 4096,
+		maxTokens: 128000,
 	} satisfies Model<"openai-completions">,
 	"z-ai/glm-5-turbo": {
 		id: "z-ai/glm-5-turbo",
@@ -4687,13 +4669,13 @@ export const OPENROUTER_MODELS = {
 		thinkingLevelMap: {"xhigh":"xhigh"},
 		input: ["text"],
 		cost: {
-			input: 0.54,
-			output: 1.76,
-			cacheRead: 0.1,
+			input: 0.924,
+			output: 2.904,
+			cacheRead: 0.1716,
 			cacheWrite: 0,
 		},
 		contextWindow: 1048576,
-		maxTokens: 101376,
+		maxTokens: 131072,
 	} satisfies Model<"openai-completions">,
 	"z-ai/glm-5v-turbo": {
 		id: "z-ai/glm-5v-turbo",
@@ -4831,9 +4813,9 @@ export const OPENROUTER_MODELS = {
 		reasoning: true,
 		input: ["text", "image"],
 		cost: {
-			input: 0.65,
+			input: 0.66,
 			output: 3.41,
-			cacheRead: 0.14,
+			cacheRead: 0.15,
 			cacheWrite: 0,
 		},
 		contextWindow: 262144,

packages/ai/src/providers/vercel-ai-gateway.models.ts

@@ -497,23 +497,6 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		contextWindow: 200000,
 		maxTokens: 4096,
 	} satisfies Model<"anthropic-messages">,
-	"anthropic/claude-3.5-haiku": {
-		id: "anthropic/claude-3.5-haiku",
-		name: "Claude 3.5 Haiku",
-		api: "anthropic-messages",
-		provider: "vercel-ai-gateway",
-		baseUrl: "https://ai-gateway.vercel.sh",
-		reasoning: false,
-		input: ["text", "image"],
-		cost: {
-			input: 0.8,
-			output: 4,
-			cacheRead: 0.08,
-			cacheWrite: 1,
-		},
-		contextWindow: 200000,
-		maxTokens: 8192,
-	} satisfies Model<"anthropic-messages">,
 	"anthropic/claude-fable-5": {
 		id: "anthropic/claude-fable-5",
 		name: "Claude Fable 5",
@@ -565,7 +548,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 			cacheWrite: 18.75,
 		},
 		contextWindow: 200000,
-		maxTokens: 32000,
+		maxTokens: 8192,
 	} satisfies Model<"anthropic-messages">,
 	"anthropic/claude-opus-4.1": {
 		id: "anthropic/claude-opus-4.1",
@@ -673,7 +656,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 			cacheWrite: 3.75,
 		},
 		contextWindow: 1000000,
-		maxTokens: 64000,
+		maxTokens: 8192,
 	} satisfies Model<"anthropic-messages">,
 	"anthropic/claude-sonnet-4.5": {
 		id: "anthropic/claude-sonnet-4.5",
@@ -730,23 +713,6 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		contextWindow: 1000000,
 		maxTokens: 128000,
 	} satisfies Model<"anthropic-messages">,
-	"arcee-ai/trinity-large-preview": {
-		id: "arcee-ai/trinity-large-preview",
-		name: "Trinity Large Preview",
-		api: "anthropic-messages",
-		provider: "vercel-ai-gateway",
-		baseUrl: "https://ai-gateway.vercel.sh",
-		reasoning: false,
-		input: ["text"],
-		cost: {
-			input: 0.25,
-			output: 1,
-			cacheRead: 0,
-			cacheWrite: 0,
-		},
-		contextWindow: 131000,
-		maxTokens: 131000,
-	} satisfies Model<"anthropic-messages">,
 	"arcee-ai/trinity-large-thinking": {
 		id: "arcee-ai/trinity-large-thinking",
 		name: "Trinity Large Thinking",
@@ -875,12 +841,12 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		reasoning: true,
 		input: ["text"],
 		cost: {
-			input: 0.6,
-			output: 1.7,
-			cacheRead: 0,
+			input: 0.21,
+			output: 0.79,
+			cacheRead: 0.13,
 			cacheWrite: 0,
 		},
-		contextWindow: 128000,
+		contextWindow: 163840,
 		maxTokens: 128000,
 	} satisfies Model<"anthropic-messages">,
 	"deepseek/deepseek-v3.1-terminus": {
@@ -945,7 +911,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		cost: {
 			input: 0.14,
 			output: 0.28,
-			cacheRead: 0.0028,
+			cacheRead: 0.028,
 			cacheWrite: 0,
 		},
 		contextWindow: 1000000,
@@ -1206,6 +1172,23 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		contextWindow: 1000000,
 		maxTokens: 32000,
 	} satisfies Model<"anthropic-messages">,
+	"kwaipilot/kat-coder-air-v2.5": {
+		id: "kwaipilot/kat-coder-air-v2.5",
+		name: "Kat Coder Air V2.5",
+		api: "anthropic-messages",
+		provider: "vercel-ai-gateway",
+		baseUrl: "https://ai-gateway.vercel.sh",
+		reasoning: true,
+		input: ["text"],
+		cost: {
+			input: 0.15,
+			output: 0.6,
+			cacheRead: 0.03,
+			cacheWrite: 0,
+		},
+		contextWindow: 256000,
+		maxTokens: 80000,
+	} satisfies Model<"anthropic-messages">,
 	"kwaipilot/kat-coder-pro-v1": {
 		id: "kwaipilot/kat-coder-pro-v1",
 		name: "KAT-Coder-Pro V1",
@@ -1240,39 +1223,22 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		contextWindow: 256000,
 		maxTokens: 256000,
 	} satisfies Model<"anthropic-messages">,
-	"meituan/longcat-flash-chat": {
-		id: "meituan/longcat-flash-chat",
-		name: "LongCat Flash Chat",
-		api: "anthropic-messages",
-		provider: "vercel-ai-gateway",
-		baseUrl: "https://ai-gateway.vercel.sh",
-		reasoning: false,
-		input: ["text"],
-		cost: {
-			input: 0,
-			output: 0,
-			cacheRead: 0,
-			cacheWrite: 0,
-		},
-		contextWindow: 128000,
-		maxTokens: 100000,
-	} satisfies Model<"anthropic-messages">,
-	"meituan/longcat-flash-thinking-2601": {
-		id: "meituan/longcat-flash-thinking-2601",
-		name: "LongCat Flash Thinking 2601",
+	"kwaipilot/kat-coder-pro-v2.5": {
+		id: "kwaipilot/kat-coder-pro-v2.5",
+		name: "Kat Coder Pro V2.5",
 		api: "anthropic-messages",
 		provider: "vercel-ai-gateway",
 		baseUrl: "https://ai-gateway.vercel.sh",
 		reasoning: true,
 		input: ["text"],
 		cost: {
-			input: 0,
-			output: 0,
-			cacheRead: 0,
+			input: 0.74,
+			output: 2.96,
+			cacheRead: 0.15,
 			cacheWrite: 0,
 		},
-		contextWindow: 32768,
-		maxTokens: 32768,
+		contextWindow: 256000,
+		maxTokens: 80000,
 	} satisfies Model<"anthropic-messages">,
 	"meta/llama-3.1-70b": {
 		id: "meta/llama-3.1-70b",
@@ -1400,7 +1366,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		provider: "vercel-ai-gateway",
 		baseUrl: "https://ai-gateway.vercel.sh",
 		reasoning: true,
-		input: ["text"],
+		input: ["text", "image"],
 		cost: {
 			input: 1.25,
 			output: 4.25,
@@ -1580,23 +1546,6 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		contextWindow: 256000,
 		maxTokens: 256000,
 	} satisfies Model<"anthropic-messages">,
-	"mistral/devstral-small": {
-		id: "mistral/devstral-small",
-		name: "Devstral Small 1.1",
-		api: "anthropic-messages",
-		provider: "vercel-ai-gateway",
-		baseUrl: "https://ai-gateway.vercel.sh",
-		reasoning: false,
-		input: ["text"],
-		cost: {
-			input: 0.1,
-			output: 0.3,
-			cacheRead: 0,
-			cacheWrite: 0,
-		},
-		contextWindow: 128000,
-		maxTokens: 64000,
-	} satisfies Model<"anthropic-messages">,
 	"mistral/devstral-small-2": {
 		id: "mistral/devstral-small-2",
 		name: "Devstral Small 2",
@@ -1801,23 +1750,6 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		contextWindow: 128000,
 		maxTokens: 4000,
 	} satisfies Model<"anthropic-messages">,
-	"mistral/pixtral-large": {
-		id: "mistral/pixtral-large",
-		name: "Pixtral Large",
-		api: "anthropic-messages",
-		provider: "vercel-ai-gateway",
-		baseUrl: "https://ai-gateway.vercel.sh",
-		reasoning: false,
-		input: ["text", "image"],
-		cost: {
-			input: 2,
-			output: 6,
-			cacheRead: 0,
-			cacheWrite: 0,
-		},
-		contextWindow: 128000,
-		maxTokens: 4000,
-	} satisfies Model<"anthropic-messages">,
 	"moonshotai/kimi-k2": {
 		id: "moonshotai/kimi-k2",
 		name: "Kimi K2 Instruct",
@@ -2972,40 +2904,6 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		contextWindow: 256000,
 		maxTokens: 256000,
 	} satisfies Model<"anthropic-messages">,
-	"xiaomi/mimo-v2-flash": {
-		id: "xiaomi/mimo-v2-flash",
-		name: "MiMo V2 Flash",
-		api: "anthropic-messages",
-		provider: "vercel-ai-gateway",
-		baseUrl: "https://ai-gateway.vercel.sh",
-		reasoning: true,
-		input: ["text"],
-		cost: {
-			input: 0.1,
-			output: 0.3,
-			cacheRead: 0.01,
-			cacheWrite: 0,
-		},
-		contextWindow: 262144,
-		maxTokens: 32000,
-	} satisfies Model<"anthropic-messages">,
-	"xiaomi/mimo-v2-pro": {
-		id: "xiaomi/mimo-v2-pro",
-		name: "MiMo V2 Pro",
-		api: "anthropic-messages",
-		provider: "vercel-ai-gateway",
-		baseUrl: "https://ai-gateway.vercel.sh",
-		reasoning: true,
-		input: ["text"],
-		cost: {
-			input: 1,
-			output: 3,
-			cacheRead: 0.2,
-			cacheWrite: 0,
-		},
-		contextWindow: 1000000,
-		maxTokens: 128000,
-	} satisfies Model<"anthropic-messages">,
 	"xiaomi/mimo-v2.5": {
 		id: "xiaomi/mimo-v2.5",
 		name: "MiMo M2.5",
@@ -3270,9 +3168,9 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		reasoning: true,
 		input: ["text"],
 		cost: {
-			input: 3,
-			output: 10.25,
-			cacheRead: 0.5,
+			input: 2.1,
+			output: 6.6,
+			cacheRead: 0.21,
 			cacheWrite: 0,
 		},
 		contextWindow: 1000000,

packages/ai/src/utils/oauth/index.ts

@@ -1,160 +0,0 @@
-/**
- * OAuth credential management for AI providers.
- *
- * This module handles login, token refresh, and credential storage
- * for OAuth-based providers:
- * - Anthropic (Claude Pro/Max)
- * - GitHub Copilot
- */
-
-// Anthropic
-export { anthropicOAuthProvider, loginAnthropic, refreshAnthropicToken } from "./anthropic.ts";
-export * from "./device-code.ts";
-// GitHub Copilot
-export {
-	getGitHubCopilotBaseUrl,
-	githubCopilotOAuthProvider,
-	loginGitHubCopilot,
-	normalizeDomain,
-	refreshGitHubCopilotToken,
-} from "./github-copilot.ts";
-// OpenAI Codex (ChatGPT OAuth)
-export {
-	loginOpenAICodex,
-	loginOpenAICodexDeviceCode,
-	OPENAI_CODEX_BROWSER_LOGIN_METHOD,
-	OPENAI_CODEX_DEVICE_CODE_LOGIN_METHOD,
-	openaiCodexOAuthProvider,
-	refreshOpenAICodexToken,
-} from "./openai-codex.ts";
-
-export * from "./types.ts";
-
-// ============================================================================
-// Provider Registry
-// ============================================================================
-
-import { anthropicOAuthProvider } from "./anthropic.ts";
-import { githubCopilotOAuthProvider } from "./github-copilot.ts";
-import { openaiCodexOAuthProvider } from "./openai-codex.ts";
-import type { OAuthCredentials, OAuthProviderId, OAuthProviderInfo, OAuthProviderInterface } from "./types.ts";
-
-const BUILT_IN_OAUTH_PROVIDERS: OAuthProviderInterface[] = [
-	anthropicOAuthProvider,
-	githubCopilotOAuthProvider,
-	openaiCodexOAuthProvider,
-];
-
-const oauthProviderRegistry = new Map<string, OAuthProviderInterface>(
-	BUILT_IN_OAUTH_PROVIDERS.map((provider) => [provider.id, provider]),
-);
-
-/**
- * Get an OAuth provider by ID
- */
-export function getOAuthProvider(id: OAuthProviderId): OAuthProviderInterface | undefined {
-	return oauthProviderRegistry.get(id);
-}
-
-/**
- * Register a custom OAuth provider
- */
-export function registerOAuthProvider(provider: OAuthProviderInterface): void {
-	oauthProviderRegistry.set(provider.id, provider);
-}
-
-/**
- * Unregister an OAuth provider.
- *
- * If the provider is built-in, restores the built-in implementation.
- * Custom providers are removed completely.
- */
-export function unregisterOAuthProvider(id: string): void {
-	const builtInProvider = BUILT_IN_OAUTH_PROVIDERS.find((provider) => provider.id === id);
-	if (builtInProvider) {
-		oauthProviderRegistry.set(id, builtInProvider);
-		return;
-	}
-	oauthProviderRegistry.delete(id);
-}
-
-/**
- * Reset OAuth providers to built-ins.
- */
-export function resetOAuthProviders(): void {
-	oauthProviderRegistry.clear();
-	for (const provider of BUILT_IN_OAUTH_PROVIDERS) {
-		oauthProviderRegistry.set(provider.id, provider);
-	}
-}
-
-/**
- * Get all registered OAuth providers
- */
-export function getOAuthProviders(): OAuthProviderInterface[] {
-	return Array.from(oauthProviderRegistry.values());
-}
-
-/**
- * @deprecated Use getOAuthProviders() which returns OAuthProviderInterface[]
- */
-export function getOAuthProviderInfoList(): OAuthProviderInfo[] {
-	return getOAuthProviders().map((p) => ({
-		id: p.id,
-		name: p.name,
-		available: true,
-	}));
-}
-
-// ============================================================================
-// High-level API (uses provider registry)
-// ============================================================================
-
-/**
- * Refresh token for any OAuth provider.
- * @deprecated Use getOAuthProvider(id).refreshToken() instead
- */
-export async function refreshOAuthToken(
-	providerId: OAuthProviderId,
-	credentials: OAuthCredentials,
-): Promise<OAuthCredentials> {
-	const provider = getOAuthProvider(providerId);
-	if (!provider) {
-		throw new Error(`Unknown OAuth provider: ${providerId}`);
-	}
-	return provider.refreshToken(credentials);
-}
-
-/**
- * Get API key for a provider from OAuth credentials.
- * Automatically refreshes expired tokens.
- *
- * @returns API key string and updated credentials, or null if no credentials
- * @throws Error if refresh fails
- */
-export async function getOAuthApiKey(
-	providerId: OAuthProviderId,
-	credentials: Record<string, OAuthCredentials>,
-): Promise<{ newCredentials: OAuthCredentials; apiKey: string } | null> {
-	const provider = getOAuthProvider(providerId);
-	if (!provider) {
-		throw new Error(`Unknown OAuth provider: ${providerId}`);
-	}
-
-	let creds = credentials[providerId];
-	if (!creds) {
-		return null;
-	}
-
-	// Refresh if expired
-	if (Date.now() >= creds.expires) {
-		try {
-			creds = await provider.refreshToken(creds);
-		} catch (_error) {
-			throw new Error(`Failed to refresh OAuth token for ${providerId}`);
-		}
-	}
-
-	const apiKey = provider.getApiKey(creds);
-	return { newCredentials: creds, apiKey };
-}

packages/ai/src/utils/oauth/types.ts

@@ -1,79 +0,0 @@
-import type { Api, Model } from "../../types.ts";
-
-export type OAuthCredentials = {
-	refresh: string;
-	access: string;
-	expires: number;
-	[key: string]: unknown;
-};
-
-export type OAuthProviderId = string;
-
-/** @deprecated Use OAuthProviderId instead */
-export type OAuthProvider = OAuthProviderId;
-
-export type OAuthPrompt = {
-	message: string;
-	placeholder?: string;
-	allowEmpty?: boolean;
-};
-
-export type OAuthAuthInfo = {
-	url: string;
-	instructions?: string;
-};
-
-export type OAuthDeviceCodeInfo = {
-	userCode: string;
-	verificationUri: string;
-	intervalSeconds?: number;
-	expiresInSeconds?: number;
-};
-
-export type OAuthSelectOption = {
-	id: string;
-	label: string;
-};
-
-export type OAuthSelectPrompt = {
-	message: string;
-	options: OAuthSelectOption[];
-};
-
-export interface OAuthLoginCallbacks {
-	onAuth: (info: OAuthAuthInfo) => void;
-	onDeviceCode: (info: OAuthDeviceCodeInfo) => void;
-	onPrompt: (prompt: OAuthPrompt) => Promise<string>;
-	onProgress?: (message: string) => void;
-	onManualCodeInput?: () => Promise<string>;
-	/** Show an interactive selector and return the selected option id, or undefined on cancel. */
-	onSelect: (prompt: OAuthSelectPrompt) => Promise<string | undefined>;
-	signal?: AbortSignal;
-}
-
-export interface OAuthProviderInterface {
-	readonly id: OAuthProviderId;
-	readonly name: string;
-
-	/** Run the login flow, return credentials to persist */
-	login(callbacks: OAuthLoginCallbacks): Promise<OAuthCredentials>;
-
-	/** Whether login uses a local callback server and supports manual code input. */
-	usesCallbackServer?: boolean;
-
-	/** Refresh expired credentials, return updated credentials to persist */
-	refreshToken(credentials: OAuthCredentials): Promise<OAuthCredentials>;
-
-	/** Convert credentials to API key string for the provider */
-	getApiKey(credentials: OAuthCredentials): string;
-
-	/** Optional: modify models for this provider (e.g., update baseUrl) */
-	modifyModels?(models: Model<Api>[], credentials: OAuthCredentials): Model<Api>[];
-}
-
-/** @deprecated Use OAuthProviderInterface instead */
-export interface OAuthProviderInfo {
-	id: OAuthProviderId;
-	name: string;
-	available: boolean;
-}

packages/ai/test/anthropic-oauth.test.ts

@@ -1,6 +1,6 @@
 import { afterEach, describe, expect, it, vi } from "vitest";
+import { anthropicOAuth } from "../src/auth/oauth/anthropic.ts";
 import type { AuthEvent, AuthPrompt } from "../src/auth/types.ts";
-import { anthropicOAuth, loginAnthropic, refreshAnthropicToken } from "../src/utils/oauth/anthropic.ts";
 
 function jsonResponse(body: unknown, status: number = 200): Response {
 	return new Response(JSON.stringify(body), {
@@ -53,18 +53,16 @@ describe.sequential("Anthropic OAuth", () => {
 		});
 		vi.stubGlobal("fetch", fetchMock);
 
-		const credentials = await loginAnthropic({
-			onAuth: (info) => {
-				authUrl = info.url;
+		const credentials = await anthropicOAuth.login({
+			notify: (event) => {
+				if (event.type === "auth_url") authUrl = event.url;
 			},
-			onPrompt: async () => "",
-			onManualCodeInput: async () => {
+			prompt: async (prompt) => {
+				if (prompt.type !== "manual_code") throw new Error(`Unexpected prompt: ${prompt.type}`);
 				const url = new URL(authUrl);
 				const state = url.searchParams.get("state");
 				const redirectUri = url.searchParams.get("redirect_uri");
-				if (!state || !redirectUri) {
-					throw new Error("Missing OAuth state or redirect_uri in auth URL");
-				}
+				if (!state || !redirectUri) throw new Error("Missing OAuth state or redirect_uri in auth URL");
 				return `${redirectUri}?code=manual-code&state=${state}`;
 			},
 		});
@@ -91,7 +89,12 @@ describe.sequential("Anthropic OAuth", () => {
 		});
 		vi.stubGlobal("fetch", fetchMock);
 
-		const credentials = await refreshAnthropicToken("refresh-token");
+		const credentials = await anthropicOAuth.refresh({
+			type: "oauth",
+			access: "old-access-token",
+			refresh: "refresh-token",
+			expires: 0,
+		});
 
 		expect(credentials.access).toBe("new-access-token");
 		expect(credentials.refresh).toBe("new-refresh-token");

packages/ai/test/cloudflare-stream.test.ts

@@ -0,0 +1,65 @@
+import { describe, expect, it } from "vitest";
+import { cloudflareStreams } from "../src/providers/cloudflare-stream.ts";
+import type { Api, Context, Model } from "../src/types.ts";
+import { AssistantMessageEventStream } from "../src/utils/event-stream.ts";
+
+const model: Model<Api> = {
+	id: "model",
+	name: "model",
+	api: "openai-completions",
+	provider: "cloudflare-ai-gateway",
+	baseUrl: "https://gateway.ai.cloudflare.com/v1/{CLOUDFLARE_ACCOUNT_ID}/{CLOUDFLARE_GATEWAY_ID}/openai",
+	reasoning: false,
+	input: ["text"],
+	cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+	contextWindow: 1000,
+	maxTokens: 100,
+};
+
+const context: Context = { messages: [] };
+
+describe("Cloudflare provider streams", () => {
+	it("materializes the model endpoint before dispatch", () => {
+		const captured: string[] = [];
+		const streams = cloudflareStreams({
+			stream: (requestModel) => {
+				captured.push(requestModel.baseUrl);
+				return new AssistantMessageEventStream();
+			},
+			streamSimple: (requestModel) => {
+				captured.push(requestModel.baseUrl);
+				return new AssistantMessageEventStream();
+			},
+		});
+		const env = {
+			CLOUDFLARE_ACCOUNT_ID: "account",
+			CLOUDFLARE_GATEWAY_ID: "gateway",
+		};
+
+		streams.stream(model, context, { env });
+		streams.streamSimple(model, context, { env });
+
+		expect(captured).toEqual([
+			"https://gateway.ai.cloudflare.com/v1/account/gateway/openai",
+			"https://gateway.ai.cloudflare.com/v1/account/gateway/openai",
+		]);
+	});
+
+	it("keeps placeholders when the provider env does not resolve them", () => {
+		let captured: string | undefined;
+		const streams = cloudflareStreams({
+			stream: (requestModel) => {
+				captured = requestModel.baseUrl;
+				return new AssistantMessageEventStream();
+			},
+			streamSimple: (requestModel) => {
+				captured = requestModel.baseUrl;
+				return new AssistantMessageEventStream();
+			},
+		});
+
+		streams.streamSimple(model, context, {});
+
+		expect(captured).toBe(model.baseUrl);
+	});
+});

packages/ai/test/codex-websocket-cached-probe.ts

@@ -9,7 +9,7 @@
 import { tmpdir } from "node:os";
 import { join, resolve } from "node:path";
 import { Type } from "typebox";
-import { AuthStorage } from "../../coding-agent/src/core/auth-storage.ts";
+import { ModelRuntime } from "../../coding-agent/src/core/model-runtime.ts";
 import {
 	closeOpenAICodexWebSocketSessions,
 	getOpenAICodexWebSocketDebugStats,
@@ -166,8 +166,9 @@ async function main(): Promise<void> {
 	const model = getModel("openai-codex", "gpt-5.5") as Model<"openai-codex-responses"> | undefined;
 	if (!model) throw new Error("Model openai-codex/gpt-5.5 not found");
 	const modelWithMaxTokens = { ...model, maxTokens: args.maxTokens };
-	const authStorage = AuthStorage.create();
-	const apiKey = (await authStorage.getApiKey("openai-codex")) ?? (await authStorage.getApiKey("openai"));
+	const modelRuntime = await ModelRuntime.create();
+	const apiKey =
+		(await modelRuntime.getAuth("openai-codex"))?.auth.apiKey ?? (await modelRuntime.getAuth("openai"))?.auth.apiKey;
 	if (!apiKey) {
 		throw new Error("No OpenAI Codex API key found in coding-agent auth storage.");
 	}

packages/ai/test/github-copilot-oauth.test.ts

@@ -1,10 +1,8 @@
 import { afterEach, describe, expect, it, vi } from "vitest";
-import { getModels } from "../src/compat.ts";
-import {
-	githubCopilotOAuthProvider,
-	loginGitHubCopilot,
-	refreshGitHubCopilotToken,
-} from "../src/utils/oauth/github-copilot.ts";
+import { InMemoryCredentialStore } from "../src/auth/credential-store.ts";
+import { githubCopilotOAuth } from "../src/auth/oauth/github-copilot.ts";
+import { createModels } from "../src/models.ts";
+import { githubCopilotProvider } from "../src/providers/github-copilot.ts";
 
 function jsonResponse(body: unknown, status: number = 200): Response {
 	return new Response(JSON.stringify(body), {
@@ -28,6 +26,33 @@ function getUrl(input: unknown): string {
 	throw new Error(`Unsupported fetch input: ${String(input)}`);
 }
 
+function loginGitHubCopilotForTest(options: {
+	onDeviceCode(info: {
+		userCode: string;
+		verificationUri: string;
+		intervalSeconds?: number;
+		expiresInSeconds?: number;
+	}): void;
+	onPrompt(prompt: { message: string; placeholder?: string; allowEmpty?: boolean }): Promise<string>;
+	onProgress?(message: string): void;
+	signal?: AbortSignal;
+}) {
+	return githubCopilotOAuth.login({
+		signal: options.signal,
+		prompt: (prompt) => {
+			if (prompt.type !== "text") throw new Error(`Unexpected prompt: ${prompt.type}`);
+			return options.onPrompt({ message: prompt.message, placeholder: prompt.placeholder, allowEmpty: true });
+		},
+		notify: (event) => {
+			if (event.type === "device_code") {
+				const { type: _, ...info } = event;
+				options.onDeviceCode(info);
+			}
+			if (event.type === "progress") options.onProgress?.(event.message);
+		},
+	});
+}
+
 describe("GitHub Copilot OAuth device flow", () => {
 	afterEach(() => {
 		vi.unstubAllGlobals();
@@ -76,13 +101,19 @@ describe("GitHub Copilot OAuth device flow", () => {
 
 		vi.stubGlobal("fetch", fetchMock);
 
-		const credentials = await refreshGitHubCopilotToken("ghu_refresh_token");
+		const credentials = await githubCopilotOAuth.refresh({
+			type: "oauth",
+			access: "old-access-token",
+			refresh: "ghu_refresh_token",
+			expires: 0,
+		});
 		expect(credentials.availableModelIds).toEqual(["gpt-4.1"]);
 
-		const modifiedModels = githubCopilotOAuthProvider.modifyModels?.(getModels("github-copilot"), credentials) ?? [];
-		expect(modifiedModels.filter((model) => model.provider === "github-copilot").map((model) => model.id)).toEqual([
-			"gpt-4.1",
-		]);
+		const store = new InMemoryCredentialStore();
+		await store.modify("github-copilot", async () => ({ ...credentials, type: "oauth" }));
+		const models = createModels({ credentials: store });
+		models.setProvider(githubCopilotProvider());
+		expect((await models.getAvailable("github-copilot")).map((model) => model.id)).toEqual(["gpt-4.1"]);
 	});
 
 	it("reports device-code details through onDeviceCode", async () => {
@@ -127,7 +158,7 @@ describe("GitHub Copilot OAuth device flow", () => {
 		vi.stubGlobal("fetch", fetchMock);
 
 		const onDeviceCode = vi.fn();
-		const loginPromise = loginGitHubCopilot({
+		const loginPromise = loginGitHubCopilotForTest({
 			onDeviceCode,
 			onPrompt: async () => "",
 		});
@@ -166,7 +197,7 @@ describe("GitHub Copilot OAuth device flow", () => {
 
 		const onDeviceCode = vi.fn();
 		await expect(
-			loginGitHubCopilot({
+			loginGitHubCopilotForTest({
 				onDeviceCode,
 				onPrompt: async () => "",
 			}),
@@ -220,7 +251,7 @@ describe("GitHub Copilot OAuth device flow", () => {
 		vi.stubGlobal("fetch", fetchMock);
 
 		const onDeviceCode = vi.fn();
-		const loginPromise = loginGitHubCopilot({
+		const loginPromise = loginGitHubCopilotForTest({
 			onDeviceCode,
 			onPrompt: async () => "",
 		});
@@ -308,7 +339,7 @@ describe("GitHub Copilot OAuth device flow", () => {
 
 		vi.stubGlobal("fetch", fetchMock);
 
-		const loginPromise = loginGitHubCopilot({
+		const loginPromise = loginGitHubCopilotForTest({
 			onDeviceCode: () => {},
 			onPrompt: async () => "",
 			onProgress: () => {},
@@ -382,7 +413,7 @@ describe("GitHub Copilot OAuth device flow", () => {
 
 		vi.stubGlobal("fetch", fetchMock);
 
-		const loginPromise = loginGitHubCopilot({
+		const loginPromise = loginGitHubCopilotForTest({
 			onDeviceCode: () => {},
 			onPrompt: async () => "",
 		});

packages/ai/test/images-models.test.ts

@@ -51,10 +51,10 @@ function testProvider(input: {
 		auth: {
 			apiKey: {
 				name: "Test key",
-				resolve: async ({ ctx }) => {
+				resolve: async ({ ctx, credential }) => {
 					if (!input.envVar) return { auth: {} };
-					const key = await ctx.env(input.envVar);
-					return key ? { auth: { apiKey: key }, source: input.envVar } : undefined;
+					const key = credential?.key ?? (await ctx.env(input.envVar));
+					return key ? { auth: { apiKey: key }, source: credential ? "stored" : input.envVar } : undefined;
 				},
 			},
 		},
@@ -93,6 +93,8 @@ describe("ImagesModels", () => {
 		const model = models.getModel("p1", "model-a")!;
 
 		expect((await models.getAuth(model))?.auth.apiKey).toBe("env-key");
+		expect((await models.getAuth(model.provider))?.auth.apiKey).toBe("env-key");
+		expect((await models.getAuth(model, { apiKey: "explicit-key" }))?.auth.apiKey).toBe("explicit-key");
 
 		const result = await models.generateImages(model, context);
 		expect(result.stopReason).toBe("stop");

packages/ai/test/models-runtime.test.ts

@@ -106,6 +106,22 @@ function testOAuth(overrides?: Partial<OAuthAuth>): OAuthAuth {
 }
 
 describe("Models runtime", () => {
+	it("enumerates credential metadata without exposing secrets", async () => {
+		const credentials = new InMemoryCredentialStore();
+		await credentials.modify("api-provider", async () => ({ type: "api_key", key: "secret" }));
+		await credentials.modify("oauth-provider", async () => ({
+			type: "oauth",
+			access: "access",
+			refresh: "refresh",
+			expires: Date.now() + 60_000,
+		}));
+
+		expect(await credentials.list()).toEqual([
+			{ providerId: "api-provider", type: "api_key" },
+			{ providerId: "oauth-provider", type: "oauth" },
+		]);
+	});
+
 	it("applies request-wide pricing tiers above the configured input threshold", () => {
 		const model = testModel("openai", "gpt-5.6-sol");
 		model.cost = {
@@ -246,8 +262,10 @@ describe("Models runtime", () => {
 		models.setProvider(testProvider({ id: "p1", auth: { apiKey: envKeyAuth("env-key"), oauth: testOAuth() } }));
 		const model = testModel("p1", "model-a");
 
-		// nothing stored: ambient env resolves
+		// model and provider-id overloads resolve the same provider-scoped auth
 		expect((await models.getAuth(model))?.auth.apiKey).toBe("env-key");
+		expect((await models.getAuth(model.provider))?.auth.apiKey).toBe("env-key");
+		expect((await models.getAuth(model, { apiKey: "explicit-key" }))?.auth.apiKey).toBe("explicit-key");
 
 		// stored oauth credential (persisted via the single write path): beats ambient env
 		await credentials.modify("p1", async () => ({
@@ -256,17 +274,69 @@ describe("Models runtime", () => {
 			refresh: "r",
 			expires: Date.now() + 100000,
 		}));
-		const resolution = await models.getAuth(model);
+		const resolution = await models.getAuth(model.provider);
 		expect(resolution?.auth.apiKey).toBe("oauth-token");
 		expect(resolution?.source).toBe("OAuth");
 
 		// stored api-key credential resolves through apiKey auth, beats env
 		await credentials.modify("p1", async () => ({ type: "api_key", key: "stored-key" }));
-		const apiKeyResolution = await models.getAuth(model);
+		const apiKeyResolution = await models.getAuth(model.provider);
 		expect(apiKeyResolution?.auth.apiKey).toBe("stored-key");
 		expect(apiKeyResolution?.source).toBe("stored");
 	});
 
+	it("checks provider auth without refreshing OAuth and filters available models", async () => {
+		const credentials = new InMemoryCredentialStore();
+		let refreshes = 0;
+		const models = createModels({ credentials });
+		models.setProvider(testProvider({ id: "ambient", auth: { apiKey: envKeyAuth("env-key") } }));
+		models.setProvider(testProvider({ id: "missing", auth: { apiKey: envKeyAuth(undefined) } }));
+		models.setProvider(
+			testProvider({
+				id: "oauth",
+				auth: {
+					oauth: testOAuth({
+						refresh: async (credential) => {
+							refreshes++;
+							return credential;
+						},
+					}),
+				},
+			}),
+		);
+		await credentials.modify("oauth", async () => ({
+			type: "oauth",
+			access: "expired",
+			refresh: "refresh",
+			expires: 0,
+		}));
+
+		expect(await models.checkAuth("ambient")).toEqual({ source: "env", type: "api_key" });
+		expect(await models.checkAuth("missing")).toBeUndefined();
+		expect(await models.checkAuth("oauth")).toEqual({ source: "OAuth", type: "oauth" });
+		expect(refreshes).toBe(0);
+		expect((await models.getAvailable()).map((model) => model.provider)).toEqual(["ambient", "oauth"]);
+		expect((await models.getAvailable("ambient")).map((model) => model.provider)).toEqual(["ambient"]);
+	});
+
+	it("runs provider login and logout through the credential store", async () => {
+		const credentials = new InMemoryCredentialStore();
+		const apiKey = envKeyAuth(undefined);
+		apiKey.login = async () => ({ type: "api_key", key: "logged-in" });
+		const models = createModels({ credentials });
+		models.setProvider(testProvider({ id: "p1", auth: { apiKey } }));
+
+		const credential = await models.login("p1", "api_key", {
+			prompt: async () => "unused",
+			notify: () => {},
+		});
+		expect(credential).toEqual({ type: "api_key", key: "logged-in" });
+		expect(await credentials.read("p1")).toEqual(credential);
+
+		await models.logout("p1");
+		expect(await credentials.read("p1")).toBeUndefined();
+	});
+
 	it("a stored credential without a matching handler blocks ambient fallback", async () => {
 		const credentials = new InMemoryCredentialStore();
 		const models = createModels({ credentials });
@@ -274,7 +344,7 @@ describe("Models runtime", () => {
 		models.setProvider(testProvider({ id: "p1", auth: { apiKey: envKeyAuth("env-key") } }));
 		await credentials.modify("p1", async () => ({ type: "oauth", access: "a", refresh: "r", expires: 0 }));
 
-		expect(await models.getAuth(testModel("p1", "model-a"))).toBeUndefined();
+		expect(await models.getAuth("p1")).toBeUndefined();
 	});
 
 	it("refreshes expired oauth credentials and persists the rotated credential", async () => {
@@ -291,7 +361,7 @@ describe("Models runtime", () => {
 			expires: 0,
 		}));
 
-		const resolution = await models.getAuth(testModel("p1", "model-a"));
+		const resolution = await models.getAuth("p1");
 		expect(resolution?.auth.apiKey).toBe("new-token");
 		expect(((await credentials.read("p1")) as { access: string }).access).toBe("new-token");
 	});
@@ -307,7 +377,7 @@ describe("Models runtime", () => {
 		models.setProvider(testProvider({ id: "p1", auth: { oauth } }));
 		await credentials.modify("p1", async () => ({ type: "oauth", access: "old", refresh: "r", expires: 0 }));
 
-		await expect(models.getAuth(testModel("p1", "model-a"))).rejects.toMatchObject({ code: "oauth" });
+		await expect(models.getAuth("p1")).rejects.toMatchObject({ code: "oauth" });
 		// credential preserved for retry / re-login
 		expect(((await credentials.read("p1")) as { access: string }).access).toBe("old");
 	});
@@ -328,7 +398,7 @@ describe("Models runtime", () => {
 		models.setProvider(testProvider({ id: "p1", auth: { oauth } }));
 		const model = testModel("p1", "model-a");
 
-		const [a, b] = await Promise.all([models.getAuth(model), models.getAuth(model)]);
+		const [a, b] = await Promise.all([models.getAuth(model.provider), models.getAuth(model.provider)]);
 		expect(refreshes).toBe(1);
 		expect(a?.auth.apiKey).toBe("new-1");
 		expect(b?.auth.apiKey).toBe("new-1");
@@ -339,6 +409,7 @@ describe("Models runtime", () => {
 		const base = new InMemoryCredentialStore();
 		const credentials: CredentialStore = {
 			read: (pid) => base.read(pid),
+			list: () => base.list(),
 			modify: (pid, fn) => {
 				modifies++;
 				return base.modify(pid, fn);
@@ -354,7 +425,7 @@ describe("Models runtime", () => {
 		const models = createModels({ credentials });
 		models.setProvider(testProvider({ id: "p1", auth: { oauth: testOAuth() } }));
 
-		expect((await models.getAuth(testModel("p1", "model-a")))?.auth.apiKey).toBe("valid");
+		expect((await models.getAuth("p1"))?.auth.apiKey).toBe("valid");
 		expect(modifies).toBe(0);
 	});
 
@@ -364,16 +435,18 @@ describe("Models runtime", () => {
 			read: async () => {
 				throw new Error("disk on fire");
 			},
+			list: async () => [],
 			modify: async () => undefined,
 			delete: async () => {},
 		};
 		const models = createModels({ credentials: readFailing });
 		models.setProvider(testProvider({ id: "p1", auth: { apiKey: envKeyAuth("env-key") } }));
-		await expect(models.getAuth(testModel("p1", "model-a"))).rejects.toMatchObject({ code: "auth" });
+		await expect(models.getAuth("p1")).rejects.toMatchObject({ code: "auth" });
 
 		// modify failure during refresh
 		const modifyFailing: CredentialStore = {
 			read: async () => ({ type: "oauth", access: "old", refresh: "r", expires: 0 }),
+			list: async () => [{ providerId: "p1", type: "oauth" }],
 			modify: async () => {
 				throw new Error("disk on fire");
 			},
@@ -381,7 +454,7 @@ describe("Models runtime", () => {
 		};
 		const oauthModels = createModels({ credentials: modifyFailing });
 		oauthModels.setProvider(testProvider({ id: "p1", auth: { oauth: testOAuth() } }));
-		await expect(oauthModels.getAuth(testModel("p1", "model-a"))).rejects.toMatchObject({ code: "auth" });
+		await expect(oauthModels.getAuth("p1")).rejects.toMatchObject({ code: "auth" });
 	});
 
 	it("wraps api-key auth failures in ModelsError", async () => {
@@ -393,7 +466,7 @@ describe("Models runtime", () => {
 		};
 		const models = createModels();
 		models.setProvider(testProvider({ id: "p1", auth: { apiKey: failing } }));
-		await expect(models.getAuth(testModel("p1", "model-a"))).rejects.toMatchObject({ code: "auth" });
+		await expect(models.getAuth("p1")).rejects.toMatchObject({ code: "auth" });
 	});
 
 	it("uses explicit request api key and env during provider auth resolution", async () => {
@@ -427,7 +500,7 @@ describe("Models runtime", () => {
 			resolve: async () => ({
 				auth: {
 					apiKey: "resolved-key",
-					headers: { "x-a": "auth", "x-b": "auth" },
+					headers: { Authorization: "Bearer resolved-key", "x-a": "auth", "x-b": "auth" },
 					baseUrl: "https://auth.test/v1",
 				},
 			}),
@@ -438,12 +511,12 @@ describe("Models runtime", () => {
 
 		const result = await models.completeSimple(model, context, {
 			apiKey: "explicit-key",
-			headers: { "x-b": "explicit" },
+			headers: { authorization: "Explicit token", "x-b": "explicit" },
 		});
 		expect(result.stopReason).toBe("stop");
 		expect(calls).toHaveLength(1);
 		expect(calls[0].options?.apiKey).toBe("explicit-key");
-		expect(calls[0].options?.headers).toEqual({ "x-a": "auth", "x-b": "explicit" });
+		expect(calls[0].options?.headers).toEqual({ authorization: "Explicit token", "x-a": "auth", "x-b": "explicit" });
 		expect(calls[0].model.baseUrl).toBe("https://auth.test/v1");
 
 		// without explicit options, resolved auth applies
@@ -452,6 +525,36 @@ describe("Models runtime", () => {
 		expect(calls[1].options?.apiKey).toBe("resolved-key");
 	});
 
+	it("adds model headers only for model auth and transforms assembled headers once", async () => {
+		const calls: ProviderCall[] = [];
+		const models = createModels();
+		models.setProvider(testProvider({ id: "p1", auth: { apiKey: envKeyAuth("key") }, calls }));
+		const model = testModel("p1", "model-a");
+		model.headers = { "x-model": "model", "x-shared": "model" };
+
+		expect((await models.getAuth("p1"))?.auth.headers).toBeUndefined();
+		expect((await models.getAuth(model))?.auth.headers).toEqual({ "x-model": "model", "x-shared": "model" });
+
+		let transforms = 0;
+		await models.completeSimple(model, context, {
+			headers: { "x-explicit": "explicit", "X-Shared": "explicit" },
+			transformHeaders: async (headers) => {
+				transforms++;
+				expect(headers).toEqual({ "x-model": "model", "x-explicit": "explicit", "X-Shared": "explicit" });
+				return { ...headers, "x-transformed": "yes" };
+			},
+		});
+
+		expect(transforms).toBe(1);
+		expect(calls[0].options?.headers).toEqual({
+			"x-model": "model",
+			"x-explicit": "explicit",
+			"X-Shared": "explicit",
+			"x-transformed": "yes",
+		});
+		expect(calls[0].options).not.toHaveProperty("transformHeaders");
+	});
+
 	it("produces an error stream for unknown providers instead of throwing", async () => {
 		const models = createModels();
 		const result = await models.completeSimple(testModel("ghost", "model-a"), context);

packages/ai/test/oauth-auth.test.ts

@@ -1,17 +1,23 @@
 import { afterEach, describe, expect, it, vi } from "vitest";
 import { InMemoryCredentialStore } from "../src/auth/credential-store.ts";
+import { anthropicOAuth } from "../src/auth/oauth/anthropic.ts";
+import { githubCopilotOAuth } from "../src/auth/oauth/github-copilot.ts";
+import { openaiCodexOAuth } from "../src/auth/oauth/openai-codex.ts";
 import { createModels } from "../src/models.ts";
+import * as extensionOAuthCompatibility from "../src/oauth.ts";
 import { anthropicProvider } from "../src/providers/anthropic.ts";
 import { githubCopilotProvider } from "../src/providers/github-copilot.ts";
-import { anthropicOAuth } from "../src/utils/oauth/anthropic.ts";
-import { githubCopilotOAuth } from "../src/utils/oauth/github-copilot.ts";
-import { openaiCodexOAuth } from "../src/utils/oauth/openai-codex.ts";
 
 function jsonResponse(body: unknown, status = 200): Response {
 	return new Response(JSON.stringify(body), { status, headers: { "Content-Type": "application/json" } });
 }
 
 describe.sequential("OAuthAuth adapters", () => {
+	it("keeps the extension OAuth barrel free of built-in flow implementations", () => {
+		expect(extensionOAuthCompatibility).not.toHaveProperty("loginAnthropic");
+		expect(extensionOAuthCompatibility).not.toHaveProperty("anthropicOAuth");
+	});
+
 	afterEach(() => {
 		vi.unstubAllGlobals();
 	});
@@ -104,7 +110,7 @@ describe("OAuth through Models.getAuth (lazy load chain)", () => {
 		models.setProvider(anthropicProvider());
 
 		const model = models.getModels("anthropic")[0];
-		const result = await models.getAuth(model);
+		const result = await models.getAuth(model.provider);
 		expect(result?.auth.apiKey).toBe("oauth-access-token");
 		expect(result?.source).toBe("OAuth");
 	});
@@ -122,7 +128,7 @@ describe("OAuth through Models.getAuth (lazy load chain)", () => {
 		models.setProvider(githubCopilotProvider());
 
 		const model = models.getModels("github-copilot")[0];
-		const result = await models.getAuth(model);
+		const result = await models.getAuth(model.provider);
 		expect(result?.auth.apiKey).toBe(access);
 		expect(result?.auth.baseUrl).toBe("https://api.business.githubcopilot.com");
 	});

packages/ai/test/oauth-device-code.test.ts

@@ -1,5 +1,5 @@
 import { afterEach, describe, expect, it, vi } from "vitest";
-import { pollOAuthDeviceCodeFlow } from "../src/utils/oauth/device-code.ts";
+import { pollOAuthDeviceCodeFlow } from "../src/auth/oauth/device-code.ts";
 
 describe("OAuth device-code polling", () => {
 	afterEach(() => {

packages/ai/test/oauth.ts

@@ -8,8 +8,8 @@
 import { chmodSync, existsSync, mkdirSync, readFileSync, writeFileSync } from "fs";
 import { homedir } from "os";
 import { dirname, join } from "path";
-import { getOAuthApiKey } from "../src/utils/oauth/index.ts";
-import type { OAuthCredentials, OAuthProvider } from "../src/utils/oauth/types.ts";
+import type { OAuthCredentials } from "../src/auth/types.ts";
+import { builtinProviders } from "../src/providers/all.ts";
 
 const AUTH_PATH = join(homedir(), ".pi", "agent", "auth.json");
 
@@ -65,28 +65,18 @@ export async function resolveApiKey(provider: string): Promise<string | undefine
 	}
 
 	if (entry.type === "oauth") {
-		// Build OAuthCredentials record for getOAuthApiKey
-		const oauthCredentials: Record<string, OAuthCredentials> = {};
-		for (const [key, value] of Object.entries(storage)) {
-			if (value.type === "oauth") {
-				const { type: _, ...creds } = value;
-				oauthCredentials[key] = creds;
-			}
-		}
-
-		let result: { newCredentials: OAuthCredentials; apiKey: string } | null = null;
+		const oauth = builtinProviders().find((candidate) => candidate.id === provider)?.auth.oauth;
+		if (!oauth) return undefined;
+		let credential = entry;
 		try {
-			result = await getOAuthApiKey(provider as OAuthProvider, oauthCredentials);
-		} catch (e) {
-			console.log(JSON.stringify(e));
+			if (Date.now() >= credential.expires) credential = await oauth.refresh(credential);
+		} catch (error) {
+			console.log(JSON.stringify(error));
+			return undefined;
 		}
-		if (!result) return undefined;
-
-		// Save refreshed credentials back to auth.json
-		storage[provider] = { type: "oauth", ...result.newCredentials };
+		storage[provider] = credential;
 		saveAuthStorage(storage);
-
-		return result.apiKey;
+		return (await oauth.toAuth(credential)).apiKey;
 	}
 
 	return undefined;

packages/ai/test/openai-codex-oauth.test.ts

@@ -1,9 +1,5 @@
 import { afterEach, describe, expect, it, vi } from "vitest";
-import {
-	loginOpenAICodexDeviceCode,
-	openaiCodexOAuthProvider,
-	refreshOpenAICodexToken,
-} from "../src/utils/oauth/openai-codex.ts";
+import { openaiCodexOAuth } from "../src/auth/oauth/openai-codex.ts";
 
 function jsonResponse(body: unknown, status: number = 200): Response {
 	return new Response(JSON.stringify(body), {
@@ -45,6 +41,30 @@ function deviceAuthPendingResponse(): Response {
 	);
 }
 
+function loginOpenAICodexDeviceCodeForTest(options: {
+	onDeviceCode(info: {
+		userCode: string;
+		verificationUri: string;
+		intervalSeconds?: number;
+		expiresInSeconds?: number;
+	}): void;
+	signal?: AbortSignal;
+}) {
+	return openaiCodexOAuth.login({
+		signal: options.signal,
+		prompt: async (prompt) => {
+			if (prompt.type !== "select") throw new Error(`Unexpected prompt: ${prompt.type}`);
+			return "device_code";
+		},
+		notify: (event) => {
+			if (event.type === "device_code") {
+				const { type: _, ...info } = event;
+				options.onDeviceCode(info);
+			}
+		},
+	});
+}
+
 describe("OpenAI Codex OAuth", () => {
 	afterEach(() => {
 		vi.restoreAllMocks();
@@ -125,7 +145,7 @@ describe("OpenAI Codex OAuth", () => {
 
 		vi.stubGlobal("fetch", fetchMock);
 
-		const credentialsPromise = loginOpenAICodexDeviceCode({
+		const credentialsPromise = loginOpenAICodexDeviceCodeForTest({
 			onDeviceCode: (info) => deviceInfos.push(info),
 		});
 
@@ -159,7 +179,7 @@ describe("OpenAI Codex OAuth", () => {
 		const accessToken = createAccessToken("account-456");
 		const selectPrompts: Array<{
 			message: string;
-			options: Array<{ id: string; label: string }>;
+			options: readonly { id: string; label: string }[];
 		}> = [];
 		const deviceInfos: Array<{
 			userCode: string;
@@ -199,20 +219,22 @@ describe("OpenAI Codex OAuth", () => {
 		);
 
 		await expect(
-			openaiCodexOAuthProvider.login({
-				onAuth: () => {
-					throw new Error("Browser login should not start");
-				},
-				onDeviceCode: (info) => deviceInfos.push(info),
-				onPrompt: async () => {
-					throw new Error("Prompt should not be used");
-				},
-				onSelect: async (prompt) => {
+			openaiCodexOAuth.login({
+				prompt: async (prompt) => {
+					if (prompt.type !== "select") throw new Error("Text prompt should not be used");
 					selectPrompts.push(prompt);
 					return "device_code";
 				},
+				notify: (event) => {
+					if (event.type === "auth_url") throw new Error("Browser login should not start");
+					if (event.type === "device_code") {
+						const { type: _, ...info } = event;
+						deviceInfos.push(info);
+					}
+				},
 			}),
 		).resolves.toMatchObject({
+			type: "oauth",
 			access: accessToken,
 			refresh: "refresh-token",
 			accountId: "account-456",
@@ -220,6 +242,7 @@ describe("OpenAI Codex OAuth", () => {
 
 		expect(selectPrompts).toEqual([
 			{
+				type: "select",
 				message: "Select OpenAI Codex login method:",
 				options: [
 					{ id: "browser", label: "Browser login (default)" },
@@ -239,11 +262,11 @@ describe("OpenAI Codex OAuth", () => {
 
 	it("cancels when OpenAI Codex login method selection is cancelled", async () => {
 		await expect(
-			openaiCodexOAuthProvider.login({
-				onAuth: () => {},
-				onDeviceCode: () => {},
-				onPrompt: async () => "",
-				onSelect: async () => undefined,
+			openaiCodexOAuth.login({
+				prompt: async () => {
+					throw new Error("Login cancelled");
+				},
+				notify: () => {},
 			}),
 		).rejects.toThrow("Login cancelled");
 	});
@@ -273,7 +296,7 @@ describe("OpenAI Codex OAuth", () => {
 			}),
 		);
 
-		const credentialsPromise = loginOpenAICodexDeviceCode({
+		const credentialsPromise = loginOpenAICodexDeviceCodeForTest({
 			onDeviceCode: () => {},
 			signal: controller.signal,
 		});
@@ -317,7 +340,7 @@ describe("OpenAI Codex OAuth", () => {
 			}),
 		);
 
-		const credentialsPromise = loginOpenAICodexDeviceCode({
+		const credentialsPromise = loginOpenAICodexDeviceCodeForTest({
 			onDeviceCode: () => {},
 		});
 		const rejectionPromise = credentialsPromise.then(
@@ -380,7 +403,7 @@ describe("OpenAI Codex OAuth", () => {
 			}),
 		);
 
-		const credentialsPromise = loginOpenAICodexDeviceCode({
+		const credentialsPromise = loginOpenAICodexDeviceCodeForTest({
 			onDeviceCode: () => {},
 		});
 
@@ -418,7 +441,7 @@ describe("OpenAI Codex OAuth", () => {
 		);
 
 		await expect(
-			loginOpenAICodexDeviceCode({
+			loginOpenAICodexDeviceCodeForTest({
 				onDeviceCode: () => {},
 			}),
 		).rejects.toThrow(
@@ -443,9 +466,14 @@ describe("OpenAI Codex OAuth", () => {
 			}),
 		);
 
-		await expect(refreshOpenAICodexToken("invalid-refresh-token")).rejects.toThrow(
-			/OpenAI Codex token refresh failed \(401\).*Could not validate your token/,
-		);
+		await expect(
+			openaiCodexOAuth.refresh({
+				type: "oauth",
+				access: "invalid-access-token",
+				refresh: "invalid-refresh-token",
+				expires: 0,
+			}),
+		).rejects.toThrow(/OpenAI Codex token refresh failed \(401\).*Could not validate your token/);
 		expect(consoleError).not.toHaveBeenCalled();
 	});
 });

packages/ai/test/openai-responses-copilot-provider.test.ts

@@ -252,11 +252,12 @@ describe("openai-responses provider defaults", () => {
 		expect(captured).toEqual({ sessionId: null, clientRequestId: null });
 	});
 
-	it.each([
-		["gpt-5.4", "priority", 2],
-		["gpt-5.5", "priority", 2.5],
-		["gpt-5.5", "flex", 0.5],
-	] as const)("applies %s %s service-tier cost multiplier", async (modelId, serviceTier, multiplier) => {
+	async function streamServiceTierUsage(
+		modelId: "gpt-5.4" | "gpt-5.5",
+		serviceTier: "priority" | "flex",
+		inputTokens: number,
+		outputTokens: number,
+	) {
 		const model = getModel("openai", modelId);
 		const sse = `${[
 			`data: ${JSON.stringify({
@@ -265,9 +266,9 @@ describe("openai-responses provider defaults", () => {
 					status: "completed",
 					service_tier: serviceTier,
 					usage: {
-						input_tokens: 1000000,
-						output_tokens: 1000000,
-						total_tokens: 2000000,
+						input_tokens: inputTokens,
+						output_tokens: outputTokens,
+						total_tokens: inputTokens + outputTokens,
 						input_tokens_details: { cached_tokens: 0 },
 					},
 				},
@@ -290,10 +291,39 @@ describe("openai-responses provider defaults", () => {
 			{ apiKey: "test-key", serviceTier },
 		);
 
-		const result = await stream.result();
+		return { model, result: await stream.result() };
+	}
 
-		expect(result.usage.cost.input).toBe(model.cost.input * multiplier);
-		expect(result.usage.cost.output).toBe(model.cost.output * multiplier);
-		expect(result.usage.cost.total).toBe((model.cost.input + model.cost.output) * multiplier);
+	it.each([
+		["gpt-5.4", "priority", 2],
+		["gpt-5.5", "priority", 2.5],
+		["gpt-5.5", "flex", 0.5],
+	] as const)("applies %s %s service-tier cost multiplier", async (modelId, serviceTier, multiplier) => {
+		// Stay below the 272K long-context tier threshold so base rates apply.
+		const inputTokens = 200000;
+		const outputTokens = 100000;
+		const { model, result } = await streamServiceTierUsage(modelId, serviceTier, inputTokens, outputTokens);
+
+		const expectedInput = (model.cost.input / 1_000_000) * inputTokens * multiplier;
+		const expectedOutput = (model.cost.output / 1_000_000) * outputTokens * multiplier;
+		expect(result.usage.cost.input).toBe(expectedInput);
+		expect(result.usage.cost.output).toBe(expectedOutput);
+		expect(result.usage.cost.total).toBe(expectedInput + expectedOutput);
+	});
+
+	it("applies the service-tier multiplier on top of long-context tier pricing", async () => {
+		// Above the 272K input threshold the long-context tier rates apply, then the multiplier.
+		const inputTokens = 1000000;
+		const outputTokens = 100000;
+		const multiplier = 2;
+		const { model, result } = await streamServiceTierUsage("gpt-5.4", "priority", inputTokens, outputTokens);
+
+		const tier = model.cost.tiers?.find((entry) => inputTokens > entry.inputTokensAbove);
+		if (!tier) throw new Error("expected gpt-5.4 to define a long-context pricing tier");
+		const expectedInput = (tier.input / 1_000_000) * inputTokens * multiplier;
+		const expectedOutput = (tier.output / 1_000_000) * outputTokens * multiplier;
+		expect(result.usage.cost.input).toBe(expectedInput);
+		expect(result.usage.cost.output).toBe(expectedOutput);
+		expect(result.usage.cost.total).toBe(expectedInput + expectedOutput);
 	});
 });

packages/ai/test/providers.test.ts

@@ -1,6 +1,6 @@
 import { describe, expect, it } from "vitest";
 import { envApiKeyAuth } from "../src/auth/helpers.ts";
-import type { AuthContext } from "../src/auth/types.ts";
+import type { AuthContext, AuthEvent } from "../src/auth/types.ts";
 import { createModels, createProvider } from "../src/models.ts";
 import { builtinModels, builtinProviders } from "../src/providers/all.ts";
 import { amazonBedrockProvider } from "../src/providers/amazon-bedrock.ts";
@@ -49,40 +49,69 @@ describe("builtin providers", () => {
 		models.setProvider(anthropicProvider());
 		const model = models.getModel("anthropic", "claude-haiku-4-5")!;
 
-		const result = await models.getAuth(model);
+		const result = await models.getAuth(model.provider);
 		expect(result?.auth.apiKey).toBe("oauth-token");
 		expect(result?.source).toBe("ANTHROPIC_OAUTH_TOKEN");
 	});
 
+	it("runs provider-owned Bedrock bearer token and AWS profile login flows", async () => {
+		const auth = amazonBedrockProvider().auth.apiKey!;
+		const bearerAnswers = ["bearer-token", "bedrock-token"];
+		expect(
+			await auth.login?.({
+				prompt: async () => bearerAnswers.shift()!,
+				notify: () => {},
+			}),
+		).toEqual({ type: "api_key", key: "bedrock-token" });
+
+		const profileAnswers = ["aws-profile", "work"];
+		const events: AuthEvent[] = [];
+		expect(
+			await auth.login?.({
+				prompt: async () => profileAnswers.shift()!,
+				notify: (event) => events.push(event),
+			}),
+		).toEqual({ type: "api_key", env: { AWS_PROFILE: "work" } });
+		expect(events).toEqual([
+			expect.objectContaining({
+				type: "info",
+				links: [expect.objectContaining({ label: "AWS credential provider chain" })],
+			}),
+		]);
+		expect(
+			await auth.resolve({
+				ctx: fakeAuthContext({}),
+				credential: { type: "api_key", env: { AWS_PROFILE: "work" } },
+			}),
+		).toMatchObject({ auth: {}, env: { AWS_PROFILE: "work" } });
+	});
+
 	it("reports bedrock as configured from ambient AWS credentials without an api key", async () => {
 		const models = createModels({ authContext: fakeAuthContext({ AWS_PROFILE: "dev" }) });
 		models.setProvider(amazonBedrockProvider());
 		const model = models.getModels("amazon-bedrock")[0];
 
-		const result = await models.getAuth(model);
+		const result = await models.getAuth(model.provider);
 		expect(result?.auth).toEqual({});
 		expect(result?.source).toBe("AWS_PROFILE");
 
 		const unconfigured = createModels({ authContext: fakeAuthContext({}) });
 		unconfigured.setProvider(amazonBedrockProvider());
-		expect(await unconfigured.getAuth(model)).toBeUndefined();
+		expect(await unconfigured.getAuth(model.provider)).toBeUndefined();
 	});
 
 	it("requires Cloudflare Workers AI account config and returns scoped env", async () => {
 		const missingAccount = createModels({ authContext: fakeAuthContext({ CLOUDFLARE_API_KEY: "cf-key" }) });
 		missingAccount.setProvider(cloudflareWorkersAIProvider());
 		const model = missingAccount.getModels("cloudflare-workers-ai")[0];
-		expect(await missingAccount.getAuth(model)).toBeUndefined();
+		expect(await missingAccount.getAuth(model.provider)).toBeUndefined();
 
 		const configured = createModels({
 			authContext: fakeAuthContext({ CLOUDFLARE_API_KEY: "cf-key", CLOUDFLARE_ACCOUNT_ID: "account-id" }),
 		});
 		configured.setProvider(cloudflareWorkersAIProvider());
-		const result = await configured.getAuth(model);
-		expect(result?.auth).toEqual({
-			apiKey: "cf-key",
-			baseUrl: "https://api.cloudflare.com/client/v4/accounts/account-id/ai/v1",
-		});
+		const result = await configured.getAuth(model.provider);
+		expect(result?.auth).toEqual({ apiKey: "cf-key" });
 		expect(result?.env).toEqual({ CLOUDFLARE_ACCOUNT_ID: "account-id" });
 	});
 
@@ -92,7 +121,7 @@ describe("builtin providers", () => {
 		});
 		missingGateway.setProvider(cloudflareAIGatewayProvider());
 		const model = missingGateway.getModels("cloudflare-ai-gateway")[0];
-		expect(await missingGateway.getAuth(model)).toBeUndefined();
+		expect(await missingGateway.getAuth(model.provider)).toBeUndefined();
 
 		const configured = createModels({
 			authContext: fakeAuthContext({
@@ -102,14 +131,13 @@ describe("builtin providers", () => {
 			}),
 		});
 		configured.setProvider(cloudflareAIGatewayProvider());
-		const result = await configured.getAuth(model);
+		const result = await configured.getAuth(model.provider);
 		expect(result?.auth).toEqual({
 			headers: {
 				"cf-aig-authorization": "Bearer cf-key",
 				Authorization: null,
 				"x-api-key": null,
 			},
-			baseUrl: "https://gateway.ai.cloudflare.com/v1/account-id/gateway-id/anthropic",
 		});
 		expect(result?.env).toEqual({
 			CLOUDFLARE_ACCOUNT_ID: "account-id",
@@ -117,6 +145,47 @@ describe("builtin providers", () => {
 		});
 	});
 
+	it("runs provider-owned Vertex API key and ADC login flows", async () => {
+		const auth = googleVertexProvider().auth.apiKey!;
+		const keyAnswers = ["api-key", "vertex-key"];
+		expect(
+			await auth.login?.({
+				prompt: async () => keyAnswers.shift()!,
+				notify: () => {},
+			}),
+		).toEqual({ type: "api_key", key: "vertex-key" });
+
+		const adcAnswers = ["adc", "project-id", "us-central1"];
+		const events: AuthEvent[] = [];
+		expect(
+			await auth.login?.({
+				prompt: async () => adcAnswers.shift()!,
+				notify: (event) => events.push(event),
+			}),
+		).toEqual({
+			type: "api_key",
+			env: { GOOGLE_CLOUD_PROJECT: "project-id", GOOGLE_CLOUD_LOCATION: "us-central1" },
+		});
+		expect(events).toEqual([
+			expect.objectContaining({
+				type: "info",
+				links: [expect.objectContaining({ label: "Application Default Credentials" })],
+			}),
+		]);
+		expect(
+			await auth.resolve({
+				ctx: fakeAuthContext({}, ["~/.config/gcloud/application_default_credentials.json"]),
+				credential: {
+					type: "api_key",
+					env: { GOOGLE_CLOUD_PROJECT: "project-id", GOOGLE_CLOUD_LOCATION: "us-central1" },
+				},
+			}),
+		).toMatchObject({
+			auth: {},
+			env: { GOOGLE_CLOUD_PROJECT: "project-id", GOOGLE_CLOUD_LOCATION: "us-central1" },
+		});
+	});
+
 	it("resolves vertex via ADC file plus project and location", async () => {
 		const adc = "~/.config/gcloud/application_default_credentials.json";
 		const configured = createModels({
@@ -125,40 +194,38 @@ describe("builtin providers", () => {
 		configured.setProvider(googleVertexProvider());
 		const model = configured.getModels("google-vertex")[0];
 
-		const result = await configured.getAuth(model);
+		const result = await configured.getAuth(model.provider);
 		expect(result?.auth).toEqual({});
 		expect(result?.source).toContain("application default");
 
 		// ADC without project/location is not configured
 		const partial = createModels({ authContext: fakeAuthContext({ GOOGLE_CLOUD_PROJECT: "proj" }, [adc]) });
 		partial.setProvider(googleVertexProvider());
-		expect(await partial.getAuth(model)).toBeUndefined();
+		expect(await partial.getAuth(model.provider)).toBeUndefined();
 
 		// explicit key wins over ADC
 		const keyed = createModels({ authContext: fakeAuthContext({ GOOGLE_CLOUD_API_KEY: "vertex-key" }) });
 		keyed.setProvider(googleVertexProvider());
-		expect((await keyed.getAuth(model))?.auth.apiKey).toBe("vertex-key");
+		expect((await keyed.getAuth(model.provider))?.auth.apiKey).toBe("vertex-key");
 	});
 });
 
 describe("envApiKeyAuth", () => {
 	it("prefers the stored credential key and falls back through env vars in order", async () => {
 		const auth = envApiKeyAuth("Test key", ["FIRST_KEY", "SECOND_KEY"]);
-		const model = { provider: "p1" } as Model<Api>;
 
 		const stored = await auth.resolve({
-			model,
 			ctx: fakeAuthContext({ FIRST_KEY: "env" }),
 			credential: { type: "api_key", key: "stored" },
 		});
 		expect(stored?.auth.apiKey).toBe("stored");
 		expect(stored?.source).toBe("stored credential");
 
-		const second = await auth.resolve({ model, ctx: fakeAuthContext({ SECOND_KEY: "second" }) });
+		const second = await auth.resolve({ ctx: fakeAuthContext({ SECOND_KEY: "second" }) });
 		expect(second?.auth.apiKey).toBe("second");
 		expect(second?.source).toBe("SECOND_KEY");
 
-		expect(await auth.resolve({ model, ctx: fakeAuthContext({}) })).toBeUndefined();
+		expect(await auth.resolve({ ctx: fakeAuthContext({}) })).toBeUndefined();
 	});
 
 	it("login prompts for a secret and returns an api-key credential", async () => {

packages/ai/test/scratch.ts

@@ -21,7 +21,7 @@ models.setProvider(anthropicProvider());
 const model = models.getModel("anthropic", "claude-haiku-4-5");
 if (!model) throw new Error("model not found");
 
-const auth = await models.getAuth(model);
+const auth = await models.getAuth(model.provider);
 console.log(`model: ${model.provider}/${model.id}`);
 console.log(`auth:  ${auth ? `configured via ${auth.source}` : "not configured"}\n`);
 if (!auth) process.exit(1);

packages/coding-agent/CHANGELOG.md

@@ -2,11 +2,76 @@
 
 ## [Unreleased]
 
+### Breaking Changes
+
+- Replaced the SDK's `CreateAgentSessionOptions.authStorage` and `modelRegistry` options with the async `modelRuntime` option. `AuthStorage` and its storage backends are no longer exported; use `ModelRuntime` (or a custom pi-ai `CredentialStore`), or `readStoredCredential()` for one-off reads of auth.json.
+- Removed redundant `ModelRuntime.getAll()`, `find()`, `getSnapshot()`, and `getAuthOptions()` projections. Use the pi-ai `Models` methods `getModels()`, `getModel()`, `getProviders()`, and `checkAuth()` directly.
+- Replaced SDK request-auth assembly through `ModelRegistry.getApiKeyAndHeaders()` with `ModelRuntime.getAuth()`. Passing a provider ID returns provider-scoped auth; passing a model also resolves built-in, `models.json`, and extension model headers.
+- Changed extension-facing `ModelRegistry.refresh()` from synchronous `void` to `Promise<void>` because `models.json` loading is asynchronous. Extensions must await it before making synchronous registry reads.
+- Removed extension OAuth `modifyModels`. Provider catalogs are now composed independently of credentials; credential-specific availability belongs to canonical provider filtering. The legacy extension OAuth callback and credential types remain available from pi-ai's root and `oauth` subpath.
+
+#### SDK migration
+
+Construct one `ModelRuntime` and pass it to `createAgentSession()`:
+
+```typescript
+// Before
+const authStorage = AuthStorage.create(authPath);
+const modelRegistry = await ModelRegistry.create(authStorage, modelsPath);
+authStorage.setRuntimeApiKey("anthropic", apiKey);
+const { session } = await createAgentSession({ authStorage, modelRegistry });
+
+// After
+const modelRuntime = await ModelRuntime.create({ authPath, modelsPath });
+// Or: ModelRuntime.create({ credentials: myCredentialStore, modelsPath })
+modelRuntime.setRuntimeApiKey("anthropic", apiKey);
+const { session } = await createAgentSession({ modelRuntime });
+```
+
+Replace `ModelRegistry` projections with the corresponding `ModelRuntime`/pi-ai `Models` methods:
+
+```typescript
+const allModels = modelRuntime.getModels();
+const model = modelRuntime.getModel(providerId, modelId);
+const availableModels = await modelRuntime.getAvailable();
+const authStatus = await modelRuntime.checkAuth(providerId);
+const requestAuth = await modelRuntime.getAuth(model); // Includes model headers
+
+modelRuntime.registerProvider(providerId, providerConfig); // Still synchronous
+await modelRuntime.reloadConfig();
+```
+
+`ModelRuntime.stream*()` resolves auth and configured headers itself. Do not call `getAuth(model)` before streaming merely to reconstruct request options. For SDK-level header interception, use the Models-only transform so auth is resolved once:
+
+```typescript
+modelRuntime.streamSimple(model, context, {
+  transformHeaders: async (headers) => ({
+    ...headers,
+    "X-Request-ID": requestId,
+  }),
+});
+```
+
+Use `ModelRuntime` for model lookup, availability, provider auth, login/logout, runtime API-key overrides, provider registration, and config refresh. `ModelRegistry` remains a synchronous-read compatibility facade for extensions; SDK code should use `ModelRuntime`. Extensions that explicitly refresh it must await completion:
+
+```typescript
+await ctx.modelRegistry.refresh();
+const models = ctx.modelRegistry.getAll();
+```
+
+
 ### Added
 
+- Added `ModelRuntime` as the canonical async SDK and internal model/auth facade while preserving the synchronous extension-facing `ModelRegistry` API. `ModelRuntime.create()` accepts any pi-ai `CredentialStore` through its `credentials` option.
+- Added provider-owned `/login` discovery directly from registered pi-ai providers, including ambient auth status and informational links.
 - Added the opt-in `max` thinking level across CLI, SDK, RPC, model selection, and themes. Custom themes can define `thinkingMax`; existing themes fall back to `thinkingXhigh`.
 - Added request-wide input-token pricing tiers to custom model costs in `models.json`, `modelOverrides`, and extension-registered providers.
 
+### Changed
+
+- Changed `ModelRuntime` to compose built-in providers, immutable `models.json` configuration, and extension overlays through ad-hoc pi-ai provider methods.
+- Changed `ModelRuntime` to own final request assembly: `getAuth(model)` includes configured model headers, stream methods resolve auth once, and `before_provider_headers` runs as the Models-only header transform before provider dispatch.
+
 ## [0.80.5] - 2026-07-09
 
 ## [0.80.4] - 2026-07-09

packages/coding-agent/README.md

@@ -453,14 +453,12 @@ See [docs/packages.md](docs/packages.md).
 ### SDK
 
 ```typescript
-import { AuthStorage, createAgentSession, ModelRegistry, SessionManager } from "@earendil-works/pi-coding-agent";
+import { createAgentSession, ModelRuntime, SessionManager } from "@earendil-works/pi-coding-agent";
 
-const authStorage = AuthStorage.create();
-const modelRegistry = ModelRegistry.create(authStorage);
+const modelRuntime = await ModelRuntime.create();
 const { session } = await createAgentSession({
   sessionManager: SessionManager.inMemory(),
-  authStorage,
-  modelRegistry,
+  modelRuntime,
 });
 
 await session.prompt("What files are in the current directory?");

packages/coding-agent/docs/custom-provider.md

@@ -253,6 +253,8 @@ pi.registerProvider("custom-api", {
 });
 ```
 
+The key is resolved for each request. An explicit request `Authorization` header takes precedence over the generated value.
+
 ## OAuth Support
 
 Add OAuth/SSO authentication that integrates with `/login`:
@@ -312,15 +314,6 @@ pi.registerProvider("corporate-ai", {
 
     getApiKey(credentials: OAuthCredentials): string {
       return credentials.access;
-    },
-
-    // Optional: modify models based on user's subscription
-    modifyModels(models, credentials) {
-      const region = decodeRegionFromToken(credentials.access);
-      return models.map(m => ({
-        ...m,
-        baseUrl: `https://${region}.ai.corp.com/v1`
-      }));
     }
   }
 });
@@ -330,7 +323,7 @@ After registration, users can authenticate via `/login corporate-ai`.
 
 ### OAuthLoginCallbacks
 
-The `callbacks` object provides three ways to authenticate:
+The `callbacks` object provides UI-neutral interactions for the provider-owned flow:
 
 ```typescript
 interface OAuthLoginCallbacks {
@@ -345,6 +338,9 @@ interface OAuthLoginCallbacks {
     expiresInSeconds?: number;
   }): void;
 
+  // Show transient progress
+  onProgress?(message: string): void;
+
   // Prompt user for input (for manual token entry)
   onPrompt(params: { message: string }): Promise<string>;
 
@@ -660,7 +656,6 @@ interface ProviderConfig {
     login(callbacks: OAuthLoginCallbacks): Promise<OAuthCredentials>;
     refreshToken(credentials: OAuthCredentials): Promise<OAuthCredentials>;
     getApiKey(credentials: OAuthCredentials): string;
-    modifyModels?(models: Model<Api>[], credentials: OAuthCredentials): Model<Api>[];
   };
 }
 ```

packages/coding-agent/docs/sdk.md

@@ -16,16 +16,12 @@ See [examples/sdk/](../examples/sdk/) for working examples from minimal to full
 ## Quick Start
 
 ```typescript
-import { AuthStorage, createAgentSession, ModelRegistry, SessionManager } from "@earendil-works/pi-coding-agent";
-
-// Set up credential storage and model registry
-const authStorage = AuthStorage.create();
-const modelRegistry = ModelRegistry.create(authStorage);
+import { createAgentSession, ModelRuntime, SessionManager } from "@earendil-works/pi-coding-agent";
 
+const modelRuntime = await ModelRuntime.create();
 const { session } = await createAgentSession({
   sessionManager: SessionManager.inMemory(),
-  authStorage,
-  modelRegistry,
+  modelRuntime,
 });
 
 session.subscribe((event) => {
@@ -369,10 +365,9 @@ When you pass a custom `ResourceLoader`, `cwd` and `agentDir` no longer control
 
 ```typescript
 import { getModel } from "@earendil-works/pi-ai";
-import { AuthStorage, ModelRegistry } from "@earendil-works/pi-coding-agent";
+import { ModelRuntime } from "@earendil-works/pi-coding-agent";
 
-const authStorage = AuthStorage.create();
-const modelRegistry = ModelRegistry.create(authStorage);
+const modelRuntime = await ModelRuntime.create();
 
 // Find specific built-in model (doesn't check if API key exists)
 const opus = getModel("anthropic", "claude-opus-4-5");
@@ -380,10 +375,10 @@ if (!opus) throw new Error("Model not found");
 
 // Find any model by provider/id, including custom models from models.json
 // (doesn't check if API key exists)
-const customModel = modelRegistry.find("my-provider", "my-model");
+const customModel = modelRuntime.getModel("my-provider", "my-model");
 
-// Get only models that have valid API keys configured
-const available = await modelRegistry.getAvailable();
+// Get only models that have valid authentication configured
+const available = await modelRuntime.getAvailable();
 
 const { session } = await createAgentSession({
   model: opus,
@@ -395,8 +390,7 @@ const { session } = await createAgentSession({
     { model: haiku, thinkingLevel: "off" },
   ],
   
-  authStorage,
-  modelRegistry,
+  modelRuntime,
 });
 ```
 
@@ -415,14 +409,14 @@ import {
 
 const cliModel = resolveCliModel({
   cliModel: "anthropic/claude-opus-4-5:high",
-  modelRegistry,
+  modelRuntime,
 });
 if (cliModel.error) throw new Error(cliModel.error);
 if (cliModel.warning) console.warn(cliModel.warning);
 
 const { scopedModels, diagnostics } = await resolveModelScopeWithDiagnostics(
   ["anthropic/*:high", "gpt-5"],
-  modelRegistry,
+  modelRuntime,
 );
 for (const diagnostic of diagnostics) {
   console.warn(diagnostic.message);
@@ -435,40 +429,41 @@ for (const diagnostic of diagnostics) {
 
 ### API Keys and OAuth
 
-API key resolution priority (handled by AuthStorage):
+Authentication resolution priority (handled by `ModelRuntime`):
 1. Runtime overrides (via `setRuntimeApiKey`, not persisted)
 2. Stored credentials in `auth.json` (API keys or OAuth tokens)
 3. Environment variables (`ANTHROPIC_API_KEY`, `OPENAI_API_KEY`, etc.)
 4. Fallback resolver (for custom provider keys from `models.json`)
 
 ```typescript
-import { AuthStorage, ModelRegistry } from "@earendil-works/pi-coding-agent";
+import { InMemoryCredentialStore } from "@earendil-works/pi-ai";
+import { createAgentSession, ModelRuntime } from "@earendil-works/pi-coding-agent";
 
 // Default: uses ~/.pi/agent/auth.json and ~/.pi/agent/models.json
-const authStorage = AuthStorage.create();
-const modelRegistry = ModelRegistry.create(authStorage);
+const modelRuntime = await ModelRuntime.create();
 
-const { session } = await createAgentSession({
-  sessionManager: SessionManager.inMemory(),
-  authStorage,
-  modelRegistry,
-});
+// Provider-owned auth methods and current status
+for (const provider of modelRuntime.getProviders()) {
+  const status = await modelRuntime.checkAuth(provider.id);
+  console.log(provider.name, provider.auth, status);
+}
 
 // Runtime API key override (not persisted to disk)
-authStorage.setRuntimeApiKey("anthropic", "sk-my-temp-key");
+modelRuntime.setRuntimeApiKey("anthropic", "sk-my-temp-key");
 
-// Custom auth storage location
-const customAuth = AuthStorage.create("/my/app/auth.json");
-const customRegistry = ModelRegistry.create(customAuth, "/my/app/models.json");
-
-const { session } = await createAgentSession({
-  sessionManager: SessionManager.inMemory(),
-  authStorage: customAuth,
-  modelRegistry: customRegistry,
+// Custom credential and model locations
+const customRuntime = await ModelRuntime.create({
+  authPath: "/my/app/auth.json",
+  modelsPath: "/my/app/models.json",
 });
 
-// No custom models.json (built-in models only)
-const simpleRegistry = ModelRegistry.inMemory(authStorage);
+// Or inject any pi-ai CredentialStore
+const credentials = new InMemoryCredentialStore();
+const inMemoryRuntime = await ModelRuntime.create({ credentials });
+
+const { session } = await createAgentSession({
+  modelRuntime: customRuntime,
+});
 ```
 
 > See [examples/sdk/09-api-keys-and-oauth.ts](../examples/sdk/09-api-keys-and-oauth.ts)
@@ -927,26 +922,22 @@ interface LoadExtensionsResult {
 import { getModel } from "@earendil-works/pi-ai";
 import { Type } from "typebox";
 import {
-  AuthStorage,
   createAgentSession,
   DefaultResourceLoader,
   defineTool,
-  ModelRegistry,
+  ModelRuntime,
   SessionManager,
   SettingsManager,
 } from "@earendil-works/pi-coding-agent";
 
-// Set up auth storage (custom location)
-const authStorage = AuthStorage.create("/custom/agent/auth.json");
-
-// Runtime API key override (not persisted)
+const modelRuntime = await ModelRuntime.create({
+  authPath: "/custom/agent/auth.json",
+  modelsPath: "/custom/agent/models.json",
+});
 if (process.env.MY_KEY) {
-  authStorage.setRuntimeApiKey("anthropic", process.env.MY_KEY);
+  modelRuntime.setRuntimeApiKey("anthropic", process.env.MY_KEY);
 }
 
-// Model registry (no custom models.json)
-const modelRegistry = ModelRegistry.create(authStorage);
-
 // Inline tool
 const statusTool = defineTool({
   name: "status",
@@ -982,8 +973,7 @@ const { session } = await createAgentSession({
 
   model,
   thinkingLevel: "off",
-  authStorage,
-  modelRegistry,
+  modelRuntime,
 
   tools: ["read", "bash", "status"],
   customTools: [statusTool],
@@ -1149,8 +1139,8 @@ createAgentSessionRuntime
 AgentSessionRuntime
 
 // Auth and Models
-AuthStorage
-ModelRegistry
+ModelRuntime // implements pi-ai Models and owns credential storage
+ModelRegistry // synchronous extension compatibility facade
 resolveCliModel
 resolveModelScopeWithDiagnostics
 

packages/coding-agent/examples/extensions/custom-provider-anthropic/index.ts

@@ -46,7 +46,7 @@ import {
 import type { ExtensionAPI } from "@earendil-works/pi-coding-agent";
 
 // =============================================================================
-// OAuth Implementation (copied from packages/ai/src/utils/oauth/anthropic.ts)
+// OAuth implementation adapted for the legacy extension compatibility interface.
 // =============================================================================
 
 const decode = (s: string) => atob(s);

packages/coding-agent/examples/sdk/02-custom-model.ts

@@ -5,11 +5,9 @@
  */
 
 import { getModel } from "@earendil-works/pi-ai/compat";
-import { AuthStorage, createAgentSession, ModelRegistry } from "@earendil-works/pi-coding-agent";
+import { createAgentSession, ModelRuntime } from "@earendil-works/pi-coding-agent";
 
-// Set up auth storage and model registry
-const authStorage = AuthStorage.create();
-const modelRegistry = ModelRegistry.create(authStorage);
+const modelRuntime = await ModelRuntime.create();
 
 // Option 1: Find a specific built-in model by provider/id
 const opus = getModel("anthropic", "claude-opus-4-5");
@@ -18,13 +16,13 @@ if (opus) {
 }
 
 // Option 2: Find model via registry (includes custom models from models.json)
-const customModel = modelRegistry.find("my-provider", "my-model");
+const customModel = modelRuntime.getModel("my-provider", "my-model");
 if (customModel) {
 	console.log(`Found custom model: ${customModel.provider}/${customModel.id}`);
 }
 
 // Option 3: Pick from available models (have valid API keys)
-const available = await modelRegistry.getAvailable();
+const available = await modelRuntime.getAvailable();
 console.log(
 	"Available models:",
 	available.map((m) => `${m.provider}/${m.id}`),
@@ -34,8 +32,7 @@ if (available.length > 0) {
 	const { session } = await createAgentSession({
 		model: available[0],
 		thinkingLevel: "medium", // off, low, medium, high
-		authStorage,
-		modelRegistry,
+		modelRuntime,
 	});
 
 	try {

packages/coding-agent/examples/sdk/09-api-keys-and-oauth.ts

@@ -1,52 +1,34 @@
 /**
  * API Keys and OAuth
  *
- * Configure API key resolution via AuthStorage and ModelRegistry.
+ * Configure provider auth through ModelRuntime.
  */
 
-import { AuthStorage, createAgentSession, ModelRegistry, SessionManager } from "@earendil-works/pi-coding-agent";
-
-// Default: AuthStorage uses ~/.pi/agent/auth.json
-// ModelRegistry loads built-in + custom models from ~/.pi/agent/models.json
-const authStorage = AuthStorage.create();
-const modelRegistry = ModelRegistry.create(authStorage);
+import { createAgentSession, ModelRuntime, SessionManager } from "@earendil-works/pi-coding-agent";
 
+const modelRuntime = await ModelRuntime.create();
 const { session: defaultAuthSession } = await createAgentSession({
 	sessionManager: SessionManager.inMemory(),
-	authStorage,
-	modelRegistry,
+	modelRuntime,
 });
-console.log("Session with default auth storage and model registry");
+console.log("Session with default model runtime");
 defaultAuthSession.dispose();
 
-// Custom auth storage location
-const customAuthStorage = AuthStorage.create("/tmp/my-app/auth.json");
-const customModelRegistry = ModelRegistry.create(customAuthStorage, "/tmp/my-app/models.json");
-
+const customRuntime = await ModelRuntime.create({
+	authPath: "/tmp/my-app/auth.json",
+	modelsPath: "/tmp/my-app/models.json",
+});
 const { session: customAuthSession } = await createAgentSession({
 	sessionManager: SessionManager.inMemory(),
-	authStorage: customAuthStorage,
-	modelRegistry: customModelRegistry,
+	modelRuntime: customRuntime,
 });
-console.log("Session with custom auth storage location");
+console.log("Session with custom auth and models locations");
 customAuthSession.dispose();
 
-// Runtime API key override (not persisted to disk)
-authStorage.setRuntimeApiKey("anthropic", "sk-my-temp-key");
+modelRuntime.setRuntimeApiKey("anthropic", "sk-my-temp-key");
 const { session: runtimeKeySession } = await createAgentSession({
 	sessionManager: SessionManager.inMemory(),
-	authStorage,
-	modelRegistry,
+	modelRuntime,
 });
 console.log("Session with runtime API key override");
 runtimeKeySession.dispose();
-
-// No models.json - only built-in models
-const simpleRegistry = ModelRegistry.inMemory(authStorage);
-const { session: builtInModelsSession } = await createAgentSession({
-	sessionManager: SessionManager.inMemory(),
-	authStorage,
-	modelRegistry: simpleRegistry,
-});
-console.log("Session with only built-in models");
-builtInModelsSession.dispose();

packages/coding-agent/examples/sdk/12-full-control.ts

@@ -6,26 +6,22 @@
 
 import { getModel } from "@earendil-works/pi-ai/compat";
 import {
-	AuthStorage,
 	createAgentSession,
 	createExtensionRuntime,
-	ModelRegistry,
+	ModelRuntime,
 	type ResourceLoader,
 	SessionManager,
 	SettingsManager,
 } from "@earendil-works/pi-coding-agent";
 
-// Custom auth storage location
-const authStorage = AuthStorage.create("/tmp/my-agent/auth.json");
-
-// Runtime API key override (not persisted)
+const modelRuntime = await ModelRuntime.create({
+	authPath: "/tmp/my-agent/auth.json",
+	modelsPath: "/tmp/my-agent/models.json",
+});
 if (process.env.MY_ANTHROPIC_KEY) {
-	authStorage.setRuntimeApiKey("anthropic", process.env.MY_ANTHROPIC_KEY);
+	modelRuntime.setRuntimeApiKey("anthropic", process.env.MY_ANTHROPIC_KEY);
 }
 
-// Model registry with no custom models.json
-const modelRegistry = ModelRegistry.inMemory(authStorage);
-
 const model = getModel("anthropic", "claude-sonnet-4-5");
 if (!model) throw new Error("Model not found");
 
@@ -55,8 +51,7 @@ const { session } = await createAgentSession({
 	agentDir: "/tmp/my-agent",
 	model,
 	thinkingLevel: "off",
-	authStorage,
-	modelRegistry,
+	modelRuntime,
 	resourceLoader,
 	tools: ["read", "bash"],
 	sessionManager: SessionManager.inMemory(cwd),

packages/coding-agent/examples/sdk/README.md

@@ -34,46 +34,44 @@ npx tsx examples/sdk/01-minimal.ts
 ```typescript
 import { getModel } from "@earendil-works/pi-ai";
 import {
-  AuthStorage,
   createAgentSession,
   DefaultResourceLoader,
-  ModelRegistry,
+  ModelRuntime,
   SessionManager,
   SettingsManager,
 } from "@earendil-works/pi-coding-agent";
 
-// Auth and models setup
-const authStorage = AuthStorage.create();
-const modelRegistry = ModelRegistry.create(authStorage);
+const modelRuntime = await ModelRuntime.create();
 
 // Minimal
-const { session } = await createAgentSession({ authStorage, modelRegistry });
+const { session } = await createAgentSession({ modelRuntime });
 
 // Custom model
 const model = getModel("anthropic", "claude-opus-4-5");
-const { session } = await createAgentSession({ model, thinkingLevel: "high", authStorage, modelRegistry });
+const { session } = await createAgentSession({ model, thinkingLevel: "high", modelRuntime });
 
 // Modify prompt
 const loader = new DefaultResourceLoader({
   systemPromptOverride: (base) => `${base}\n\nBe concise.`,
 });
 await loader.reload();
-const { session } = await createAgentSession({ resourceLoader: loader, authStorage, modelRegistry });
+const { session } = await createAgentSession({ resourceLoader: loader, modelRuntime });
 
 // Read-only
-const { session } = await createAgentSession({ tools: ["read", "grep", "find", "ls"], authStorage, modelRegistry });
+const { session } = await createAgentSession({ tools: ["read", "grep", "find", "ls"], modelRuntime });
 
 // In-memory
 const { session } = await createAgentSession({
   sessionManager: SessionManager.inMemory(),
-  authStorage,
-  modelRegistry,
+  modelRuntime,
 });
 
 // Full control
-const customAuth = AuthStorage.create("/my/app/auth.json");
-customAuth.setRuntimeApiKey("anthropic", process.env.MY_KEY!);
-const customRegistry = ModelRegistry.create(customAuth);
+const customRuntime = await ModelRuntime.create({
+  authPath: "/my/app/auth.json",
+  modelsPath: "/my/app/models.json",
+});
+customRuntime.setRuntimeApiKey("anthropic", process.env.MY_KEY!);
 
 const resourceLoader = new DefaultResourceLoader({
   systemPromptOverride: () => "You are helpful.",
@@ -86,8 +84,7 @@ await resourceLoader.reload();
 
 const { session } = await createAgentSession({
   model,
-  authStorage: customAuth,
-  modelRegistry: customRegistry,
+  modelRuntime: customRuntime,
   resourceLoader,
   tools: ["read", "bash", "my_tool"],
   customTools: [myTool],
@@ -108,8 +105,7 @@ await session.prompt("Hello");
 
 | Option | Default | Description |
 |--------|---------|-------------|
-| `authStorage` | `AuthStorage.create()` | Credential storage |
-| `modelRegistry` | `ModelRegistry.create(authStorage)` | Model registry |
+| `modelRuntime` | Runtime using `agentDir/auth.json` and `models.json` | Canonical model and authentication runtime |
 | `cwd` | `process.cwd()` | Working directory |
 | `agentDir` | `~/.pi/agent` | Config directory |
 | `model` | From settings/first available | Model to use |

packages/coding-agent/src/cli/list-models.ts

@@ -6,7 +6,7 @@ import type { Api, Model } from "@earendil-works/pi-ai";
 import { fuzzyFilter } from "@earendil-works/pi-tui";
 import chalk from "chalk";
 import { formatNoModelsAvailableMessage } from "../core/auth-guidance.ts";
-import type { ModelRegistry } from "../core/model-registry.ts";
+import type { ModelRuntime } from "../core/model-runtime.ts";
 
 /**
  * Format a number as human-readable (e.g., 200000 -> "200K", 1000000 -> "1M")
@@ -26,13 +26,13 @@ function formatTokenCount(count: number): string {
 /**
  * List available models, optionally filtered by search pattern
  */
-export async function listModels(modelRegistry: ModelRegistry, searchPattern?: string): Promise<void> {
-	const loadError = modelRegistry.getError();
+export async function listModels(modelRuntime: ModelRuntime, searchPattern?: string): Promise<void> {
+	const loadError = modelRuntime.getError();
 	if (loadError) {
 		console.error(chalk.yellow(`Warning: errors loading models.json:\n${loadError}`));
 	}
 
-	const models = modelRegistry.getAvailable();
+	const models = [...(await modelRuntime.getAvailable())];
 
 	if (models.length === 0) {
 		console.log(formatNoModelsAvailableMessage());

packages/coding-agent/src/core/agent-session-services.ts

@@ -3,9 +3,8 @@ import type { ThinkingLevel } from "@earendil-works/pi-agent-core";
 import type { Model } from "@earendil-works/pi-ai";
 import { getAgentDir } from "../config.ts";
 import { resolvePath } from "../utils/paths.ts";
-import { AuthStorage } from "./auth-storage.ts";
 import type { SessionStartEvent, ToolDefinition } from "./extensions/index.ts";
-import { ModelRegistry } from "./model-registry.ts";
+import { ModelRuntime } from "./model-runtime.ts";
 import {
 	DefaultResourceLoader,
 	type DefaultResourceLoaderOptions,
@@ -38,9 +37,8 @@ export interface AgentSessionRuntimeDiagnostic {
 export interface CreateAgentSessionServicesOptions {
 	cwd: string;
 	agentDir?: string;
-	authStorage?: AuthStorage;
 	settingsManager?: SettingsManager;
-	modelRegistry?: ModelRegistry;
+	modelRuntime?: ModelRuntime;
 	extensionFlagValues?: Map<string, boolean | string>;
 	resourceLoaderOptions?: Omit<DefaultResourceLoaderOptions, "cwd" | "agentDir" | "settingsManager">;
 	resourceLoaderReloadOptions?: ResourceLoaderReloadOptions;
@@ -74,9 +72,8 @@ export interface CreateAgentSessionFromServicesOptions {
 export interface AgentSessionServices {
 	cwd: string;
 	agentDir: string;
-	authStorage: AuthStorage;
+	modelRuntime: ModelRuntime;
 	settingsManager: SettingsManager;
-	modelRegistry: ModelRegistry;
 	resourceLoader: ResourceLoader;
 	diagnostics: AgentSessionRuntimeDiagnostic[];
 }
@@ -139,9 +136,13 @@ export async function createAgentSessionServices(
 ): Promise<AgentSessionServices> {
 	const cwd = resolvePath(options.cwd);
 	const agentDir = options.agentDir ? resolvePath(options.agentDir) : getAgentDir();
-	const authStorage = options.authStorage ?? AuthStorage.create(join(agentDir, "auth.json"));
+	const modelRuntime =
+		options.modelRuntime ??
+		(await ModelRuntime.create({
+			authPath: join(agentDir, "auth.json"),
+			modelsPath: join(agentDir, "models.json"),
+		}));
 	const settingsManager = options.settingsManager ?? SettingsManager.create(cwd, agentDir);
-	const modelRegistry = options.modelRegistry ?? ModelRegistry.create(authStorage, join(agentDir, "models.json"));
 	const resourceLoader = new DefaultResourceLoader({
 		...(options.resourceLoaderOptions ?? {}),
 		cwd,
@@ -154,7 +155,7 @@ export async function createAgentSessionServices(
 	const extensionsResult = resourceLoader.getExtensions();
 	for (const { name, config, extensionPath } of extensionsResult.runtime.pendingProviderRegistrations) {
 		try {
-			modelRegistry.registerProvider(name, config);
+			modelRuntime.registerProvider(name, config);
 		} catch (error) {
 			const message = error instanceof Error ? error.message : String(error);
 			diagnostics.push({
@@ -169,9 +170,8 @@ export async function createAgentSessionServices(
 	return {
 		cwd,
 		agentDir,
-		authStorage,
+		modelRuntime,
 		settingsManager,
-		modelRegistry,
 		resourceLoader,
 		diagnostics,
 	};
@@ -190,9 +190,8 @@ export async function createAgentSessionFromServices(
 	return createAgentSession({
 		cwd: options.services.cwd,
 		agentDir: options.services.agentDir,
-		authStorage: options.services.authStorage,
+		modelRuntime: options.services.modelRuntime,
 		settingsManager: options.services.settingsManager,
-		modelRegistry: options.services.modelRegistry,
 		resourceLoader: options.services.resourceLoader,
 		sessionManager: options.sessionManager,
 		model: options.model,

packages/coding-agent/src/core/agent-session.ts

@@ -24,7 +24,15 @@ import type {
 	PrepareNextTurnContext,
 	ThinkingLevel,
 } from "@earendil-works/pi-agent-core";
-import type { AssistantMessage, ImageContent, Message, Model, TextContent } from "@earendil-works/pi-ai/compat";
+import type {
+	AssistantMessage,
+	AuthResult,
+	ImageContent,
+	Message,
+	Model,
+	ProviderHeaders,
+	TextContent,
+} from "@earendil-works/pi-ai/compat";
 import {
 	clampThinkingLevel,
 	cleanupSessionResources,
@@ -83,7 +91,8 @@ import {
 } from "./extensions/index.ts";
 import { emitSessionShutdownEvent } from "./extensions/runner.ts";
 import type { BashExecutionMessage, CustomMessage } from "./messages.ts";
-import type { ModelRegistry } from "./model-registry.ts";
+import { ModelRegistry } from "./model-registry.ts";
+import type { ModelRuntime } from "./model-runtime.ts";
 import { expandPromptTemplate, type PromptTemplate } from "./prompt-templates.ts";
 import type { ResourceExtensionPaths, ResourceLoader } from "./resource-loader.ts";
 import type { BranchSummaryEntry, CompactionEntry, SessionEntry, SessionManager } from "./session-manager.ts";
@@ -159,6 +168,12 @@ export type AgentSessionEventListener = (event: AgentSessionEvent) => void;
 // Types
 // ============================================================================
 
+function withoutDeletedHeaders(headers: ProviderHeaders | undefined): Record<string, string> | undefined {
+	return headers
+		? Object.fromEntries(Object.entries(headers).filter((entry): entry is [string, string] => entry[1] !== null))
+		: undefined;
+}
+
 export interface AgentSessionConfig {
 	agent: Agent;
 	sessionManager: SessionManager;
@@ -170,8 +185,8 @@ export interface AgentSessionConfig {
 	resourceLoader: ResourceLoader;
 	/** SDK custom tools registered outside extensions */
 	customTools?: ToolDefinition[];
-	/** Model registry for API key resolution and model discovery */
-	modelRegistry: ModelRegistry;
+	/** Canonical model/auth runtime used by coding-agent internals. */
+	modelRuntime: ModelRuntime;
 	/** Initial active built-in tool names. Default: [read, bash, edit, write] */
 	initialActiveToolNames?: string[];
 	/** Optional allowlist of tool names. When provided, only these tool names are exposed. */
@@ -325,8 +340,7 @@ export class AgentSession {
 	private _extensionErrorListener?: ExtensionErrorListener;
 	private _extensionErrorUnsubscriber?: () => void;
 
-	// Model registry for API key resolution
-	private _modelRegistry: ModelRegistry;
+	private _modelRuntime: ModelRuntime;
 
 	// Tool registry for extension getTools/setTools
 	private _toolRegistry: Map<string, AgentTool> = new Map();
@@ -347,7 +361,7 @@ export class AgentSession {
 		this._resourceLoader = config.resourceLoader;
 		this._customTools = config.customTools ?? [];
 		this._cwd = config.cwd;
-		this._modelRegistry = config.modelRegistry;
+		this._modelRuntime = config.modelRuntime;
 		this._extensionRunnerRef = config.extensionRunnerRef;
 		this._initialActiveToolNames = config.initialActiveToolNames;
 		this._allowedToolNames = config.allowedToolNames ? new Set(config.allowedToolNames) : undefined;
@@ -367,9 +381,8 @@ export class AgentSession {
 		});
 	}
 
-	/** Model registry for API key resolution and model discovery */
-	get modelRegistry(): ModelRegistry {
-		return this._modelRegistry;
+	get modelRuntime(): ModelRuntime {
+		return this._modelRuntime;
 	}
 
 	private async _getRequiredRequestAuth(model: Model<any>): Promise<{
@@ -377,18 +390,25 @@ export class AgentSession {
 		headers?: Record<string, string>;
 		env?: Record<string, string>;
 	}> {
-		const result = await this._modelRegistry.getApiKeyAndHeaders(model);
-		if (!result.ok) {
-			if (result.error.startsWith("No API key found")) {
+		let result: AuthResult | undefined;
+		try {
+			result = await this._modelRuntime.getAuth(model);
+		} catch (error) {
+			const cause = error instanceof Error ? error.cause : undefined;
+			if (cause instanceof Error && cause.message === "authHeader requires a resolved API key") {
 				throw new Error(formatNoApiKeyFoundMessage(model.provider));
 			}
-			throw new Error(result.error);
+			throw error;
 		}
-		if (result.apiKey) {
-			return { apiKey: result.apiKey, headers: result.headers, env: result.env };
+		if (result?.auth.apiKey) {
+			return {
+				apiKey: result.auth.apiKey,
+				headers: withoutDeletedHeaders(result.auth.headers),
+				env: result.env,
+			};
 		}
 
-		const isOAuth = this._modelRegistry.isUsingOAuth(model);
+		const isOAuth = this._modelRuntime.isUsingOAuth(model.provider);
 		if (isOAuth) {
 			throw new Error(
 				`Authentication failed for "${model.provider}". ` +
@@ -408,8 +428,14 @@ export class AgentSession {
 			return this._getRequiredRequestAuth(model);
 		}
 
-		const result = await this._modelRegistry.getApiKeyAndHeaders(model);
-		return result.ok ? { apiKey: result.apiKey, headers: result.headers, env: result.env } : {};
+		try {
+			const result = await this._modelRuntime.getAuth(model);
+			return result
+				? { apiKey: result.auth.apiKey, headers: withoutDeletedHeaders(result.auth.headers), env: result.env }
+				: {};
+		} catch {
+			return {};
+		}
 	}
 
 	/**
@@ -1141,8 +1167,11 @@ export class AgentSession {
 				throw new Error(formatNoModelSelectedMessage());
 			}
 
-			if (!this._modelRegistry.hasConfiguredAuth(this.model)) {
-				const isOAuth = this._modelRegistry.isUsingOAuth(this.model);
+			const hasConfiguredAuth =
+				this._modelRuntime.hasConfiguredAuth(this.model.provider) ||
+				(await this._modelRuntime.checkAuth(this.model.provider)) !== undefined;
+			if (!hasConfiguredAuth) {
+				const isOAuth = this._modelRuntime.isUsingOAuth(this.model.provider);
 				if (isOAuth) {
 					throw new Error(
 						`Authentication failed for "${this.model.provider}". ` +
@@ -1535,7 +1564,7 @@ export class AgentSession {
 	 * @throws Error if no auth is configured for the model
 	 */
 	async setModel(model: Model<any>): Promise<void> {
-		if (!this._modelRegistry.hasConfiguredAuth(model)) {
+		if (!(await this._modelRuntime.checkAuth(model.provider))) {
 			throw new Error(`No API key for ${model.provider}/${model.id}`);
 		}
 
@@ -1565,7 +1594,13 @@ export class AgentSession {
 	}
 
 	private async _cycleScopedModel(direction: "forward" | "backward"): Promise<ModelCycleResult | undefined> {
-		const scopedModels = this._scopedModels.filter((scoped) => this._modelRegistry.hasConfiguredAuth(scoped.model));
+		const checks = await Promise.all(
+			this._scopedModels.map(async (scoped) => ({
+				scoped,
+				auth: await this._modelRuntime.checkAuth(scoped.model.provider),
+			})),
+		);
+		const scopedModels = checks.filter(({ auth }) => auth !== undefined).map(({ scoped }) => scoped);
 		if (scopedModels.length <= 1) return undefined;
 
 		const currentModel = this.model;
@@ -1594,7 +1629,7 @@ export class AgentSession {
 	}
 
 	private async _cycleAvailableModel(direction: "forward" | "backward"): Promise<ModelCycleResult | undefined> {
-		const availableModels = await this._modelRegistry.getAvailable();
+		const availableModels = await this._modelRuntime.getAvailable();
 		if (availableModels.length <= 1) return undefined;
 
 		const currentModel = this.model;
@@ -2004,12 +2039,10 @@ export class AgentSession {
 			let headers: Record<string, string> | undefined;
 			let env: Record<string, string> | undefined;
 			if (this.agent.streamFn === streamSimple) {
-				const authResult = await this._modelRegistry.getApiKeyAndHeaders(this.model);
-				if (!authResult.ok || !authResult.apiKey) {
-					return false;
-				}
-				apiKey = authResult.apiKey;
-				headers = authResult.headers;
+				const authResult = await this._modelRuntime.getAuth(this.model);
+				if (!authResult?.auth.apiKey) return false;
+				apiKey = authResult.auth.apiKey;
+				headers = withoutDeletedHeaders(authResult.auth.headers);
 				env = authResult.env;
 			} else {
 				({ apiKey, headers, env } = await this._getCompactionRequestAuth(this.model));
@@ -2267,7 +2300,7 @@ export class AgentSession {
 			return;
 		}
 
-		const refreshedModel = this._modelRegistry.find(currentModel.provider, currentModel.id);
+		const refreshedModel = this._modelRuntime.getModel(currentModel.provider, currentModel.id);
 		if (!refreshedModel || refreshedModel === currentModel) {
 			return;
 		}
@@ -2343,7 +2376,7 @@ export class AgentSession {
 				refreshTools: () => this._refreshToolRegistry(),
 				getCommands,
 				setModel: async (model) => {
-					if (!this.modelRegistry.hasConfiguredAuth(model)) return false;
+					if (!this._modelRuntime.hasConfiguredAuth(model.provider)) return false;
 					await this.setModel(model);
 					return true;
 				},
@@ -2383,11 +2416,11 @@ export class AgentSession {
 			},
 			{
 				registerProvider: (name, config) => {
-					this._modelRegistry.registerProvider(name, config);
+					this._modelRuntime.registerProvider(name, config);
 					this._refreshCurrentModelFromRegistry();
 				},
 				unregisterProvider: (name) => {
-					this._modelRegistry.unregisterProvider(name);
+					this._modelRuntime.unregisterProvider(name);
 					this._refreshCurrentModelFromRegistry();
 				},
 			},
@@ -2523,7 +2556,7 @@ export class AgentSession {
 			extensionsResult.runtime,
 			this._cwd,
 			this.sessionManager,
-			this._modelRegistry,
+			new ModelRegistry(this._modelRuntime),
 		);
 		if (this._extensionRunnerRef) {
 			this._extensionRunnerRef.current = this._extensionRunner;

packages/coding-agent/src/core/auth-storage.ts

@@ -1,19 +1,9 @@
 /**
- * Credential storage for API keys and OAuth tokens.
- * Handles loading, saving, and refreshing credentials from auth.json.
- *
- * Uses file locking to prevent race conditions when multiple pi instances
- * try to refresh tokens simultaneously.
+ * CredentialStore implementation backed by auth.json.
+ * Provider auth orchestration belongs to ModelRuntime and pi-ai Models.
  */
 
-import {
-	findEnvKeys,
-	getEnvApiKey,
-	type OAuthCredentials,
-	type OAuthLoginCallbacks,
-	type OAuthProviderId,
-} from "@earendil-works/pi-ai/compat";
-import { getOAuthApiKey, getOAuthProvider, getOAuthProviders } from "@earendil-works/pi-ai/oauth";
+import type { Credential, CredentialInfo, CredentialStore } from "@earendil-works/pi-ai";
 import { chmodSync, existsSync, mkdirSync, readFileSync, writeFileSync } from "fs";
 import { dirname, join } from "path";
 import lockfile from "proper-lockfile";
@@ -21,29 +11,7 @@ import { getAgentDir } from "../config.ts";
 import { normalizePath } from "../utils/paths.ts";
 import { resolveConfigValue } from "./resolve-config-value.ts";
 
-export type ApiKeyCredential = {
-	type: "api_key";
-	key: string;
-	env?: Record<string, string>;
-};
-
-export type OAuthCredential = {
-	type: "oauth";
-} & OAuthCredentials;
-
-export type AuthCredential = ApiKeyCredential | OAuthCredential;
-
-export type AuthStorageData = Record<string, AuthCredential>;
-
-export type AuthStatus = {
-	configured: boolean;
-	source?: "stored" | "runtime" | "environment" | "fallback" | "models_json_key" | "models_json_command";
-	label?: string;
-};
-
-export interface GetApiKeyOptions {
-	includeFallback?: boolean;
-}
+type AuthStorageData = Record<string, Credential>;
 
 type LockResult<T> = {
 	result: T;
@@ -200,11 +168,8 @@ export class InMemoryAuthStorageBackend implements AuthStorageBackend {
 /**
  * Credential storage backed by a JSON file.
  */
-export class AuthStorage {
+export class AuthStorage implements CredentialStore {
 	private data: AuthStorageData = {};
-	private runtimeOverrides: Map<string, string> = new Map();
-	private loadError: Error | null = null;
-	private errors: Error[] = [];
 	private storage: AuthStorageBackend;
 
 	private constructor(storage: AuthStorageBackend) {
@@ -226,26 +191,6 @@ export class AuthStorage {
 		return AuthStorage.fromStorage(storage);
 	}
 
-	/**
-	 * Set a runtime API key override (not persisted to disk).
-	 * Used for CLI --api-key flag.
-	 */
-	setRuntimeApiKey(provider: string, apiKey: string): void {
-		this.runtimeOverrides.set(provider, apiKey);
-	}
-
-	/**
-	 * Remove a runtime API key override.
-	 */
-	removeRuntimeApiKey(provider: string): void {
-		this.runtimeOverrides.delete(provider);
-	}
-
-	private recordError(error: unknown): void {
-		const normalizedError = error instanceof Error ? error : new Error(String(error));
-		this.errors.push(normalizedError);
-	}
-
 	private parseStorageData(content: string | undefined): AuthStorageData {
 		if (!content) {
 			return {};
@@ -264,276 +209,63 @@ export class AuthStorage {
 				return { result: undefined };
 			});
 			this.data = this.parseStorageData(content);
-			this.loadError = null;
-		} catch (error) {
-			this.loadError = error as Error;
-			this.recordError(error);
+		} catch {
+			// Preserve the last valid in-memory snapshot.
 		}
 	}
 
-	private persistProviderChange(provider: string, credential: AuthCredential | undefined): AuthStorageData {
-		if (this.loadError) {
-			this.reload();
-		}
-
-		if (this.loadError) {
-			const error = new Error(
-				`Cannot update auth storage because it could not be loaded: ${this.loadError.message}`,
-			);
-			this.recordError(error);
-			throw error;
-		}
-
-		try {
-			let persistedData: AuthStorageData = {};
-			this.storage.withLock((current) => {
-				const currentData = this.parseStorageData(current);
-				const merged: AuthStorageData = { ...currentData };
-				if (credential) {
-					merged[provider] = credential;
-				} else {
-					delete merged[provider];
-				}
-				persistedData = merged;
-				return { result: undefined, next: JSON.stringify(merged, null, 2) };
-			});
-			this.loadError = null;
-			return persistedData;
-		} catch (error) {
-			this.recordError(error);
-			throw error;
-		}
+	async read(provider: string): Promise<Credential | undefined> {
+		const credential = this.data[provider];
+		if (credential?.type !== "api_key") return credential;
+		if (credential.key === undefined) return credential;
+		return { ...credential, key: resolveConfigValue(credential.key, credential.env) };
 	}
 
-	/**
-	 * Get credential for a provider.
-	 */
-	get(provider: string): AuthCredential | undefined {
-		return this.data[provider] ?? undefined;
-	}
-
-	/**
-	 * Get provider-scoped environment values for an API key credential.
-	 */
-	getProviderEnv(provider: string): Record<string, string> | undefined {
-		const cred = this.data[provider];
-		return cred?.type === "api_key" && cred.env ? { ...cred.env } : undefined;
-	}
-
-	/**
-	 * Set credential for a provider.
-	 */
-	set(provider: string, credential: AuthCredential): void {
-		this.data = this.persistProviderChange(provider, credential);
-	}
-
-	/**
-	 * Remove credential for a provider.
-	 */
-	remove(provider: string): void {
-		this.data = this.persistProviderChange(provider, undefined);
-	}
-
-	/**
-	 * List all providers with credentials.
-	 */
-	list(): string[] {
-		return Object.keys(this.data);
-	}
-
-	/**
-	 * Check if credentials exist for a provider in auth.json.
-	 */
-	has(provider: string): boolean {
-		return provider in this.data;
-	}
-
-	/**
-	 * Check if any form of auth is configured for a provider.
-	 * Unlike getApiKey(), this doesn't refresh OAuth tokens.
-	 */
-	hasAuth(provider: string): boolean {
-		if (this.runtimeOverrides.has(provider)) return true;
-		if (this.data[provider]) return true;
-		if (getEnvApiKey(provider)) return true;
-		return false;
-	}
-
-	/**
-	 * Return auth status without exposing credential values or refreshing tokens.
-	 */
-	getAuthStatus(provider: string): AuthStatus {
-		if (this.data[provider]) {
-			return { configured: true, source: "stored" };
-		}
-
-		if (this.runtimeOverrides.has(provider)) {
-			return { configured: false, source: "runtime", label: "--api-key" };
-		}
-
-		const envKeys = findEnvKeys(provider);
-		if (envKeys?.[0]) {
-			return { configured: false, source: "environment", label: envKeys[0] };
-		}
-
-		return { configured: false };
-	}
-
-	/**
-	 * Get all credentials (for passing to getOAuthApiKey).
-	 */
-	getAll(): AuthStorageData {
-		return { ...this.data };
-	}
-
-	drainErrors(): Error[] {
-		const drained = [...this.errors];
-		this.errors = [];
-		return drained;
-	}
-
-	/**
-	 * Login to an OAuth provider.
-	 */
-	async login(providerId: OAuthProviderId, callbacks: OAuthLoginCallbacks): Promise<void> {
-		const provider = getOAuthProvider(providerId);
-		if (!provider) {
-			throw new Error(`Unknown OAuth provider: ${providerId}`);
-		}
-
-		const credentials = await provider.login(callbacks);
-		this.set(providerId, { type: "oauth", ...credentials });
-	}
-
-	/**
-	 * Logout from a provider.
-	 */
-	logout(provider: string): void {
-		this.remove(provider);
-	}
-
-	/**
-	 * Refresh OAuth token with backend locking to prevent race conditions.
-	 * Multiple pi instances may try to refresh simultaneously when tokens expire.
-	 */
-	private async refreshOAuthTokenWithLock(
-		providerId: OAuthProviderId,
-	): Promise<{ apiKey: string; newCredentials: OAuthCredentials } | null> {
-		const provider = getOAuthProvider(providerId);
-		if (!provider) {
-			return null;
-		}
-
-		const result = await this.storage.withLockAsync(async (current) => {
-			const currentData = this.parseStorageData(current);
-			this.data = currentData;
-			this.loadError = null;
-
-			const cred = currentData[providerId];
-			if (cred?.type !== "oauth") {
-				return { result: null };
+	async modify(
+		provider: string,
+		fn: (current: Credential | undefined) => Promise<Credential | undefined>,
+	): Promise<Credential | undefined> {
+		return this.storage.withLockAsync(async (content) => {
+			const currentData = this.parseStorageData(content);
+			const next = await fn(currentData[provider]);
+			if (next === undefined) {
+				this.data = currentData;
+				return { result: currentData[provider] };
 			}
 
-			if (Date.now() < cred.expires) {
-				return { result: { apiKey: provider.getApiKey(cred), newCredentials: cred } };
-			}
-
-			const oauthCreds: Record<string, OAuthCredentials> = {};
-			for (const [key, value] of Object.entries(currentData)) {
-				if (value.type === "oauth") {
-					oauthCreds[key] = value;
-				}
-			}
-
-			const refreshed = await getOAuthApiKey(providerId, oauthCreds);
-			if (!refreshed) {
-				return { result: null };
-			}
-
-			const merged: AuthStorageData = {
-				...currentData,
-				[providerId]: { type: "oauth", ...refreshed.newCredentials },
-			};
+			const merged: AuthStorageData = { ...currentData, [provider]: next };
 			this.data = merged;
-			this.loadError = null;
-			return { result: refreshed, next: JSON.stringify(merged, null, 2) };
+			return { result: next, next: JSON.stringify(merged, null, 2) };
 		});
-
-		return result;
 	}
 
-	/**
-	 * Get API key for a provider.
-	 * Priority:
-	 * 1. Runtime override (CLI --api-key)
-	 * 2. API key from auth.json
-	 * 3. OAuth token from auth.json (auto-refreshed with locking)
-	 * 4. Environment variable
-	 */
-	async getApiKey(providerId: string, options: GetApiKeyOptions = {}): Promise<string | undefined> {
-		// Runtime override takes highest priority
-		const runtimeKey = this.runtimeOverrides.get(providerId);
-		if (runtimeKey) {
-			return runtimeKey;
-		}
-
-		const cred = this.data[providerId];
-
-		if (cred?.type === "api_key") {
-			return resolveConfigValue(cred.key, cred.env);
-		}
-
-		if (cred?.type === "oauth") {
-			const provider = getOAuthProvider(providerId);
-			if (!provider) {
-				// Unknown OAuth provider, can't get API key
-				return undefined;
-			}
-
-			// Check if token needs refresh
-			const needsRefresh = Date.now() >= cred.expires;
-
-			if (needsRefresh) {
-				// Use locked refresh to prevent race conditions
-				try {
-					const result = await this.refreshOAuthTokenWithLock(providerId);
-					if (result) {
-						return result.apiKey;
-					}
-				} catch (error) {
-					this.recordError(error);
-					// Refresh failed - re-read file to check if another instance succeeded
-					this.reload();
-					const updatedCred = this.data[providerId];
-
-					if (updatedCred?.type === "oauth" && Date.now() < updatedCred.expires) {
-						// Another instance refreshed successfully, use those credentials
-						return provider.getApiKey(updatedCred);
-					}
-
-					// Refresh truly failed - return undefined so model discovery skips this provider
-					// User can /login to re-authenticate (credentials preserved for retry)
-					return undefined;
-				}
-			} else {
-				// Token not expired, use current access token
-				return provider.getApiKey(cred);
-			}
-		}
-
-		if (options.includeFallback === false) return undefined;
-
-		// Fall back to environment variable
-		const envKey = getEnvApiKey(providerId);
-		if (envKey) return envKey;
-
-		return undefined;
+	async delete(provider: string): Promise<void> {
+		await this.storage.withLockAsync(async (content) => {
+			const currentData = this.parseStorageData(content);
+			delete currentData[provider];
+			this.data = currentData;
+			return { result: undefined, next: JSON.stringify(currentData, null, 2) };
+		});
 	}
 
-	/**
-	 * Get all registered OAuth providers
-	 */
-	getOAuthProviders() {
-		return getOAuthProviders();
+	/** List credential metadata without resolving configured key values. */
+	async list(): Promise<readonly CredentialInfo[]> {
+		return Object.entries(this.data).map(([providerId, credential]) => ({ providerId, type: credential.type }));
+	}
+}
+
+/**
+ * One-off synchronous read of a stored credential from an auth.json file,
+ * without instantiating a store or resolving configured key values.
+ */
+export function readStoredCredential(
+	providerId: string,
+	authPath: string = join(getAgentDir(), "auth.json"),
+): Credential | undefined {
+	try {
+		const data = JSON.parse(readFileSync(normalizePath(authPath), "utf-8")) as AuthStorageData;
+		return data[providerId];
+	} catch {
+		return undefined;
 	}
 }

packages/coding-agent/src/core/cache-stats.ts

@@ -29,9 +29,9 @@ export interface CacheWasteTotals {
 	missCount: number;
 }
 
-/** Minimal pricing lookup, satisfied by ModelRegistry. Cost is $/million tokens. */
+/** Minimal pricing lookup, satisfied by ModelRuntime. Cost is $/million tokens. */
 export interface ModelPriceSource {
-	find(provider: string, modelId: string): { cost: { cacheRead: number } } | undefined;
+	getModel(provider: string, modelId: string): { cost: { cacheRead: number } } | undefined;
 }
 
 /** The last request seen by the scan; everything in its prompt should be cached. */
@@ -79,7 +79,7 @@ function detectMiss(
 	const readPerToken =
 		usage.cacheRead > 0
 			? usage.cost.cacheRead / usage.cacheRead
-			: (models.find(message.provider, message.model)?.cost.cacheRead ?? 0) / 1_000_000;
+			: (models.getModel(message.provider, message.model)?.cost.cacheRead ?? 0) / 1_000_000;
 
 	return {
 		missedTokens,

packages/coding-agent/src/core/extensions/loader.ts

@@ -10,6 +10,7 @@ import { fileURLToPath } from "node:url";
 import * as _bundledPiAgentCore from "@earendil-works/pi-agent-core";
 import * as _bundledPiAiCompat from "@earendil-works/pi-ai/compat";
 import * as _bundledPiAiOauth from "@earendil-works/pi-ai/oauth";
+import * as _bundledPiAiProviders from "@earendil-works/pi-ai/providers/all";
 import type { KeyId } from "@earendil-works/pi-tui";
 import * as _bundledPiTui from "@earendil-works/pi-tui";
 import { createJiti } from "jiti/static";
@@ -58,12 +59,14 @@ const VIRTUAL_MODULES: Record<string, unknown> = {
 	"@earendil-works/pi-ai": _bundledPiAiCompat,
 	"@earendil-works/pi-ai/compat": _bundledPiAiCompat,
 	"@earendil-works/pi-ai/oauth": _bundledPiAiOauth,
+	"@earendil-works/pi-ai/providers/all": _bundledPiAiProviders,
 	"@earendil-works/pi-coding-agent": _bundledPiCodingAgent,
 	"@mariozechner/pi-agent-core": _bundledPiAgentCore,
 	"@mariozechner/pi-tui": _bundledPiTui,
 	"@mariozechner/pi-ai": _bundledPiAiCompat,
 	"@mariozechner/pi-ai/compat": _bundledPiAiCompat,
 	"@mariozechner/pi-ai/oauth": _bundledPiAiOauth,
+	"@mariozechner/pi-ai/providers/all": _bundledPiAiProviders,
 	"@mariozechner/pi-coding-agent": _bundledPiCodingAgent,
 };
 
@@ -102,20 +105,26 @@ function getAliases(): Record<string, string> {
 	// global API keep working at runtime until compat is removed.
 	const piAiCompatEntry = resolveWorkspaceOrImport("ai/dist/compat.js", "@earendil-works/pi-ai/compat");
 	const piAiOauthEntry = resolveWorkspaceOrImport("ai/dist/oauth.js", "@earendil-works/pi-ai/oauth");
+	const piAiProvidersEntry = resolveWorkspaceOrImport(
+		"ai/dist/providers/all.js",
+		"@earendil-works/pi-ai/providers/all",
+	);
 
 	_aliases = {
 		"@earendil-works/pi-coding-agent": piCodingAgentEntry,
 		"@earendil-works/pi-agent-core": piAgentCoreEntry,
 		"@earendil-works/pi-tui": piTuiEntry,
-		"@earendil-works/pi-ai": piAiCompatEntry,
+		"@earendil-works/pi-ai/providers/all": piAiProvidersEntry,
 		"@earendil-works/pi-ai/compat": piAiCompatEntry,
 		"@earendil-works/pi-ai/oauth": piAiOauthEntry,
+		"@earendil-works/pi-ai": piAiCompatEntry,
 		"@mariozechner/pi-coding-agent": piCodingAgentEntry,
 		"@mariozechner/pi-agent-core": piAgentCoreEntry,
 		"@mariozechner/pi-tui": piTuiEntry,
-		"@mariozechner/pi-ai": piAiCompatEntry,
+		"@mariozechner/pi-ai/providers/all": piAiProvidersEntry,
 		"@mariozechner/pi-ai/compat": piAiCompatEntry,
 		"@mariozechner/pi-ai/oauth": piAiOauthEntry,
+		"@mariozechner/pi-ai": piAiCompatEntry,
 		typebox: typeboxEntry,
 		"typebox/compile": typeboxCompileEntry,
 		"typebox/value": typeboxValueEntry,

packages/coding-agent/src/core/extensions/runner.ts

@@ -602,6 +602,10 @@ export class ExtensionRunner {
 		});
 	}
 
+	getModelRegistry(): ModelRegistry {
+		return this.modelRegistry;
+	}
+
 	getRegisteredCommands(): ResolvedCommand[] {
 		this.commandDiagnostics = [];
 		return this.resolveRegisteredCommands();

packages/coding-agent/src/core/extensions/types.ts

@@ -1424,14 +1424,14 @@ export interface ProviderConfig {
 	oauth?: {
 		/** Display name for the provider in login UI. */
 		name: string;
+		/** @deprecated Retained for source compatibility; canonical auth flows ignore it. */
+		usesCallbackServer?: boolean;
 		/** Run the login flow, return credentials to persist. */
 		login(callbacks: OAuthLoginCallbacks): Promise<OAuthCredentials>;
 		/** Refresh expired credentials, return updated credentials to persist. */
 		refreshToken(credentials: OAuthCredentials): Promise<OAuthCredentials>;
 		/** Convert credentials to API key string for the provider. */
 		getApiKey(credentials: OAuthCredentials): string;
-		/** Optional: modify models for this provider (e.g., update baseUrl based on credentials). */
-		modifyModels?(models: Model<Api>[], credentials: OAuthCredentials): Model<Api>[];
 	};
 }
 

packages/coding-agent/src/core/model-config.ts

@@ -0,0 +1,277 @@
+/** Immutable, credential-blind models.json snapshot. */
+
+import { readFile } from "node:fs/promises";
+import { type Static, Type } from "typebox";
+import { Compile } from "typebox/compile";
+import type { TLocalizedValidationError } from "typebox/error";
+import { stripJsonComments } from "../utils/json.ts";
+import { normalizePath } from "../utils/paths.ts";
+
+const PercentileCutoffsSchema = Type.Object({
+	p50: Type.Optional(Type.Number()),
+	p75: Type.Optional(Type.Number()),
+	p90: Type.Optional(Type.Number()),
+	p99: Type.Optional(Type.Number()),
+});
+
+const OpenRouterRoutingSchema = Type.Object({
+	allow_fallbacks: Type.Optional(Type.Boolean()),
+	require_parameters: Type.Optional(Type.Boolean()),
+	data_collection: Type.Optional(Type.Union([Type.Literal("deny"), Type.Literal("allow")])),
+	zdr: Type.Optional(Type.Boolean()),
+	enforce_distillable_text: Type.Optional(Type.Boolean()),
+	order: Type.Optional(Type.Array(Type.String())),
+	only: Type.Optional(Type.Array(Type.String())),
+	ignore: Type.Optional(Type.Array(Type.String())),
+	quantizations: Type.Optional(Type.Array(Type.String())),
+	sort: Type.Optional(
+		Type.Union([
+			Type.String(),
+			Type.Object({
+				by: Type.Optional(Type.String()),
+				partition: Type.Optional(Type.Union([Type.String(), Type.Null()])),
+			}),
+		]),
+	),
+	max_price: Type.Optional(
+		Type.Object({
+			prompt: Type.Optional(Type.Union([Type.Number(), Type.String()])),
+			completion: Type.Optional(Type.Union([Type.Number(), Type.String()])),
+			image: Type.Optional(Type.Union([Type.Number(), Type.String()])),
+			audio: Type.Optional(Type.Union([Type.Number(), Type.String()])),
+			request: Type.Optional(Type.Union([Type.Number(), Type.String()])),
+		}),
+	),
+	preferred_min_throughput: Type.Optional(Type.Union([Type.Number(), PercentileCutoffsSchema])),
+	preferred_max_latency: Type.Optional(Type.Union([Type.Number(), PercentileCutoffsSchema])),
+});
+
+const VercelGatewayRoutingSchema = Type.Object({
+	only: Type.Optional(Type.Array(Type.String())),
+	order: Type.Optional(Type.Array(Type.String())),
+});
+
+const ThinkingLevelMapValueSchema = Type.Union([Type.String(), Type.Null()]);
+const ThinkingLevelMapSchema = Type.Object({
+	off: Type.Optional(ThinkingLevelMapValueSchema),
+	minimal: Type.Optional(ThinkingLevelMapValueSchema),
+	low: Type.Optional(ThinkingLevelMapValueSchema),
+	medium: Type.Optional(ThinkingLevelMapValueSchema),
+	high: Type.Optional(ThinkingLevelMapValueSchema),
+	xhigh: Type.Optional(ThinkingLevelMapValueSchema),
+	max: Type.Optional(ThinkingLevelMapValueSchema),
+});
+
+const ChatTemplateKwargScalarSchema = Type.Union([Type.String(), Type.Number(), Type.Boolean(), Type.Null()]);
+const ChatTemplateKwargVariableSchema = Type.Object({
+	$var: Type.Union([Type.Literal("thinking.enabled"), Type.Literal("thinking.effort")]),
+	omitWhenOff: Type.Optional(Type.Boolean()),
+});
+const ChatTemplateKwargSchema = Type.Union([ChatTemplateKwargScalarSchema, ChatTemplateKwargVariableSchema]);
+
+const OpenAICompletionsCompatSchema = Type.Object({
+	supportsStore: Type.Optional(Type.Boolean()),
+	supportsDeveloperRole: Type.Optional(Type.Boolean()),
+	supportsReasoningEffort: Type.Optional(Type.Boolean()),
+	supportsUsageInStreaming: Type.Optional(Type.Boolean()),
+	maxTokensField: Type.Optional(Type.Union([Type.Literal("max_completion_tokens"), Type.Literal("max_tokens")])),
+	requiresToolResultName: Type.Optional(Type.Boolean()),
+	requiresAssistantAfterToolResult: Type.Optional(Type.Boolean()),
+	requiresThinkingAsText: Type.Optional(Type.Boolean()),
+	requiresReasoningContentOnAssistantMessages: Type.Optional(Type.Boolean()),
+	thinkingFormat: Type.Optional(
+		Type.Union([
+			Type.Literal("openai"),
+			Type.Literal("openrouter"),
+			Type.Literal("together"),
+			Type.Literal("deepseek"),
+			Type.Literal("zai"),
+			Type.Literal("qwen"),
+			Type.Literal("chat-template"),
+			Type.Literal("qwen-chat-template"),
+			Type.Literal("string-thinking"),
+			Type.Literal("ant-ling"),
+		]),
+	),
+	chatTemplateKwargs: Type.Optional(Type.Record(Type.String(), ChatTemplateKwargSchema)),
+	cacheControlFormat: Type.Optional(Type.Literal("anthropic")),
+	openRouterRouting: Type.Optional(OpenRouterRoutingSchema),
+	vercelGatewayRouting: Type.Optional(VercelGatewayRoutingSchema),
+	supportsStrictMode: Type.Optional(Type.Boolean()),
+	supportsLongCacheRetention: Type.Optional(Type.Boolean()),
+});
+
+const OpenAIResponsesCompatSchema = Type.Object({
+	supportsDeveloperRole: Type.Optional(Type.Boolean()),
+	sendSessionIdHeader: Type.Optional(Type.Boolean()),
+	supportsLongCacheRetention: Type.Optional(Type.Boolean()),
+});
+
+const AnthropicMessagesCompatSchema = Type.Object({
+	supportsEagerToolInputStreaming: Type.Optional(Type.Boolean()),
+	supportsLongCacheRetention: Type.Optional(Type.Boolean()),
+	sendSessionAffinityHeaders: Type.Optional(Type.Boolean()),
+	supportsCacheControlOnTools: Type.Optional(Type.Boolean()),
+	forceAdaptiveThinking: Type.Optional(Type.Boolean()),
+});
+
+const ProviderCompatSchema = Type.Union([
+	OpenAICompletionsCompatSchema,
+	OpenAIResponsesCompatSchema,
+	AnthropicMessagesCompatSchema,
+]);
+
+const ModelCostRatesSchema = {
+	input: Type.Number(),
+	output: Type.Number(),
+	cacheRead: Type.Number(),
+	cacheWrite: Type.Number(),
+};
+const ModelCostTierSchema = Type.Object({
+	inputTokensAbove: Type.Number(),
+	...ModelCostRatesSchema,
+});
+const ModelCostSchema = Type.Object({
+	...ModelCostRatesSchema,
+	tiers: Type.Optional(Type.Array(ModelCostTierSchema)),
+});
+
+const ModelDefinitionSchema = Type.Object({
+	id: Type.String({ minLength: 1 }),
+	name: Type.Optional(Type.String({ minLength: 1 })),
+	api: Type.Optional(Type.String({ minLength: 1 })),
+	baseUrl: Type.Optional(Type.String({ minLength: 1 })),
+	reasoning: Type.Optional(Type.Boolean()),
+	thinkingLevelMap: Type.Optional(ThinkingLevelMapSchema),
+	input: Type.Optional(Type.Array(Type.Union([Type.Literal("text"), Type.Literal("image")]))),
+	cost: Type.Optional(ModelCostSchema),
+	contextWindow: Type.Optional(Type.Number()),
+	maxTokens: Type.Optional(Type.Number()),
+	headers: Type.Optional(Type.Record(Type.String(), Type.String())),
+	compat: Type.Optional(ProviderCompatSchema),
+});
+
+const ModelOverrideSchema = Type.Object({
+	name: Type.Optional(Type.String({ minLength: 1 })),
+	reasoning: Type.Optional(Type.Boolean()),
+	thinkingLevelMap: Type.Optional(ThinkingLevelMapSchema),
+	input: Type.Optional(Type.Array(Type.Union([Type.Literal("text"), Type.Literal("image")]))),
+	cost: Type.Optional(
+		Type.Object({
+			input: Type.Optional(Type.Number()),
+			output: Type.Optional(Type.Number()),
+			cacheRead: Type.Optional(Type.Number()),
+			cacheWrite: Type.Optional(Type.Number()),
+			tiers: Type.Optional(Type.Array(ModelCostTierSchema)),
+		}),
+	),
+	contextWindow: Type.Optional(Type.Number()),
+	maxTokens: Type.Optional(Type.Number()),
+	headers: Type.Optional(Type.Record(Type.String(), Type.String())),
+	compat: Type.Optional(ProviderCompatSchema),
+});
+
+const ProviderConfigSchema = Type.Object({
+	name: Type.Optional(Type.String({ minLength: 1 })),
+	baseUrl: Type.Optional(Type.String({ minLength: 1 })),
+	apiKey: Type.Optional(Type.String({ minLength: 1 })),
+	api: Type.Optional(Type.String({ minLength: 1 })),
+	headers: Type.Optional(Type.Record(Type.String(), Type.String())),
+	compat: Type.Optional(ProviderCompatSchema),
+	authHeader: Type.Optional(Type.Boolean()),
+	models: Type.Optional(Type.Array(ModelDefinitionSchema)),
+	modelOverrides: Type.Optional(Type.Record(Type.String(), ModelOverrideSchema)),
+});
+
+const ModelsConfigSchema = Type.Object({
+	providers: Type.Record(Type.String(), ProviderConfigSchema),
+});
+const validateModelsConfig = Compile(ModelsConfigSchema);
+
+export type ModelsJsonModel = Static<typeof ModelDefinitionSchema>;
+export type ModelsJsonModelOverride = Static<typeof ModelOverrideSchema>;
+export type ModelsJsonProvider = Static<typeof ProviderConfigSchema>;
+type ModelsJson = Static<typeof ModelsConfigSchema>;
+
+function formatValidationPath(error: TLocalizedValidationError): string {
+	if (error.keyword === "required") {
+		const requiredProperties = (error.params as { requiredProperties?: string[] }).requiredProperties;
+		const requiredProperty = requiredProperties?.[0];
+		if (requiredProperty) {
+			const basePath = error.instancePath.replace(/^\//, "").replace(/\//g, ".");
+			return basePath ? `${basePath}.${requiredProperty}` : requiredProperty;
+		}
+	}
+	const path = error.instancePath.replace(/^\//, "").replace(/\//g, ".");
+	return path || "root";
+}
+
+function deepFreeze<T>(value: T): T {
+	if (typeof value !== "object" || value === null || Object.isFrozen(value)) return value;
+	for (const child of Object.values(value)) deepFreeze(child);
+	return Object.freeze(value);
+}
+
+/** One immutable load of models.json. */
+export class ModelConfig {
+	private readonly providers: ReadonlyMap<string, ModelsJsonProvider>;
+	private readonly error: string | undefined;
+
+	private constructor(providers: ReadonlyMap<string, ModelsJsonProvider>, error?: string) {
+		this.providers = providers;
+		this.error = error;
+	}
+
+	static async load(modelsJsonPath: string | undefined): Promise<ModelConfig> {
+		if (!modelsJsonPath) return new ModelConfig(new Map());
+		const path = normalizePath(modelsJsonPath);
+		let content: string;
+		try {
+			content = await readFile(path, "utf-8");
+		} catch (error) {
+			if ((error as NodeJS.ErrnoException).code === "ENOENT") return new ModelConfig(new Map());
+			return new ModelConfig(
+				new Map(),
+				`Failed to load models.json: ${error instanceof Error ? error.message : error}\n\nFile: ${path}`,
+			);
+		}
+
+		let parsed: unknown;
+		try {
+			parsed = JSON.parse(stripJsonComments(content));
+		} catch (error) {
+			return new ModelConfig(
+				new Map(),
+				`Failed to parse models.json: ${error instanceof Error ? error.message : error}\n\nFile: ${path}`,
+			);
+		}
+
+		if (!validateModelsConfig.Check(parsed)) {
+			const errors =
+				validateModelsConfig
+					.Errors(parsed)
+					.map((error) => `  - ${formatValidationPath(error)}: ${error.message}`)
+					.join("\n") || "Unknown schema error";
+			return new ModelConfig(new Map(), `Invalid models.json schema:\n${errors}\n\nFile: ${path}`);
+		}
+
+		const config = parsed as ModelsJson;
+		const providers = new Map<string, ModelsJsonProvider>();
+		for (const [providerId, provider] of Object.entries(config.providers)) {
+			providers.set(providerId, deepFreeze(structuredClone(provider)));
+		}
+		return new ModelConfig(providers);
+	}
+
+	getProvider(providerId: string): ModelsJsonProvider | undefined {
+		return this.providers.get(providerId);
+	}
+
+	getProviderIds(): readonly string[] {
+		return [...this.providers.keys()];
+	}
+
+	getError(): string | undefined {
+		return this.error;
+	}
+}

packages/coding-agent/src/core/model-resolver.ts

@@ -8,7 +8,7 @@ import chalk from "chalk";
 import { minimatch } from "minimatch";
 import { isValidThinkingLevel } from "../cli/args.ts";
 import { DEFAULT_THINKING_LEVEL } from "./defaults.ts";
-import type { ModelRegistry } from "./model-registry.ts";
+import type { ModelRuntime } from "./model-runtime.ts";
 
 /** Default model IDs for each known provider */
 export const defaultModelPerProvider: Record<KnownProvider, string> = {
@@ -268,9 +268,9 @@ export interface ResolveModelScopeResult {
 
 export async function resolveModelScopeWithDiagnostics(
 	patterns: string[],
-	modelRegistry: ModelRegistry,
+	modelRuntime: ModelRuntime,
 ): Promise<ResolveModelScopeResult> {
-	const availableModels = await modelRegistry.getAvailable();
+	const availableModels = [...(await modelRuntime.getAvailable())];
 	const scopedModels: ScopedModel[] = [];
 	const diagnostics: ModelScopeDiagnostic[] = [];
 
@@ -330,8 +330,8 @@ export async function resolveModelScopeWithDiagnostics(
 	return { scopedModels, diagnostics };
 }
 
-export async function resolveModelScope(patterns: string[], modelRegistry: ModelRegistry): Promise<ScopedModel[]> {
-	const { scopedModels, diagnostics } = await resolveModelScopeWithDiagnostics(patterns, modelRegistry);
+export async function resolveModelScope(patterns: string[], modelRuntime: ModelRuntime): Promise<ScopedModel[]> {
+	const { scopedModels, diagnostics } = await resolveModelScopeWithDiagnostics(patterns, modelRuntime);
 	for (const diagnostic of diagnostics) {
 		console.warn(chalk.yellow(`Warning: ${diagnostic.message}`));
 	}
@@ -364,9 +364,9 @@ export function resolveCliModel(options: {
 	cliProvider?: string;
 	cliModel?: string;
 	cliThinking?: ThinkingLevel;
-	modelRegistry: ModelRegistry;
+	modelRuntime: ModelRuntime;
 }): ResolveCliModelResult {
-	const { cliProvider, cliModel, cliThinking, modelRegistry } = options;
+	const { cliProvider, cliModel, cliThinking, modelRuntime } = options;
 
 	if (!cliModel) {
 		return { model: undefined, warning: undefined, error: undefined };
@@ -374,7 +374,7 @@ export function resolveCliModel(options: {
 
 	// Important: use *all* models here, not just models with pre-configured auth.
 	// This allows "--api-key" to be used for first-time setup.
-	const availableModels = modelRegistry.getAll();
+	const availableModels = [...modelRuntime.getModels()];
 	if (availableModels.length === 0) {
 		return {
 			model: undefined,
@@ -454,8 +454,8 @@ export function resolveCliModel(options: {
 			const rawExactMatches = availableModels.filter(
 				(m) => m.id.toLowerCase() === cliModel.toLowerCase() && !modelsAreEqual(m, model),
 			);
-			if (rawExactMatches.length > 0 && !modelRegistry.hasConfiguredAuth(model)) {
-				const authenticatedRawMatches = rawExactMatches.filter((m) => modelRegistry.hasConfiguredAuth(m));
+			if (rawExactMatches.length > 0 && !modelRuntime.hasConfiguredAuth(model.provider)) {
+				const authenticatedRawMatches = rawExactMatches.filter((m) => modelRuntime.hasConfiguredAuth(m.provider));
 				if (authenticatedRawMatches.length === 1) {
 					return {
 						model: authenticatedRawMatches[0],
@@ -555,7 +555,7 @@ export async function findInitialModel(options: {
 	defaultProvider?: string;
 	defaultModelId?: string;
 	defaultThinkingLevel?: ThinkingLevel;
-	modelRegistry: ModelRegistry;
+	modelRuntime: ModelRuntime;
 }): Promise<InitialModelResult> {
 	const {
 		cliProvider,
@@ -565,7 +565,7 @@ export async function findInitialModel(options: {
 		defaultProvider,
 		defaultModelId,
 		defaultThinkingLevel,
-		modelRegistry,
+		modelRuntime,
 	} = options;
 
 	let model: Model<Api> | undefined;
@@ -576,7 +576,7 @@ export async function findInitialModel(options: {
 		const resolved = resolveCliModel({
 			cliProvider,
 			cliModel,
-			modelRegistry,
+			modelRuntime,
 		});
 		if (resolved.error) {
 			console.error(chalk.red(resolved.error));
@@ -598,8 +598,8 @@ export async function findInitialModel(options: {
 
 	// 3. Try saved default from settings if auth is configured.
 	if (defaultProvider && defaultModelId) {
-		const found = modelRegistry.find(defaultProvider, defaultModelId);
-		if (found && modelRegistry.hasConfiguredAuth(found)) {
+		const found = modelRuntime.getModel(defaultProvider, defaultModelId);
+		if (found && modelRuntime.hasConfiguredAuth(found.provider)) {
 			model = found;
 			if (defaultThinkingLevel) {
 				thinkingLevel = defaultThinkingLevel;
@@ -609,7 +609,7 @@ export async function findInitialModel(options: {
 	}
 
 	// 4. Try first available model with valid API key
-	const availableModels = await modelRegistry.getAvailable();
+	const availableModels = [...(await modelRuntime.getAvailable())];
 
 	if (availableModels.length > 0) {
 		// Try to find a default model from known providers
@@ -637,12 +637,12 @@ export async function restoreModelFromSession(
 	savedModelId: string,
 	currentModel: Model<Api> | undefined,
 	shouldPrintMessages: boolean,
-	modelRegistry: ModelRegistry,
+	modelRuntime: ModelRuntime,
 ): Promise<{ model: Model<Api> | undefined; fallbackMessage: string | undefined }> {
-	const restoredModel = modelRegistry.find(savedProvider, savedModelId);
+	const restoredModel = modelRuntime.getModel(savedProvider, savedModelId);
 
 	// Check if restored model exists and still has auth configured
-	const hasConfiguredAuth = restoredModel ? modelRegistry.hasConfiguredAuth(restoredModel) : false;
+	const hasConfiguredAuth = restoredModel ? modelRuntime.hasConfiguredAuth(restoredModel.provider) : false;
 
 	if (restoredModel && hasConfiguredAuth) {
 		if (shouldPrintMessages) {
@@ -670,7 +670,7 @@ export async function restoreModelFromSession(
 	}
 
 	// Try to find any available model
-	const availableModels = await modelRegistry.getAvailable();
+	const availableModels = [...(await modelRuntime.getAvailable())];
 
 	if (availableModels.length > 0) {
 		// Try to find a default model from known providers

packages/coding-agent/src/core/model-runtime.ts

@@ -0,0 +1,489 @@
+import { join } from "node:path";
+import {
+	type Api,
+	type ApiStreamOptions,
+	type AssistantMessage,
+	type AssistantMessageEventStream,
+	type AuthCheck,
+	type AuthInteraction,
+	type AuthResult,
+	type AuthType,
+	type Context,
+	type Credential,
+	type CredentialInfo,
+	type CredentialStore,
+	createModels,
+	lazyStream,
+	type Model,
+	type Models,
+	type ModelsApiStreamOptions,
+	ModelsError,
+	type ModelsSimpleStreamOptions,
+	type ModelsStreamTransforms,
+	type MutableModels,
+	type Provider,
+	type ProviderHeaders,
+	type SimpleStreamOptions,
+	type StreamOptions,
+} from "@earendil-works/pi-ai";
+import { builtinProviders } from "@earendil-works/pi-ai/providers/all";
+import { getAgentDir } from "../config.ts";
+import { AuthStorage as DefaultAuthStorage } from "./auth-storage.ts";
+import { ModelConfig } from "./model-config.ts";
+import {
+	type AuthStatus,
+	type CompatibilityRequestConfig,
+	composeModelProvider,
+	configuredRequestAuthStatus,
+	type ProviderConfigInput,
+	resolveCompatibilityRequestConfig,
+	resolveConfiguredModelHeaders,
+	validateExtensionProvider,
+} from "./provider-composer.ts";
+import { RuntimeCredentials } from "./runtime-credentials.ts";
+
+interface ModelRuntimeSnapshot {
+	all: readonly Model<Api>[];
+	available: readonly Model<Api>[];
+	configuredProviders: ReadonlySet<string>;
+	storedProviders: ReadonlySet<string>;
+	auth: ReadonlyMap<string, AuthCheck | undefined>;
+}
+
+export interface CreateModelRuntimeOptions {
+	/** Credential storage. Defaults to the file at authPath. */
+	credentials?: CredentialStore;
+	authPath?: string;
+	modelsPath?: string | null;
+}
+
+export interface ModelRuntimeAuthOverrides {
+	apiKey?: string;
+	env?: Record<string, string>;
+}
+
+function mergeHeaders(
+	base: ProviderHeaders | undefined,
+	override: ProviderHeaders | undefined,
+): ProviderHeaders | undefined {
+	if (!base && !override) return undefined;
+	const merged = { ...base };
+	for (const [name, value] of Object.entries(override ?? {})) {
+		const lowerName = name.toLowerCase();
+		for (const existingName of Object.keys(merged)) {
+			if (existingName.toLowerCase() === lowerName) delete merged[existingName];
+		}
+		merged[name] = value;
+	}
+	return merged;
+}
+
+/** Configured pi-ai Models collection used by coding-agent and SDK consumers. */
+export class ModelRuntime implements Models {
+	private readonly models: MutableModels;
+	private readonly credentials: RuntimeCredentials;
+	private readonly builtins: ReadonlyMap<string, Provider>;
+	private readonly extensionProviders = new Map<string, ProviderConfigInput>();
+	private readonly compositionErrors = new Map<string, string>();
+	private readonly modelsPath: string | undefined;
+	private config: ModelConfig;
+	private snapshot: ModelRuntimeSnapshot = {
+		all: [],
+		available: [],
+		configuredProviders: new Set(),
+		storedProviders: new Set(),
+		auth: new Map(),
+	};
+	private availabilityRefresh: Promise<void> | undefined;
+	private availabilityError: string | undefined;
+
+	private constructor(
+		credentials: RuntimeCredentials,
+		config: ModelConfig,
+		modelsPath: string | undefined,
+		providers: readonly Provider[],
+	) {
+		this.credentials = credentials;
+		this.config = config;
+		this.modelsPath = modelsPath;
+		this.builtins = new Map(providers.map((provider) => [provider.id, provider]));
+		this.models = createModels({ credentials });
+		this.rebuildProviders();
+	}
+
+	static async create(options: CreateModelRuntimeOptions = {}): Promise<ModelRuntime> {
+		const credentials = new RuntimeCredentials(options.credentials ?? DefaultAuthStorage.create(options.authPath));
+		const modelsPath =
+			options.modelsPath === null ? undefined : (options.modelsPath ?? join(getAgentDir(), "models.json"));
+		const config = await ModelConfig.load(modelsPath);
+		const runtime = new ModelRuntime(credentials, config, modelsPath, builtinProviders());
+		await runtime.refreshAvailability();
+		return runtime;
+	}
+
+	private providerIds(): Set<string> {
+		return new Set([...this.builtins.keys(), ...this.config.getProviderIds(), ...this.extensionProviders.keys()]);
+	}
+
+	private recomposeProvider(providerId: string): void {
+		const base = this.builtins.get(providerId);
+		const extension = this.extensionProviders.get(providerId);
+		if (!base && !this.config.getProvider(providerId) && !extension) {
+			this.models.deleteProvider(providerId);
+			this.compositionErrors.delete(providerId);
+			return;
+		}
+		if (base && !this.config.getProvider(providerId) && !extension) {
+			// No overlays: use the builtin untouched so its auth/login/stream behavior is exact.
+			this.models.setProvider(base);
+			this.compositionErrors.delete(providerId);
+			return;
+		}
+		try {
+			this.models.setProvider(composeModelProvider(providerId, base, this.config, extension));
+			this.compositionErrors.delete(providerId);
+		} catch (error) {
+			this.compositionErrors.set(providerId, error instanceof Error ? error.message : String(error));
+			if (base) this.models.setProvider(base);
+			else this.models.deleteProvider(providerId);
+		}
+	}
+
+	private rebuildProviders(): void {
+		this.models.clearProviders();
+		this.compositionErrors.clear();
+		for (const providerId of this.providerIds()) this.recomposeProvider(providerId);
+		this.updateModelSnapshot();
+	}
+
+	private updateModelSnapshot(): void {
+		const all = [...this.models.getModels()];
+		this.snapshot = {
+			...this.snapshot,
+			all,
+			available: all.filter((model) => this.snapshot.configuredProviders.has(model.provider)),
+		};
+	}
+
+	private async runAvailabilityRefresh(): Promise<void> {
+		const providers = this.models.getProviders();
+		const [available, checks, credentials] = await Promise.all([
+			this.models.getAvailable(),
+			Promise.all(
+				providers.map(
+					async (provider): Promise<[string, AuthCheck | undefined]> => [
+						provider.id,
+						await this.models.checkAuth(provider.id),
+					],
+				),
+			),
+			this.credentials.list(),
+		]);
+		const auth = new Map(checks);
+		const configuredProviders = new Set(
+			checks
+				.filter((entry): entry is [string, AuthCheck] => entry[1] !== undefined)
+				.map(([providerId]) => providerId),
+		);
+		this.snapshot = {
+			all: [...this.models.getModels()],
+			available: [...available],
+			configuredProviders,
+			storedProviders: new Set(credentials.map((entry) => entry.providerId)),
+			auth,
+		};
+		this.availabilityError = undefined;
+	}
+
+	private queueAvailabilityRefresh(after: Promise<void> | undefined): Promise<void> {
+		const refresh = (after ?? Promise.resolve()).catch(() => {}).then(() => this.runAvailabilityRefresh());
+		const recorded = refresh.catch((error) => {
+			this.availabilityError = error instanceof Error ? error.message : String(error);
+			throw error;
+		});
+		const tracked = recorded.finally(() => {
+			if (this.availabilityRefresh === tracked) this.availabilityRefresh = undefined;
+		});
+		this.availabilityRefresh = tracked;
+		return tracked;
+	}
+
+	/** Coalesce concurrent readers onto the pending refresh. */
+	private refreshAvailability(): Promise<void> {
+		return this.availabilityRefresh ?? this.queueAvailabilityRefresh(undefined);
+	}
+
+	/** Mutations must not observe an in-flight refresh started before them. */
+	private forceRefreshAvailability(): Promise<void> {
+		return this.queueAvailabilityRefresh(this.availabilityRefresh);
+	}
+
+	getProviders(): readonly Provider[] {
+		return this.models.getProviders();
+	}
+
+	getProvider(providerId: string): Provider | undefined {
+		return this.models.getProvider(providerId);
+	}
+
+	getModels(providerId?: string): readonly Model<Api>[] {
+		return this.models.getModels(providerId);
+	}
+
+	getModel(providerId: string, modelId: string): Model<Api> | undefined {
+		return this.models.getModel(providerId, modelId);
+	}
+
+	async checkAuth(providerId: string): Promise<AuthCheck | undefined> {
+		return this.models.checkAuth(providerId);
+	}
+
+	async getAvailable(providerId?: string): Promise<readonly Model<Api>[]> {
+		if (providerId) {
+			if (this.availabilityRefresh) {
+				await this.availabilityRefresh;
+				return this.snapshot.available.filter((model) => model.provider === providerId);
+			}
+			try {
+				return await this.models.getAvailable(providerId);
+			} catch (error) {
+				this.availabilityError = error instanceof Error ? error.message : String(error);
+				throw error;
+			}
+		}
+		await this.refreshAvailability();
+		return this.snapshot.available;
+	}
+
+	getAvailableSnapshot(): readonly Model<Api>[] {
+		return this.snapshot.available;
+	}
+
+	getError(): string | undefined {
+		const errors: string[] = [];
+		const configError = this.config.getError();
+		if (configError) errors.push(configError);
+		for (const [providerId, error] of this.compositionErrors) {
+			errors.push(`Provider "${providerId}": ${error}`);
+		}
+		if (this.availabilityError) errors.push(`Availability refresh: ${this.availabilityError}`);
+		return errors.length > 0 ? errors.join("\n\n") : undefined;
+	}
+
+	getRegisteredProviderConfig(providerId: string): ProviderConfigInput | undefined {
+		return this.extensionProviders.get(providerId);
+	}
+
+	getRegisteredProviderIds(): readonly string[] {
+		return [...this.extensionProviders.keys()];
+	}
+
+	/** @internal Compatibility fallback for ModelRegistry when provider auth is unconfigured. */
+	getCompatibilityRequestConfig(model: Model<Api>): CompatibilityRequestConfig {
+		return resolveCompatibilityRequestConfig(
+			model,
+			this.config.getProvider(model.provider),
+			this.extensionProviders.get(model.provider),
+		);
+	}
+
+	isUsingOAuth(providerId: string): boolean {
+		return this.snapshot.auth.get(providerId)?.type === "oauth";
+	}
+
+	hasConfiguredAuth(providerId: string): boolean {
+		return this.snapshot.configuredProviders.has(providerId);
+	}
+
+	getAuth(providerId: string, overrides?: ModelRuntimeAuthOverrides): Promise<AuthResult | undefined>;
+	getAuth(model: Model<Api>, overrides?: ModelRuntimeAuthOverrides): Promise<AuthResult | undefined>;
+	async getAuth(
+		providerOrModel: string | Model<Api>,
+		overrides: ModelRuntimeAuthOverrides = {},
+	): Promise<AuthResult | undefined> {
+		if (typeof providerOrModel === "string") return this.models.getAuth(providerOrModel, overrides);
+		const resolution = await this.models.getAuth(providerOrModel, overrides);
+		if (!resolution) return undefined;
+		const configuredHeaders = resolveConfiguredModelHeaders(
+			providerOrModel,
+			this.config.getProvider(providerOrModel.provider),
+			this.extensionProviders.get(providerOrModel.provider),
+			{ ...(resolution.env ?? {}), ...(overrides.env ?? {}) },
+		);
+		return {
+			...resolution,
+			auth: {
+				...resolution.auth,
+				headers: mergeHeaders(resolution.auth.headers, configuredHeaders),
+			},
+		};
+	}
+
+	setRuntimeApiKey(providerId: string, apiKey: string): void {
+		this.credentials.setRuntimeApiKey(providerId, apiKey);
+		const auth = new Map(this.snapshot.auth).set(providerId, { type: "api_key", source: "runtime API key" });
+		const configuredProviders = new Set(this.snapshot.configuredProviders).add(providerId);
+		const storedProviders = new Set(this.snapshot.storedProviders).add(providerId);
+		this.snapshot = {
+			...this.snapshot,
+			auth,
+			configuredProviders,
+			storedProviders,
+			available: this.snapshot.all.filter((model) => configuredProviders.has(model.provider)),
+		};
+		void this.forceRefreshAvailability().catch(() => {});
+	}
+
+	removeRuntimeApiKey(providerId: string): void {
+		this.credentials.removeRuntimeApiKey(providerId);
+		void this.forceRefreshAvailability().catch(() => {});
+	}
+
+	listCredentials(): Promise<readonly CredentialInfo[]> {
+		return this.credentials.list();
+	}
+
+	getProviderAuthStatus(providerId: string): AuthStatus {
+		if (this.credentials.hasRuntimeApiKey(providerId)) return { configured: true, source: "runtime" };
+		if (this.snapshot.storedProviders.has(providerId)) return { configured: true, source: "stored" };
+		const configured = configuredRequestAuthStatus(
+			this.config.getProvider(providerId),
+			this.extensionProviders.get(providerId),
+		);
+		if (configured) return configured;
+		const check = this.snapshot.auth.get(providerId);
+		return check ? { configured: true, source: "environment", label: check.source } : { configured: false };
+	}
+
+	private async prepareRequest(
+		model: Model<Api>,
+		options: (StreamOptions & ModelsStreamTransforms) | undefined,
+	): Promise<{ provider: Provider; model: Model<Api>; options: StreamOptions }> {
+		const provider = this.models.getProvider(model.provider);
+		if (!provider) throw new ModelsError("provider", `Unknown provider: ${model.provider}`);
+		const resolution = await this.getAuth(model, { apiKey: options?.apiKey, env: options?.env });
+		if (!resolution) throw new ModelsError("auth", `Provider is not configured: ${model.provider}`);
+
+		const { transformHeaders, ...providerOptions } = options ?? {};
+		let headers = mergeHeaders(resolution.auth.headers, providerOptions.headers);
+		if (transformHeaders) headers = await transformHeaders(headers ?? {});
+		const env =
+			resolution.env || providerOptions.env
+				? { ...(resolution.env ?? {}), ...(providerOptions.env ?? {}) }
+				: undefined;
+		return {
+			provider,
+			model: resolution.auth.baseUrl ? { ...model, baseUrl: resolution.auth.baseUrl } : model,
+			options: {
+				...providerOptions,
+				apiKey: providerOptions.apiKey ?? resolution.auth.apiKey,
+				headers,
+				env,
+			},
+		};
+	}
+
+	stream<TApi extends Api>(
+		model: Model<TApi>,
+		context: Context,
+		options?: ModelsApiStreamOptions<TApi>,
+	): AssistantMessageEventStream {
+		return lazyStream(model, async () => {
+			const prepared = await this.prepareRequest(
+				model,
+				options as (StreamOptions & ModelsStreamTransforms) | undefined,
+			);
+			return prepared.provider.stream(
+				prepared.model as Model<TApi>,
+				context,
+				prepared.options as ApiStreamOptions<TApi>,
+			);
+		});
+	}
+
+	complete<TApi extends Api>(
+		model: Model<TApi>,
+		context: Context,
+		options?: ModelsApiStreamOptions<TApi>,
+	): Promise<AssistantMessage> {
+		return this.stream(model, context, options).result();
+	}
+
+	streamSimple(model: Model<Api>, context: Context, options?: ModelsSimpleStreamOptions): AssistantMessageEventStream {
+		return lazyStream(model, async () => {
+			const prepared = await this.prepareRequest(model, options);
+			return prepared.provider.streamSimple(prepared.model, context, prepared.options as SimpleStreamOptions);
+		});
+	}
+
+	completeSimple(model: Model<Api>, context: Context, options?: ModelsSimpleStreamOptions): Promise<AssistantMessage> {
+		return this.streamSimple(model, context, options).result();
+	}
+
+	async login(providerId: string, type: AuthType, interaction: AuthInteraction): Promise<Credential> {
+		const credential = await this.models.login(providerId, type, interaction);
+		await this.forceRefreshAvailability();
+		return credential;
+	}
+
+	async logout(providerId: string): Promise<void> {
+		await this.models.logout(providerId);
+		await this.forceRefreshAvailability();
+	}
+
+	async reloadConfig(): Promise<void> {
+		this.config = await ModelConfig.load(this.modelsPath);
+		this.rebuildProviders();
+		await this.forceRefreshAvailability();
+	}
+
+	async refresh(providerId?: string): Promise<void> {
+		await this.models.refresh(providerId);
+		this.updateModelSnapshot();
+		await this.forceRefreshAvailability();
+	}
+
+	registerProvider(providerId: string, config: ProviderConfigInput): void {
+		// Validate the incoming registration on its own, like the legacy registry:
+		// a broken re-registration must throw without touching the stored config.
+		validateExtensionProvider(providerId, this.builtins.get(providerId), this.config.getProvider(providerId), config);
+		// Re-registration merges defined values over the previous registration and
+		// preserves undefined ones, matching the legacy ModelRegistry contract.
+		const previous = this.extensionProviders.get(providerId);
+		const effective: ProviderConfigInput = { ...previous };
+		for (const [key, value] of Object.entries(config)) {
+			if (value !== undefined) (effective as Record<string, unknown>)[key] = value;
+		}
+		this.extensionProviders.set(providerId, effective);
+		this.recomposeProvider(providerId);
+		this.updateModelSnapshot();
+		if (
+			this.snapshot.storedProviders.has(providerId) ||
+			configuredRequestAuthStatus(this.config.getProvider(providerId), effective)?.configured
+		) {
+			const configuredProviders = new Set(this.snapshot.configuredProviders).add(providerId);
+			const auth = new Map(this.snapshot.auth);
+			// Provisional entry until the async refresh lands; never clobber a real check result.
+			if (!auth.get(providerId)) {
+				auth.set(providerId, {
+					type: effective.oauth && !effective.apiKey ? "oauth" : "api_key",
+					source: "configured provider",
+				});
+			}
+			this.snapshot = {
+				...this.snapshot,
+				auth,
+				configuredProviders,
+				available: this.snapshot.all.filter((model) => configuredProviders.has(model.provider)),
+			};
+		}
+		void this.forceRefreshAvailability().catch(() => {});
+	}
+
+	unregisterProvider(providerId: string): void {
+		this.extensionProviders.delete(providerId);
+		this.recomposeProvider(providerId);
+		this.updateModelSnapshot();
+		void this.forceRefreshAvailability().catch(() => {});
+	}
+}

packages/coding-agent/src/core/provider-composer.ts

@@ -0,0 +1,513 @@
+import {
+	type Api,
+	type ApiKeyAuth,
+	type AssistantMessageEventStream,
+	type AuthContext,
+	type AuthInteraction,
+	type AuthResult,
+	type Context,
+	type Credential,
+	lazyStream,
+	type Model,
+	type ModelAuth,
+	type OAuthAuth,
+	type OAuthCredentials,
+	type OAuthLoginCallbacks,
+	type Provider,
+	type ProviderHeaders,
+	type SimpleStreamOptions,
+	type StreamOptions,
+} from "@earendil-works/pi-ai";
+import { getApiProvider } from "@earendil-works/pi-ai/compat";
+import type { ModelConfig, ModelsJsonModel, ModelsJsonModelOverride, ModelsJsonProvider } from "./model-config.ts";
+import {
+	clearConfigValueCache,
+	getConfigValueEnvVarNames,
+	isCommandConfigValue,
+	isConfigValueConfigured,
+	resolveConfigValueOrThrow,
+	resolveHeadersOrThrow,
+} from "./resolve-config-value.ts";
+
+export interface ExtensionOAuthConfig {
+	name: string;
+	/** @deprecated Retained for extension source compatibility; ignored by canonical auth flows. */
+	usesCallbackServer?: boolean;
+	login(callbacks: OAuthLoginCallbacks): Promise<OAuthCredentials>;
+	refreshToken(credentials: OAuthCredentials): Promise<OAuthCredentials>;
+	getApiKey(credentials: OAuthCredentials): string;
+}
+
+/** Input type for the extension registerProvider API. */
+export interface ProviderConfigInput {
+	name?: string;
+	baseUrl?: string;
+	apiKey?: string;
+	api?: Api;
+	streamSimple?: (model: Model<Api>, context: Context, options?: SimpleStreamOptions) => AssistantMessageEventStream;
+	headers?: Record<string, string>;
+	authHeader?: boolean;
+	oauth?: ExtensionOAuthConfig;
+	models?: Array<{
+		id: string;
+		name: string;
+		api?: Api;
+		baseUrl?: string;
+		reasoning: boolean;
+		thinkingLevelMap?: Model<Api>["thinkingLevelMap"];
+		input: ("text" | "image")[];
+		cost: Model<Api>["cost"];
+		contextWindow: number;
+		maxTokens: number;
+		headers?: Record<string, string>;
+		compat?: Model<Api>["compat"];
+	}>;
+}
+
+export type AuthStatus = {
+	configured: boolean;
+	source?: "stored" | "runtime" | "environment" | "fallback" | "models_json_key" | "models_json_command";
+	label?: string;
+};
+
+export const clearApiKeyCache = clearConfigValueCache;
+
+function mergeCompat(
+	base: Model<Api>["compat"],
+	override: Model<Api>["compat"] | ModelsJsonModelOverride["compat"],
+): Model<Api>["compat"] {
+	if (!override) return base;
+	const merged = { ...base, ...override } as NonNullable<Model<Api>["compat"]>;
+	const baseNested = base as Record<string, unknown> | undefined;
+	const overrideNested = override as Record<string, unknown>;
+	const mergedNested = merged as Record<string, unknown>;
+	for (const key of ["openRouterRouting", "vercelGatewayRouting", "chatTemplateKwargs"] as const) {
+		const baseValue = baseNested?.[key];
+		const overrideValue = overrideNested[key];
+		if (
+			(typeof baseValue === "object" && baseValue !== null) ||
+			(typeof overrideValue === "object" && overrideValue !== null)
+		) {
+			mergedNested[key] = { ...(baseValue as object | undefined), ...(overrideValue as object | undefined) };
+		}
+	}
+	return merged;
+}
+
+function applyModelOverride(model: Model<Api>, override: ModelsJsonModelOverride): Model<Api> {
+	return {
+		...model,
+		name: override.name ?? model.name,
+		reasoning: override.reasoning ?? model.reasoning,
+		thinkingLevelMap: override.thinkingLevelMap
+			? { ...model.thinkingLevelMap, ...override.thinkingLevelMap }
+			: model.thinkingLevelMap,
+		input: (override.input as ("text" | "image")[] | undefined) ?? model.input,
+		cost: override.cost
+			? {
+					input: override.cost.input ?? model.cost.input,
+					output: override.cost.output ?? model.cost.output,
+					cacheRead: override.cost.cacheRead ?? model.cost.cacheRead,
+					cacheWrite: override.cost.cacheWrite ?? model.cost.cacheWrite,
+					tiers: override.cost.tiers ?? model.cost.tiers,
+				}
+			: model.cost,
+		contextWindow: override.contextWindow ?? model.contextWindow,
+		maxTokens: override.maxTokens ?? model.maxTokens,
+		compat: mergeCompat(model.compat, override.compat),
+	};
+}
+
+function modelFromJson(
+	providerId: string,
+	definition: ModelsJsonModel,
+	providerConfig: ModelsJsonProvider,
+	defaults: Model<Api> | undefined,
+): Model<Api> {
+	const api = definition.api ?? providerConfig.api ?? defaults?.api;
+	if (!api) {
+		throw new Error(
+			`Provider ${providerId}, model ${definition.id}: no "api" specified. Set at provider or model level.`,
+		);
+	}
+	const baseUrl = definition.baseUrl ?? providerConfig.baseUrl ?? defaults?.baseUrl;
+	if (!baseUrl) throw new Error(`Provider ${providerId}: "baseUrl" is required when defining custom models.`);
+	if (definition.contextWindow !== undefined && definition.contextWindow <= 0) {
+		throw new Error(`Provider ${providerId}, model ${definition.id}: invalid contextWindow`);
+	}
+	if (definition.maxTokens !== undefined && definition.maxTokens <= 0) {
+		throw new Error(`Provider ${providerId}, model ${definition.id}: invalid maxTokens`);
+	}
+	return {
+		id: definition.id,
+		name: definition.name ?? definition.id,
+		api: api as Api,
+		provider: providerId,
+		baseUrl,
+		reasoning: definition.reasoning ?? false,
+		thinkingLevelMap: definition.thinkingLevelMap,
+		input: (definition.input ?? ["text"]) as ("text" | "image")[],
+		cost: definition.cost ?? { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+		contextWindow: definition.contextWindow ?? 128000,
+		maxTokens: definition.maxTokens ?? 16384,
+		headers: undefined,
+		compat: mergeCompat(providerConfig.compat, definition.compat),
+	};
+}
+
+function applyModelsJson(
+	providerId: string,
+	baseModels: readonly Model<Api>[],
+	config: ModelsJsonProvider | undefined,
+): Model<Api>[] {
+	if (!config) return [...baseModels];
+	const hasOverrides = config.modelOverrides && Object.keys(config.modelOverrides).length > 0;
+	if (
+		!config.models?.length &&
+		!config.baseUrl &&
+		!config.headers &&
+		!config.compat &&
+		!hasOverrides &&
+		!config.apiKey &&
+		config.authHeader === undefined
+	) {
+		throw new Error(
+			`Provider ${providerId}: must specify "baseUrl", "headers", "compat", "modelOverrides", or "models".`,
+		);
+	}
+
+	const models: Model<Api>[] = baseModels.map((model) => ({
+		...model,
+		baseUrl: config.baseUrl ?? model.baseUrl,
+		compat: mergeCompat(model.compat, config.compat),
+	}));
+	for (const definition of config.models ?? []) {
+		const existingIndex = models.findIndex((model) => model.id === definition.id);
+		const defaults = existingIndex >= 0 ? models[existingIndex] : models[0];
+		const model = modelFromJson(providerId, definition, config, defaults);
+		if (existingIndex >= 0) models[existingIndex] = model;
+		else models.push(model);
+	}
+	return models;
+}
+
+function applyExtension(
+	providerId: string,
+	models: readonly Model<Api>[],
+	config: ProviderConfigInput | undefined,
+): Model<Api>[] {
+	if (!config) return [...models];
+	if (!config.models) {
+		return config.baseUrl ? models.map((model) => ({ ...model, baseUrl: config.baseUrl! })) : [...models];
+	}
+	return config.models.map((definition) => {
+		const defaults = models.find((model) => model.id === definition.id) ?? models[0];
+		const api = definition.api ?? config.api ?? defaults?.api;
+		if (!api) {
+			throw new Error(
+				`Provider ${providerId}, model ${definition.id}: no "api" specified. Set at provider or model level.`,
+			);
+		}
+		const baseUrl = definition.baseUrl ?? config.baseUrl ?? defaults?.baseUrl;
+		if (!baseUrl) throw new Error(`Provider ${providerId}: "baseUrl" is required when defining custom models.`);
+		return {
+			...definition,
+			api,
+			provider: providerId,
+			baseUrl,
+			headers: undefined,
+		};
+	});
+}
+
+function adaptOAuth(config: ExtensionOAuthConfig): OAuthAuth {
+	return {
+		name: config.name,
+		login: async (callbacks) => {
+			const credential = await config.login({
+				onAuth: (info) => callbacks.notify({ type: "auth_url", ...info }),
+				onDeviceCode: (info) => callbacks.notify({ type: "device_code", ...info }),
+				onPrompt: (prompt) => callbacks.prompt({ type: "text", ...prompt }),
+				onProgress: (message) => callbacks.notify({ type: "progress", message }),
+				onManualCodeInput: () => callbacks.prompt({ type: "manual_code", message: "Paste the authorization code" }),
+				onSelect: (prompt) => callbacks.prompt({ type: "select", ...prompt }),
+				signal: callbacks.signal,
+			});
+			return { ...credential, type: "oauth" };
+		},
+		refresh: async (credential) => ({ ...(await config.refreshToken(credential)), type: "oauth" }),
+		toAuth: async (credential) => ({ apiKey: config.getApiKey(credential) }),
+	};
+}
+
+function withConfiguredAuth(
+	auth: ModelAuth,
+	headers: Record<string, string> | undefined,
+	authHeader: boolean,
+): ModelAuth {
+	let mergedHeaders: ProviderHeaders | undefined =
+		auth.headers || headers ? { ...auth.headers, ...headers } : undefined;
+	if (authHeader) {
+		if (!auth.apiKey) throw new Error("authHeader requires a resolved API key");
+		mergedHeaders = { ...mergedHeaders, Authorization: `Bearer ${auth.apiKey}` };
+	}
+	return { ...auth, headers: mergedHeaders };
+}
+
+function configuredApiKey(
+	config: ModelsJsonProvider | undefined,
+	extension: ProviderConfigInput | undefined,
+): string | undefined {
+	return extension?.apiKey ?? config?.apiKey;
+}
+
+function configuredHeaders(
+	config: ModelsJsonProvider | undefined,
+	extension: ProviderConfigInput | undefined,
+): Record<string, string> | undefined {
+	if (!config?.headers && !extension?.headers) return undefined;
+	return { ...config?.headers, ...extension?.headers };
+}
+
+async function configContextEnv(
+	values: readonly string[],
+	ctx: AuthContext,
+	explicit?: Record<string, string>,
+): Promise<Record<string, string> | undefined> {
+	const env = { ...explicit };
+	for (const name of new Set(values.flatMap(getConfigValueEnvVarNames))) {
+		if (env[name] !== undefined) continue;
+		const value = await ctx.env(name);
+		if (value !== undefined) env[name] = value;
+	}
+	return Object.keys(env).length > 0 ? env : undefined;
+}
+
+function composeApiKeyAuth(
+	providerId: string,
+	base: Provider | undefined,
+	config: ModelsJsonProvider | undefined,
+	extension: ProviderConfigInput | undefined,
+): ApiKeyAuth | undefined {
+	const inherited = base?.auth.apiKey;
+	const rawKey = configuredApiKey(config, extension);
+	const oauth = extension?.oauth ?? base?.auth.oauth;
+	// OAuth-only providers get no fabricated API-key login method.
+	if (!inherited && rawKey === undefined && oauth) return undefined;
+	const rawHeaders = configuredHeaders(config, extension);
+	const authHeader = extension?.authHeader ?? config?.authHeader ?? false;
+	return {
+		name: inherited?.name ?? "API key",
+		login:
+			inherited?.login ??
+			(async (interaction: AuthInteraction) => ({
+				type: "api_key",
+				key: await interaction.prompt({ type: "secret", message: "Enter API key" }),
+			})),
+		check: async (input) => {
+			if (input.credential) {
+				if (inherited?.check) return inherited.check(input);
+				if (input.credential.key) return { type: "api_key", source: "stored credential" };
+				const resolved = await inherited?.resolve(input);
+				return resolved ? { type: "api_key", source: resolved.source } : undefined;
+			}
+			if (rawKey !== undefined) {
+				if (isCommandConfigValue(rawKey)) return { type: "api_key", source: "configured API key" };
+				const envNames = getConfigValueEnvVarNames(rawKey);
+				for (const name of envNames) {
+					if ((await input.ctx.env(name)) === undefined) return undefined;
+				}
+				return { type: "api_key", source: "configured API key" };
+			}
+			if (inherited?.check) return inherited.check(input);
+			const resolved = await inherited?.resolve(input);
+			return resolved ? { type: "api_key", source: resolved.source } : undefined;
+		},
+		resolve: async (input) => {
+			let result: AuthResult | undefined;
+			if (input.credential) {
+				result = inherited
+					? await inherited.resolve(input)
+					: input.credential.key
+						? { auth: { apiKey: input.credential.key }, env: input.credential.env, source: "stored credential" }
+						: undefined;
+			} else if (rawKey !== undefined) {
+				const env = await configContextEnv([rawKey], input.ctx);
+				const key = resolveConfigValueOrThrow(rawKey, `API key for provider "${providerId}"`, env);
+				result = inherited
+					? await inherited.resolve({ ...input, credential: { type: "api_key", key } })
+					: { auth: { apiKey: key }, source: "configured API key" };
+			} else {
+				result = await inherited?.resolve(input);
+			}
+			if (!result) return undefined;
+			const explicitEnv = { ...(input.credential?.env ?? {}), ...(result.env ?? {}) };
+			const headerEnv = await configContextEnv(Object.values(rawHeaders ?? {}), input.ctx, explicitEnv);
+			const headers = resolveHeadersOrThrow(rawHeaders, `provider "${providerId}"`, headerEnv);
+			return { ...result, auth: withConfiguredAuth(result.auth, headers, authHeader) };
+		},
+	};
+}
+
+function composeOAuthAuth(
+	providerId: string,
+	base: Provider | undefined,
+	config: ModelsJsonProvider | undefined,
+	extension: ProviderConfigInput | undefined,
+): OAuthAuth | undefined {
+	const oauth = extension?.oauth ? adaptOAuth(extension.oauth) : base?.auth.oauth;
+	if (!oauth) return undefined;
+	const rawHeaders = configuredHeaders(config, extension);
+	const authHeader = extension?.authHeader ?? config?.authHeader ?? false;
+	return {
+		...oauth,
+		toAuth: async (credential) => {
+			const auth = await oauth.toAuth(credential);
+			const env = credential.env;
+			const headers = resolveHeadersOrThrow(
+				rawHeaders,
+				`provider "${providerId}"`,
+				typeof env === "object" && env !== null ? (env as Record<string, string>) : undefined,
+			);
+			return withConfiguredAuth(auth, headers, authHeader);
+		},
+	};
+}
+
+function rawModelHeaders(
+	model: Model<Api>,
+	config: ModelsJsonProvider | undefined,
+	extension: ProviderConfigInput | undefined,
+): Record<string, string> | undefined {
+	const definition = config?.models?.find((entry) => entry.id === model.id);
+	const extensionModel = extension?.models?.find((entry) => entry.id === model.id);
+	const headers = {
+		...config?.modelOverrides?.[model.id]?.headers,
+		...definition?.headers,
+		...extensionModel?.headers,
+	};
+	return Object.keys(headers).length > 0 ? headers : undefined;
+}
+
+export function validateExtensionProvider(
+	providerId: string,
+	base: Provider | undefined,
+	modelsConfig: ModelsJsonProvider | undefined,
+	extension: ProviderConfigInput,
+): void {
+	if (extension.streamSimple && !extension.api) {
+		throw new Error(`Provider ${providerId}: "api" is required when registering streamSimple.`);
+	}
+	applyExtension(providerId, applyModelsJson(providerId, base?.getModels() ?? [], modelsConfig), extension);
+}
+
+/** Compose built-in, models.json, and extension layers without reading credentials. */
+export function composeModelProvider(
+	providerId: string,
+	base: Provider | undefined,
+	modelConfig: ModelConfig,
+	extension: ProviderConfigInput | undefined,
+): Provider {
+	const config = modelConfig.getProvider(providerId);
+	// models.json modelOverrides are the topmost user-config layer: they apply once,
+	// after custom-model upserts and extension model replacement.
+	const getModels = () =>
+		applyExtension(providerId, applyModelsJson(providerId, base?.getModels() ?? [], config), extension).map(
+			(model) => {
+				const override = config?.modelOverrides?.[model.id];
+				return override ? applyModelOverride(model, override) : model;
+			},
+		);
+	// Validate eagerly so registration/reload reports structural errors immediately.
+	getModels();
+	const apiKey = composeApiKeyAuth(providerId, base, config, extension);
+	const oauth = composeOAuthAuth(providerId, base, config, extension);
+	if (!apiKey && !oauth) throw new Error(`Provider ${providerId}: no authentication method configured.`);
+
+	const supportsBaseApi = (model: Model<Api>) => base?.getModels().some((entry) => entry.api === model.api) ?? false;
+	const streamWith = (
+		model: Model<Api>,
+		context: Context,
+		options: StreamOptions | undefined,
+		simple: boolean,
+	): AssistantMessageEventStream =>
+		lazyStream(model, async () => {
+			if (extension?.streamSimple && model.api === extension.api) {
+				return extension.streamSimple(model, context, options as SimpleStreamOptions);
+			}
+			if (base && supportsBaseApi(model)) {
+				return simple
+					? base.streamSimple(model, context, options as SimpleStreamOptions)
+					: base.stream(model, context, options);
+			}
+			const api = getApiProvider(model.api);
+			if (!api) throw new Error(`No API provider registered for api: ${model.api}`);
+			return simple
+				? api.streamSimple(model, context, options as SimpleStreamOptions)
+				: api.stream(model, context, options);
+		});
+
+	return {
+		id: providerId,
+		name: extension?.name ?? config?.name ?? base?.name ?? extension?.oauth?.name ?? providerId,
+		baseUrl: extension?.baseUrl ?? config?.baseUrl ?? base?.baseUrl,
+		headers: base?.headers,
+		auth: { ...(apiKey ? { apiKey } : {}), ...(oauth ? { oauth } : {}) },
+		getModels,
+		refreshModels: base?.refreshModels ? () => base.refreshModels!() : undefined,
+		filterModels: base?.filterModels
+			? (models, credential: Credential | undefined) => base.filterModels!(models, credential)
+			: undefined,
+		stream: (model, context, options) => streamWith(model, context, options, false),
+		streamSimple: (model, context, options) => streamWith(model, context, options, true),
+	};
+}
+
+export function resolveConfiguredModelHeaders(
+	model: Model<Api>,
+	config: ModelsJsonProvider | undefined,
+	extension: ProviderConfigInput | undefined,
+	env?: Record<string, string>,
+): Record<string, string> | undefined {
+	return resolveHeadersOrThrow(
+		rawModelHeaders(model, config, extension),
+		`model "${model.provider}/${model.id}"`,
+		env,
+	);
+}
+
+export interface CompatibilityRequestConfig {
+	headers?: ProviderHeaders;
+	authHeader: boolean;
+}
+
+export function resolveCompatibilityRequestConfig(
+	model: Model<Api>,
+	config: ModelsJsonProvider | undefined,
+	extension: ProviderConfigInput | undefined,
+): CompatibilityRequestConfig {
+	const configured = resolveHeadersOrThrow(
+		{ ...configuredHeaders(config, extension), ...rawModelHeaders(model, config, extension) },
+		`model "${model.provider}/${model.id}"`,
+	);
+	return {
+		headers: model.headers || configured ? { ...model.headers, ...configured } : undefined,
+		authHeader: extension?.authHeader ?? config?.authHeader ?? false,
+	};
+}
+
+export function configuredRequestAuthStatus(
+	config: ModelsJsonProvider | undefined,
+	extension: ProviderConfigInput | undefined,
+): AuthStatus | undefined {
+	const value = configuredApiKey(config, extension);
+	if (value === undefined) return undefined;
+	if (isCommandConfigValue(value)) return { configured: true, source: "models_json_command" };
+	const names = getConfigValueEnvVarNames(value);
+	if (names.length > 0) {
+		return isConfigValueConfigured(value)
+			? { configured: true, source: "environment", label: names.join(", ") }
+			: { configured: false };
+	}
+	return { configured: true, source: extension?.apiKey !== undefined ? "fallback" : "models_json_key" };
+}

packages/coding-agent/src/core/provider-display-names.ts

@@ -1,35 +0,0 @@
-export const BUILT_IN_PROVIDER_DISPLAY_NAMES: Record<string, string> = {
-	anthropic: "Anthropic",
-	"amazon-bedrock": "Amazon Bedrock",
-	"ant-ling": "Ant Ling",
-	"azure-openai-responses": "Azure OpenAI Responses",
-	cerebras: "Cerebras",
-	"cloudflare-ai-gateway": "Cloudflare AI Gateway",
-	"cloudflare-workers-ai": "Cloudflare Workers AI",
-	deepseek: "DeepSeek",
-	fireworks: "Fireworks",
-	google: "Google Gemini",
-	"google-vertex": "Google Vertex AI",
-	groq: "Groq",
-	huggingface: "Hugging Face",
-	"kimi-coding": "Kimi For Coding",
-	mistral: "Mistral",
-	minimax: "MiniMax",
-	"minimax-cn": "MiniMax (China)",
-	moonshotai: "Moonshot AI",
-	"moonshotai-cn": "Moonshot AI (China)",
-	nvidia: "NVIDIA NIM",
-	opencode: "OpenCode Zen",
-	"opencode-go": "OpenCode Go",
-	openai: "OpenAI",
-	openrouter: "OpenRouter",
-	together: "Together AI",
-	"vercel-ai-gateway": "Vercel AI Gateway",
-	xai: "xAI",
-	zai: "ZAI Coding Plan (Global)",
-	"zai-coding-cn": "ZAI Coding Plan (China)",
-	xiaomi: "Xiaomi MiMo",
-	"xiaomi-token-plan-cn": "Xiaomi MiMo Token Plan (China)",
-	"xiaomi-token-plan-ams": "Xiaomi MiMo Token Plan (Amsterdam)",
-	"xiaomi-token-plan-sgp": "Xiaomi MiMo Token Plan (Singapore)",
-};

packages/coding-agent/src/core/runtime-credentials.ts

@@ -0,0 +1,48 @@
+import type { Credential, CredentialInfo, CredentialStore } from "@earendil-works/pi-ai";
+
+/** Async credential store overlay for non-persistent runtime API keys. */
+export class RuntimeCredentials implements CredentialStore {
+	private readonly store: CredentialStore;
+	private readonly overrides = new Map<string, string>();
+
+	constructor(store: CredentialStore) {
+		this.store = store;
+	}
+
+	setRuntimeApiKey(providerId: string, apiKey: string): void {
+		this.overrides.set(providerId, apiKey);
+	}
+
+	removeRuntimeApiKey(providerId: string): void {
+		this.overrides.delete(providerId);
+	}
+
+	hasRuntimeApiKey(providerId: string): boolean {
+		return this.overrides.has(providerId);
+	}
+
+	async read(providerId: string): Promise<Credential | undefined> {
+		const override = this.overrides.get(providerId);
+		return override ? { type: "api_key", key: override } : this.store.read(providerId);
+	}
+
+	async list(): Promise<readonly CredentialInfo[]> {
+		const entries = new Map((await this.store.list()).map((entry) => [entry.providerId, entry]));
+		for (const providerId of this.overrides.keys()) {
+			entries.set(providerId, { providerId, type: "api_key" });
+		}
+		return [...entries.values()];
+	}
+
+	modify(
+		providerId: string,
+		fn: (current: Credential | undefined) => Promise<Credential | undefined>,
+	): Promise<Credential | undefined> {
+		return this.store.modify(providerId, fn);
+	}
+
+	async delete(providerId: string): Promise<void> {
+		this.overrides.delete(providerId);
+		await this.store.delete(providerId);
+	}
+}

packages/coding-agent/src/core/sdk.ts

@@ -1,16 +1,15 @@
 import { join } from "node:path";
 import { Agent, type AgentMessage, type ThinkingLevel } from "@earendil-works/pi-agent-core";
-import { clampThinkingLevel, type Message, type Model, streamSimple } from "@earendil-works/pi-ai/compat";
+import { clampThinkingLevel, type Message, type Model } from "@earendil-works/pi-ai/compat";
 import { getAgentDir } from "../config.ts";
 import { resolvePath } from "../utils/paths.ts";
 import { AgentSession } from "./agent-session.ts";
 import { formatNoModelsAvailableMessage } from "./auth-guidance.ts";
-import { AuthStorage } from "./auth-storage.ts";
 import { DEFAULT_THINKING_LEVEL } from "./defaults.ts";
 import type { ExtensionRunner, LoadExtensionsResult, SessionStartEvent, ToolDefinition } from "./extensions/index.ts";
 import { convertToLlm } from "./messages.ts";
-import { ModelRegistry } from "./model-registry.ts";
 import { findInitialModel } from "./model-resolver.ts";
+import { ModelRuntime } from "./model-runtime.ts";
 import { mergeProviderAttributionHeaders } from "./provider-attribution.ts";
 import type { ResourceLoader } from "./resource-loader.ts";
 import { DefaultResourceLoader } from "./resource-loader.ts";
@@ -37,10 +36,8 @@ export interface CreateAgentSessionOptions {
 	/** Global config directory. Default: ~/.pi/agent */
 	agentDir?: string;
 
-	/** Auth storage for credentials. Default: AuthStorage.create(agentDir/auth.json) */
-	authStorage?: AuthStorage;
-	/** Model registry. Default: ModelRegistry.create(authStorage, agentDir/models.json) */
-	modelRegistry?: ModelRegistry;
+	/** Canonical model/auth runtime. Defaults to a runtime using agentDir/auth.json and models.json. */
+	modelRuntime?: ModelRuntime;
 
 	/** Model to use. Default: from settings, else first available */
 	model?: Model<any>;
@@ -169,11 +166,9 @@ export async function createAgentSession(options: CreateAgentSessionOptions = {}
 	const agentDir = options.agentDir ? resolvePath(options.agentDir) : getDefaultAgentDir();
 	let resourceLoader = options.resourceLoader;
 
-	// Use provided or create AuthStorage and ModelRegistry
 	const authPath = options.agentDir ? join(agentDir, "auth.json") : undefined;
 	const modelsPath = options.agentDir ? join(agentDir, "models.json") : undefined;
-	const authStorage = options.authStorage ?? AuthStorage.create(authPath);
-	const modelRegistry = options.modelRegistry ?? ModelRegistry.create(authStorage, modelsPath);
+	const modelRuntime = options.modelRuntime ?? (await ModelRuntime.create({ authPath, modelsPath }));
 
 	const settingsManager = options.settingsManager ?? SettingsManager.create(cwd, agentDir);
 	const sessionManager = options.sessionManager ?? SessionManager.create(cwd, getDefaultSessionDir(cwd, agentDir));
@@ -194,8 +189,8 @@ export async function createAgentSession(options: CreateAgentSessionOptions = {}
 
 	// If session has data, try to restore model from it
 	if (!model && hasExistingSession && existingSession.model) {
-		const restoredModel = modelRegistry.find(existingSession.model.provider, existingSession.model.modelId);
-		if (restoredModel && modelRegistry.hasConfiguredAuth(restoredModel)) {
+		const restoredModel = modelRuntime.getModel(existingSession.model.provider, existingSession.model.modelId);
+		if (restoredModel && modelRuntime.hasConfiguredAuth(restoredModel.provider)) {
 			model = restoredModel;
 		}
 		if (!model) {
@@ -211,7 +206,7 @@ export async function createAgentSession(options: CreateAgentSessionOptions = {}
 			defaultProvider: settingsManager.getDefaultProvider(),
 			defaultModelId: settingsManager.getDefaultModel(),
 			defaultThinkingLevel: settingsManager.getDefaultThinkingLevel(),
-			modelRegistry,
+			modelRuntime,
 		});
 		model = result.model;
 		if (!model) {
@@ -300,11 +295,6 @@ export async function createAgentSession(options: CreateAgentSessionOptions = {}
 		},
 		convertToLlm: convertToLlmWithBlockImages,
 		streamFn: async (model, context, options) => {
-			const auth = await modelRegistry.getApiKeyAndHeaders(model);
-			if (!auth.ok) {
-				throw new Error(auth.error);
-			}
-			const env = auth.env || options?.env ? { ...(auth.env ?? {}), ...(options?.env ?? {}) } : undefined;
 			const providerRetrySettings = settingsManager.getProviderRetrySettings();
 			const httpIdleTimeoutMs = settingsManager.getHttpIdleTimeoutMs();
 			// SDKs treat timeout=0 as 0ms (immediate timeout), not "no timeout".
@@ -313,28 +303,24 @@ export async function createAgentSession(options: CreateAgentSessionOptions = {}
 			const timeoutMs = options?.timeoutMs ?? providerRetrySettings.timeoutMs ?? effectiveTimeoutMs;
 			const websocketConnectTimeoutMs =
 				options?.websocketConnectTimeoutMs ?? settingsManager.getWebSocketConnectTimeoutMs();
-			let headers = mergeProviderAttributionHeaders(
-				model,
-				settingsManager,
-				options?.sessionId,
-				auth.headers,
-				options?.headers,
-			);
-			// Let extensions inject/adjust per-request headers (e.g. tracing, session correlation)
-			// after static assembly, before the provider HTTP call.
 			const headerRunner = extensionRunnerRef.current;
-			if (headerRunner?.hasHandlers("before_provider_headers")) {
-				headers = await headerRunner.emitBeforeProviderHeaders(headers ?? {});
-			}
-			return streamSimple(model, context, {
+			return modelRuntime.streamSimple(model, context, {
 				...options,
-				apiKey: auth.apiKey,
-				env,
 				timeoutMs,
 				websocketConnectTimeoutMs,
 				maxRetries: options?.maxRetries ?? providerRetrySettings.maxRetries,
 				maxRetryDelayMs: options?.maxRetryDelayMs ?? providerRetrySettings.maxRetryDelayMs,
-				headers,
+				transformHeaders: async (requestHeaders) => {
+					const headers = mergeProviderAttributionHeaders(
+						model,
+						settingsManager,
+						options?.sessionId,
+						requestHeaders,
+					);
+					return headerRunner?.hasHandlers("before_provider_headers")
+						? headerRunner.emitBeforeProviderHeaders(headers ?? {})
+						: (headers ?? {});
+				},
 			});
 		},
 		onPayload: async (payload, _model) => {
@@ -390,7 +376,7 @@ export async function createAgentSession(options: CreateAgentSessionOptions = {}
 		scopedModels: options.scopedModels,
 		resourceLoader,
 		customTools: options.customTools,
-		modelRegistry,
+		modelRuntime,
 		initialActiveToolNames,
 		allowedToolNames,
 		excludedToolNames,

packages/coding-agent/src/index.ts

@@ -23,17 +23,7 @@ export {
 	parseSkillBlock,
 	type SessionStats,
 } from "./core/agent-session.ts";
-// Auth and model registry
-export {
-	type ApiKeyCredential,
-	type AuthCredential,
-	type AuthStatus,
-	AuthStorage,
-	type AuthStorageBackend,
-	FileAuthStorageBackend,
-	InMemoryAuthStorageBackend,
-	type OAuthCredential,
-} from "./core/auth-storage.ts";
+export { readStoredCredential } from "./core/auth-storage.ts";
 // Compaction
 export {
 	type BranchPreparation,
@@ -178,6 +168,11 @@ export {
 	resolveModelScopeWithDiagnostics,
 	type ScopedModel,
 } from "./core/model-resolver.ts";
+export {
+	type CreateModelRuntimeOptions,
+	ModelRuntime,
+	type ModelRuntimeAuthOverrides,
+} from "./core/model-runtime.ts";
 export type {
 	PackageManager,
 	PathMetadata,

packages/coding-agent/src/main.ts

@@ -23,12 +23,11 @@ import {
 	createAgentSessionServices,
 } from "./core/agent-session-services.ts";
 import { formatNoModelsAvailableMessage } from "./core/auth-guidance.ts";
-import { AuthStorage } from "./core/auth-storage.ts";
 import { exportFromFile } from "./core/export-html/index.ts";
 import type { InlineExtension } from "./core/extensions/types.ts";
 import { applyHttpProxySettings, configureHttpDispatcher } from "./core/http-dispatcher.ts";
-import type { ModelRegistry } from "./core/model-registry.ts";
 import { resolveCliModel, resolveModelScope, type ScopedModel } from "./core/model-resolver.ts";
+import type { ModelRuntime } from "./core/model-runtime.ts";
 import { restoreStdout, takeOverStdout } from "./core/output-guard.ts";
 import { type AppMode, resolveProjectTrusted } from "./core/project-trust.ts";
 import type { CreateAgentSessionOptions } from "./core/sdk.ts";
@@ -358,7 +357,7 @@ function buildSessionOptions(
 	parsed: Args,
 	scopedModels: ScopedModel[],
 	hasExistingSession: boolean,
-	modelRegistry: ModelRegistry,
+	modelRuntime: ModelRuntime,
 	settingsManager: SettingsManager,
 ): {
 	options: CreateAgentSessionOptions;
@@ -377,7 +376,7 @@ function buildSessionOptions(
 			cliProvider: parsed.provider,
 			cliModel: parsed.model,
 			cliThinking: parsed.thinking,
-			modelRegistry,
+			modelRuntime,
 		});
 		if (resolved.warning) {
 			diagnostics.push({ type: "warning", message: resolved.warning });
@@ -400,7 +399,7 @@ function buildSessionOptions(
 		// Check if saved default is in scoped models - use it if so, otherwise first scoped model
 		const savedProvider = settingsManager.getDefaultProvider();
 		const savedModelId = settingsManager.getDefaultModel();
-		const savedModel = savedProvider && savedModelId ? modelRegistry.find(savedProvider, savedModelId) : undefined;
+		const savedModel = savedProvider && savedModelId ? modelRuntime.getModel(savedProvider, savedModelId) : undefined;
 		const savedInScope = savedModel ? scopedModels.find((sm) => modelsAreEqual(sm.model, savedModel)) : undefined;
 
 		if (savedInScope) {
@@ -433,7 +432,7 @@ function buildSessionOptions(
 		}));
 	}
 
-	// API key from CLI - set in authStorage
+	// API key from CLI - set as a non-persistent runtime override
 	// (handled by caller before createAgentSession)
 
 	// Tools
@@ -611,7 +610,6 @@ export async function main(args: string[], options?: MainOptions) {
 	const resolvedSkillPaths = resolveCliPaths(cwd, parsed.skills);
 	const resolvedPromptTemplatePaths = resolveCliPaths(cwd, parsed.promptTemplates);
 	const resolvedThemePaths = resolveCliPaths(cwd, parsed.themes);
-	const authStorage = AuthStorage.create();
 	const createRuntime: CreateAgentSessionRuntimeFactory = async ({
 		cwd,
 		agentDir,
@@ -634,7 +632,6 @@ export async function main(args: string[], options?: MainOptions) {
 		const services = await createAgentSessionServices({
 			cwd,
 			agentDir,
-			authStorage,
 			settingsManager: runtimeSettingsManager,
 			extensionFlagValues: parsed.unknownFlags,
 			resourceLoaderReloadOptions: shouldResolveProjectTrust
@@ -676,7 +673,7 @@ export async function main(args: string[], options?: MainOptions) {
 				extensionFactories: options?.extensionFactories,
 			},
 		});
-		const { settingsManager, modelRegistry, resourceLoader } = services;
+		const { settingsManager, modelRuntime, resourceLoader } = services;
 		const diagnostics: AgentSessionRuntimeDiagnostic[] = [
 			...projectTrustDiagnostics,
 			...services.diagnostics,
@@ -689,7 +686,7 @@ export async function main(args: string[], options?: MainOptions) {
 
 		const modelPatterns = parsed.models ?? settingsManager.getEnabledModels();
 		const scopedModels =
-			modelPatterns && modelPatterns.length > 0 ? await resolveModelScope(modelPatterns, modelRegistry) : [];
+			modelPatterns && modelPatterns.length > 0 ? await resolveModelScope(modelPatterns, modelRuntime) : [];
 		const {
 			options: sessionOptions,
 			cliThinkingFromModel,
@@ -698,7 +695,7 @@ export async function main(args: string[], options?: MainOptions) {
 			parsed,
 			scopedModels,
 			sessionManager.buildSessionContext().messages.length > 0,
-			modelRegistry,
+			modelRuntime,
 			settingsManager,
 		);
 		diagnostics.push(...sessionOptionDiagnostics);
@@ -710,7 +707,8 @@ export async function main(args: string[], options?: MainOptions) {
 					message: "--api-key requires a model to be specified via --model, --provider/--model, or --models",
 				});
 			} else {
-				authStorage.setRuntimeApiKey(sessionOptions.model.provider, parsed.apiKey);
+				modelRuntime.setRuntimeApiKey(sessionOptions.model.provider, parsed.apiKey);
+				await services.modelRuntime.getAvailable();
 			}
 		}
 
@@ -745,7 +743,7 @@ export async function main(args: string[], options?: MainOptions) {
 	});
 	time("createAgentSessionRuntime");
 	const { services, session, modelFallbackMessage } = runtime;
-	const { settingsManager, modelRegistry, resourceLoader } = services;
+	const { settingsManager, modelRuntime, resourceLoader } = services;
 	applyHttpProxySettings(settingsManager.getGlobalSettings().httpProxy);
 	configureHttpDispatcher(settingsManager.getHttpIdleTimeoutMs());
 
@@ -759,7 +757,7 @@ export async function main(args: string[], options?: MainOptions) {
 
 	if (parsed.listModels !== undefined) {
 		const searchPattern = typeof parsed.listModels === "string" ? parsed.listModels : undefined;
-		await listModels(modelRegistry, searchPattern);
+		await listModels(modelRuntime, searchPattern);
 		process.exit(0);
 	}
 

packages/coding-agent/src/modes/interactive/components/footer.ts

@@ -138,7 +138,7 @@ export class FooterComponent implements Component {
 			statsParts.push(`CH${latestCacheHitRate.toFixed(1)}%`);
 		}
 		// Show cost with "(sub)" indicator if using OAuth subscription
-		const usingSubscription = state.model ? this.session.modelRegistry.isUsingOAuth(state.model) : false;
+		const usingSubscription = state.model ? this.session.modelRuntime.isUsingOAuth(state.model.provider) : false;
 		if (totalCost || usingSubscription) {
 			const costStr = `$${totalCost.toFixed(3)}${usingSubscription ? " (sub)" : ""}`;
 			statsParts.push(costStr);

packages/coding-agent/src/modes/interactive/components/login-dialog.ts

@@ -1,4 +1,4 @@
-import { getOAuthProviders, type OAuthDeviceCodeInfo } from "@earendil-works/pi-ai/oauth";
+import type { AuthInfoLink, OAuthDeviceCodeInfo } from "@earendil-works/pi-ai";
 import { Container, type Focusable, getKeybindings, Input, Spacer, Text, type TUI } from "@earendil-works/pi-tui";
 import { openBrowser } from "../../../utils/open-browser.ts";
 import { theme } from "../theme/theme.ts";
@@ -38,8 +38,7 @@ export class LoginDialogComponent extends Container implements Focusable {
 		this.tui = tui;
 		this.onComplete = onComplete;
 
-		const providerInfo = getOAuthProviders().find((p) => p.id === providerId);
-		const providerName = providerNameOverride || providerInfo?.name || providerId;
+		const providerName = providerNameOverride || providerId;
 		const title = titleOverride ?? `Login to ${providerName}`;
 
 		// Top border
@@ -176,17 +175,19 @@ export class LoginDialogComponent extends Container implements Focusable {
 		});
 	}
 
-	/**
-	 * Show informational text without prompting for input.
-	 */
-	showInfo(lines: string[]): void {
-		this.contentContainer.clear();
+	/** Show provider-owned information and links without starting an auth callback flow. */
+	showInfo(message: string, links: readonly AuthInfoLink[] = [], showCloseHint = false): void {
 		this.contentContainer.addChild(new Spacer(1));
-		for (const line of lines) {
-			this.contentContainer.addChild(new Text(line, 1, 0));
+		this.contentContainer.addChild(new Text(theme.fg("text", message), 1, 0));
+		for (const link of links) {
+			const text = link.label ? `${link.label}: ${link.url}` : link.url;
+			const hyperlink = `\x1b]8;;${link.url}\x07${text}\x1b]8;;\x07`;
+			this.contentContainer.addChild(new Text(theme.fg("accent", hyperlink), 1, 0));
+		}
+		if (showCloseHint) {
+			this.contentContainer.addChild(new Spacer(1));
+			this.contentContainer.addChild(new Text(`(${keyHint("tui.select.cancel", "to close")})`, 1, 0));
 		}
-		this.contentContainer.addChild(new Spacer(1));
-		this.contentContainer.addChild(new Text(`(${keyHint("tui.select.cancel", "to close")})`, 1, 0));
 		this.tui.requestRender();
 	}
 

packages/coding-agent/src/modes/interactive/components/model-selector.ts

@@ -9,7 +9,7 @@ import {
 	Text,
 	type TUI,
 } from "@earendil-works/pi-tui";
-import type { ModelRegistry } from "../../../core/model-registry.ts";
+import type { ModelRuntime } from "../../../core/model-runtime.ts";
 import type { SettingsManager } from "../../../core/settings-manager.ts";
 import { getModelSelectorSearchText } from "../model-search.ts";
 import { theme } from "../theme/theme.ts";
@@ -52,7 +52,7 @@ export class ModelSelectorComponent extends Container implements Focusable {
 	private selectedIndex: number = 0;
 	private currentModel?: Model<any>;
 	private settingsManager: SettingsManager;
-	private modelRegistry: ModelRegistry;
+	private modelRuntime: ModelRuntime;
 	private onSelectCallback: (model: Model<any>) => void;
 	private onCancelCallback: () => void;
 	private errorMessage?: string;
@@ -66,7 +66,7 @@ export class ModelSelectorComponent extends Container implements Focusable {
 		tui: TUI,
 		currentModel: Model<any> | undefined,
 		settingsManager: SettingsManager,
-		modelRegistry: ModelRegistry,
+		modelRuntime: ModelRuntime,
 		scopedModels: ReadonlyArray<ScopedModelItem>,
 		onSelect: (model: Model<any>) => void,
 		onCancel: () => void,
@@ -77,7 +77,7 @@ export class ModelSelectorComponent extends Container implements Focusable {
 		this.tui = tui;
 		this.currentModel = currentModel;
 		this.settingsManager = settingsManager;
-		this.modelRegistry = modelRegistry;
+		this.modelRuntime = modelRuntime;
 		this.scopedModels = scopedModels;
 		this.scope = scopedModels.length > 0 ? "scoped" : "all";
 		this.onSelectCallback = onSelect;
@@ -139,17 +139,17 @@ export class ModelSelectorComponent extends Container implements Focusable {
 		let models: ModelItem[];
 
 		// Refresh to pick up any changes to models.json
-		this.modelRegistry.refresh();
+		await this.modelRuntime.refresh();
 
 		// Check for models.json errors
-		const loadError = this.modelRegistry.getError();
+		const loadError = this.modelRuntime.getError();
 		if (loadError) {
 			this.errorMessage = loadError;
 		}
 
 		// Load available models (built-in models still work even if models.json failed)
 		try {
-			const availableModels = await this.modelRegistry.getAvailable();
+			const availableModels = await this.modelRuntime.getAvailable();
 			models = availableModels.map((model: Model<any>) => ({
 				provider: model.provider,
 				id: model.id,
@@ -166,7 +166,7 @@ export class ModelSelectorComponent extends Container implements Focusable {
 
 		this.allModels = this.sortModels(models);
 		this.scopedModels = this.scopedModels.map((scoped) => {
-			const refreshed = this.modelRegistry.find(scoped.model.provider, scoped.model.id);
+			const refreshed = this.modelRuntime.getModel(scoped.model.provider, scoped.model.id);
 			return refreshed ? { ...scoped, model: refreshed } : scoped;
 		});
 		this.scopedModelItems = this.scopedModels.map((scoped) => ({

packages/coding-agent/src/modes/interactive/components/oauth-selector.ts

@@ -1,3 +1,4 @@
+import type { ApiKeyAuth, AuthCheck, OAuthAuth } from "@earendil-works/pi-ai";
 import {
 	Container,
 	type Focusable,
@@ -7,7 +8,6 @@ import {
 	Spacer,
 	TruncatedText,
 } from "@earendil-works/pi-tui";
-import type { AuthStatus, AuthStorage } from "../../../core/auth-storage.ts";
 import { theme } from "../theme/theme.ts";
 import { DynamicBorder } from "./dynamic-border.ts";
 
@@ -15,6 +15,8 @@ export type AuthSelectorProvider = {
 	id: string;
 	name: string;
 	authType: "oauth" | "api_key";
+	method?: ApiKeyAuth | OAuthAuth;
+	status?: AuthCheck;
 };
 
 export function formatAuthSelectorProviderType(authType: AuthSelectorProvider["authType"]): string {
@@ -42,26 +44,20 @@ export class OAuthSelectorComponent extends Container implements Focusable {
 	private filteredProviders: AuthSelectorProvider[];
 	private selectedIndex: number = 0;
 	private mode: "login" | "logout";
-	private authStorage: AuthStorage;
-	private getAuthStatus: (providerId: string) => AuthStatus;
 	private onSelectCallback: (providerId: string, authType: AuthSelectorProvider["authType"]) => void;
 	private onCancelCallback: () => void;
 	private showAuthTypeLabels: boolean;
 
 	constructor(
 		mode: "login" | "logout",
-		authStorage: AuthStorage,
 		providers: AuthSelectorProvider[],
 		onSelect: (providerId: string, authType: AuthSelectorProvider["authType"]) => void,
 		onCancel: () => void,
-		getAuthStatus?: (providerId: string) => AuthStatus,
 		initialSearchInput?: string,
 	) {
 		super();
 
 		this.mode = mode;
-		this.authStorage = authStorage;
-		this.getAuthStatus = getAuthStatus ?? ((providerId) => this.authStorage.getAuthStatus(providerId));
 		this.allProviders = providers;
 		this.filteredProviders = providers;
 		this.showAuthTypeLabels = new Set(providers.map((provider) => provider.authType)).size > 1;
@@ -105,7 +101,11 @@ export class OAuthSelectorComponent extends Container implements Focusable {
 
 	private filterProviders(query: string): void {
 		this.filteredProviders = query
-			? fuzzyFilter(this.allProviders, query, (provider) => `${provider.name} ${provider.id} ${provider.authType}`)
+			? fuzzyFilter(
+					this.allProviders,
+					query,
+					(provider) => `${provider.name} ${provider.id} ${provider.authType} ${provider.method?.name ?? ""}`,
+				)
 			: this.allProviders;
 		this.selectedIndex = Math.max(0, Math.min(this.selectedIndex, Math.max(0, this.filteredProviders.length - 1)));
 		this.updateList();
@@ -162,29 +162,22 @@ export class OAuthSelectorComponent extends Container implements Focusable {
 	}
 
 	private formatStatusIndicator(provider: AuthSelectorProvider): string {
-		const credential = this.authStorage.get(provider.id);
-		if (credential?.type === provider.authType) return theme.fg("success", " ✓ configured");
-		if (credential) {
-			const label = credential.type === "oauth" ? "subscription configured" : "API key configured";
+		if (!provider.status) return theme.fg("muted", " • unconfigured");
+		if (provider.status.type !== provider.authType) {
+			const label = provider.status.type === "oauth" ? "subscription configured" : "API key configured";
 			return theme.fg("muted", " • ") + theme.fg("warning", label);
 		}
-		if (provider.authType !== "api_key") return theme.fg("muted", " • unconfigured");
-
-		const status = this.getAuthStatus(provider.id);
-		switch (status.source) {
-			case "environment":
-				return theme.fg("success", ` ✓ env: ${status.label ?? "API key"}`);
-			case "runtime":
-				return theme.fg("success", " ✓ runtime API key");
-			case "fallback":
-				return theme.fg("success", " ✓ custom API key");
-			case "models_json_key":
-				return theme.fg("success", " ✓ key in models.json");
-			case "models_json_command":
-				return theme.fg("success", " ✓ command in models.json");
-			default:
-				return theme.fg("muted", " • unconfigured");
+		if (
+			!provider.status.source ||
+			provider.status.source === "OAuth" ||
+			provider.status.source === "stored credential"
+		) {
+			return theme.fg("success", " ✓ configured");
 		}
+		const source = /^[A-Z][A-Z0-9_]*(?:, [A-Z][A-Z0-9_]*)*$/.test(provider.status.source)
+			? `env: ${provider.status.source}`
+			: provider.status.source;
+		return theme.fg("success", ` ✓ ${source}`);
 	}
 
 	handleInput(keyData: string): void {

packages/coding-agent/src/modes/interactive/interactive-mode.ts

@@ -8,15 +8,8 @@ import * as fs from "node:fs";
 import * as os from "node:os";
 import * as path from "node:path";
 import type { AgentMessage } from "@earendil-works/pi-agent-core";
-import {
-	type AssistantMessage,
-	getProviders,
-	type ImageContent,
-	type Message,
-	type Model,
-	type OAuthProviderId,
-	type OAuthSelectPrompt,
-} from "@earendil-works/pi-ai/compat";
+import type { AuthEvent, AuthPrompt } from "@earendil-works/pi-ai";
+import type { AssistantMessage, ImageContent, Message, Model } from "@earendil-works/pi-ai/compat";
 import type {
 	AutocompleteItem,
 	AutocompleteProvider,
@@ -54,7 +47,6 @@ import {
 	getAgentDir,
 	getAuthPath,
 	getDebugLogPath,
-	getDocsPath,
 	getShareViewerUrl,
 	VERSION,
 } from "../../config.ts";
@@ -85,7 +77,6 @@ import { type AppKeybinding, KeybindingsManager } from "../../core/keybindings.t
 import { createCompactionSummaryMessage } from "../../core/messages.ts";
 import { defaultModelPerProvider, findExactModelReferenceMatch, resolveModelScope } from "../../core/model-resolver.ts";
 import { DefaultPackageManager } from "../../core/package-manager.ts";
-import { BUILT_IN_PROVIDER_DISPLAY_NAMES } from "../../core/provider-display-names.ts";
 import type { ResourceDiagnostic } from "../../core/resource-loader.ts";
 import { formatMissingSessionCwdPrompt, MissingSessionCwdError } from "../../core/session-cwd.ts";
 import { type SessionEntry, SessionManager, sessionEntryToContextMessages } from "../../core/session-manager.ts";
@@ -212,7 +203,7 @@ function isDeadTerminalError(error: unknown): boolean {
 }
 
 const ANTHROPIC_SUBSCRIPTION_AUTH_WARNING =
-	"Anthropic subscription auth is active. Third-party harness usage draws from extra usage and is billed per token, not your Claude plan limits. Manage extra usage at https://claude.ai/settings/usage.";
+	"Anthropic subscription auth is active. Third-party harness usage draws from extra usage and is billed per token, not your Claude plan limits. Manage extra usage at https://claude.ai/settings/usage. Disable this warning in /settings.";
 
 function isAnthropicSubscriptionAuthKey(apiKey: string | undefined): boolean {
 	return typeof apiKey === "string" && apiKey.startsWith("sk-ant-oat");
@@ -248,24 +239,6 @@ function hasDefaultModelProvider(providerId: string): providerId is keyof typeof
 	return providerId in defaultModelPerProvider;
 }
 
-const BEDROCK_PROVIDER_ID = "amazon-bedrock";
-
-const BUILT_IN_MODEL_PROVIDERS = new Set<string>(getProviders());
-
-export function isApiKeyLoginProvider(
-	providerId: string,
-	oauthProviderIds: ReadonlySet<string>,
-	builtInProviderIds: ReadonlySet<string> = BUILT_IN_MODEL_PROVIDERS,
-): boolean {
-	if (BUILT_IN_PROVIDER_DISPLAY_NAMES[providerId]) {
-		return true;
-	}
-	if (builtInProviderIds.has(providerId)) {
-		return false;
-	}
-	return !oauthProviderIds.has(providerId);
-}
-
 type LoginProviderCompletionOption = {
 	id: string;
 	name: string;
@@ -571,12 +544,12 @@ export class InteractiveMode {
 
 		const modelCommand = slashCommands.find((command) => command.name === "model");
 		if (modelCommand) {
-			modelCommand.getArgumentCompletions = (prefix: string): AutocompleteItem[] | null => {
+			modelCommand.getArgumentCompletions = async (prefix: string): Promise<AutocompleteItem[] | null> => {
 				// Get available models (scoped or from registry)
 				const models =
 					this.session.scopedModels.length > 0
 						? this.session.scopedModels.map((s) => s.model)
-						: this.session.modelRegistry.getAvailable();
+						: await this.session.modelRuntime.getAvailable();
 
 				if (models.length === 0) return null;
 
@@ -879,7 +852,7 @@ export class InteractiveMode {
 			this.showWarning(`Migrated credentials to auth.json: ${migratedProviders.join(", ")}`);
 		}
 
-		const modelsJsonError = this.session.modelRegistry.getError();
+		const modelsJsonError = this.session.modelRuntime.getError();
 		if (modelsJsonError) {
 			this.showError(`models.json error: ${modelsJsonError}`);
 		}
@@ -1779,7 +1752,7 @@ export class InteractiveMode {
 			hasUI: true,
 			cwd: this.sessionManager.getCwd(),
 			sessionManager: this.sessionManager,
-			modelRegistry: this.session.modelRegistry,
+			modelRegistry: extensionRunner.getModelRegistry(),
 			model: this.session.model,
 			isIdle: () => this.session.isIdle,
 			isProjectTrusted: () => this.settingsManager.isProjectTrusted(),
@@ -3288,7 +3261,7 @@ export class InteractiveMode {
 		// Cache-miss notices are not persisted; re-derive them from the full entry
 		// list and re-inject them after the assistant messages that paid for them.
 		const cacheMisses = this.settingsManager.getShowCacheMissNotices()
-			? collectCacheMisses(this.sessionManager.getEntries(), this.session.modelRegistry)
+			? collectCacheMisses(this.sessionManager.getEntries(), this.session.modelRuntime)
 			: new Map<AssistantMessage, CacheMiss>();
 
 		if (options.updateFooter) {
@@ -3392,7 +3365,7 @@ export class InteractiveMode {
 		if (!this.settingsManager.getShowCacheMissNotices()) return;
 
 		// Entries don't contain `message` yet: message_end fires before persistence.
-		const miss = detectCacheMiss(this.sessionManager.getEntries(), message, this.session.modelRegistry);
+		const miss = detectCacheMiss(this.sessionManager.getEntries(), message, this.session.modelRuntime);
 		if (miss) this.addCacheMissNotice(miss);
 	}
 
@@ -4325,9 +4298,9 @@ export class InteractiveMode {
 			return this.session.scopedModels.map((scoped) => scoped.model);
 		}
 
-		this.session.modelRegistry.refresh();
 		try {
-			return await this.session.modelRegistry.getAvailable();
+			await this.session.modelRuntime.refresh();
+			return [...(await this.session.modelRuntime.getAvailable())];
 		} catch {
 			return [];
 		}
@@ -4353,15 +4326,13 @@ export class InteractiveMode {
 			return;
 		}
 
-		const storedCredential = this.session.modelRegistry.authStorage.get("anthropic");
-		if (storedCredential?.type === "oauth") {
-			this.anthropicSubscriptionWarningShown = true;
-			this.showWarning(ANTHROPIC_SUBSCRIPTION_AUTH_WARNING);
-			return;
-		}
-
 		try {
-			const apiKey = await this.session.modelRegistry.getApiKeyForProvider(model.provider);
+			if ((await this.session.modelRuntime.checkAuth("anthropic"))?.type === "oauth") {
+				this.anthropicSubscriptionWarningShown = true;
+				this.showWarning(ANTHROPIC_SUBSCRIPTION_AUTH_WARNING);
+				return;
+			}
+			const apiKey = (await this.session.modelRuntime.getAuth(model.provider))?.auth.apiKey;
 			if (!isAnthropicSubscriptionAuthKey(apiKey)) {
 				return;
 			}
@@ -4429,7 +4400,7 @@ export class InteractiveMode {
 				this.ui,
 				this.session.model,
 				this.settingsManager,
-				this.session.modelRegistry,
+				this.session.modelRuntime,
 				this.session.scopedModels,
 				async (model) => {
 					try {
@@ -4457,8 +4428,8 @@ export class InteractiveMode {
 
 	private async showModelsSelector(): Promise<void> {
 		// Get all available models
-		this.session.modelRegistry.refresh();
-		const allModels = this.session.modelRegistry.getAvailable();
+		await this.session.modelRuntime.refresh();
+		const allModels = [...(await this.session.modelRuntime.getAvailable())];
 
 		if (allModels.length === 0) {
 			this.showStatus("No models available");
@@ -4479,7 +4450,7 @@ export class InteractiveMode {
 			// Fall back to settings
 			const patterns = this.settingsManager.getEnabledModels();
 			if (patterns !== undefined && patterns.length > 0) {
-				const scopedModels = await resolveModelScope(patterns, this.session.modelRegistry);
+				const scopedModels = await resolveModelScope(patterns, this.session.modelRuntime);
 				currentEnabledIds = scopedModels.map((scoped) => `${scoped.model.provider}/${scoped.model.id}`);
 			}
 		}
@@ -4488,7 +4459,7 @@ export class InteractiveMode {
 		const updateSessionModels = async (enabledIds: string[] | null) => {
 			currentEnabledIds = enabledIds === null ? null : [...enabledIds];
 			if (enabledIds && enabledIds.length > 0 && enabledIds.length < allModels.length) {
-				const newScopedModels = await resolveModelScope(enabledIds, this.session.modelRegistry);
+				const newScopedModels = await resolveModelScope(enabledIds, this.session.modelRuntime);
 				this.session.setScopedModels(
 					newScopedModels.map((sm) => ({
 						model: sm.model,
@@ -4790,48 +4761,46 @@ export class InteractiveMode {
 	}
 
 	private getLoginProviderOptions(authType?: "oauth" | "api_key"): AuthSelectorProvider[] {
-		const authStorage = this.session.modelRegistry.authStorage;
-		const oauthProviders = authStorage.getOAuthProviders();
-		const oauthProviderIds = new Set(oauthProviders.map((provider) => provider.id));
-		const options: AuthSelectorProvider[] = oauthProviders.map((provider) => ({
-			id: provider.id,
-			name: provider.name,
-			authType: "oauth",
-		}));
-
-		const modelProviders = new Set(this.session.modelRegistry.getAll().map((model) => model.provider));
-		for (const providerId of modelProviders) {
-			if (!isApiKeyLoginProvider(providerId, oauthProviderIds)) {
-				continue;
+		const options: AuthSelectorProvider[] = [];
+		for (const provider of this.session.modelRuntime.getProviders()) {
+			const authStatus = this.session.modelRuntime.getProviderAuthStatus(provider.id);
+			const status = authStatus.configured
+				? {
+						type: this.session.modelRuntime.isUsingOAuth(provider.id) ? ("oauth" as const) : ("api_key" as const),
+						source: authStatus.label ?? authStatus.source,
+					}
+				: undefined;
+			if ((!authType || authType === "oauth") && provider.auth.oauth) {
+				options.push({
+					id: provider.id,
+					name: provider.name,
+					authType: "oauth",
+					method: provider.auth.oauth,
+					status,
+				});
+			}
+			if ((!authType || authType === "api_key") && provider.auth.apiKey) {
+				options.push({
+					id: provider.id,
+					name: provider.name,
+					authType: "api_key",
+					method: provider.auth.apiKey,
+					status,
+				});
 			}
-			options.push({
-				id: providerId,
-				name: this.session.modelRegistry.getProviderDisplayName(providerId),
-				authType: "api_key",
-			});
 		}
-
-		const filteredOptions = authType ? options.filter((option) => option.authType === authType) : options;
-		return filteredOptions.sort((a, b) => a.name.localeCompare(b.name));
+		return options.sort((a, b) => a.name.localeCompare(b.name));
 	}
 
-	private getLogoutProviderOptions(): AuthSelectorProvider[] {
-		const authStorage = this.session.modelRegistry.authStorage;
-		const options: AuthSelectorProvider[] = [];
-
-		for (const providerId of authStorage.list()) {
-			const credential = authStorage.get(providerId);
-			if (!credential) {
-				continue;
-			}
-			options.push({
+	private async getLogoutProviderOptions(): Promise<AuthSelectorProvider[]> {
+		return (await this.session.modelRuntime.listCredentials())
+			.map(({ providerId, type }) => ({
 				id: providerId,
-				name: this.session.modelRegistry.getProviderDisplayName(providerId),
-				authType: credential.type,
-			});
-		}
-
-		return options.sort((a, b) => a.name.localeCompare(b.name));
+				name: this.session.modelRuntime.getProvider(providerId)?.name ?? providerId,
+				authType: type,
+				status: { type, source: "stored credential" },
+			}))
+			.sort((a, b) => a.name.localeCompare(b.name));
 	}
 
 	private findLoginProviderOptions(providerRef: string): AuthSelectorProvider[] {
@@ -4848,6 +4817,7 @@ export class InteractiveMode {
 	}
 
 	private async handleLoginCommand(providerRef?: string): Promise<void> {
+		await this.session.modelRuntime.getAvailable();
 		if (!providerRef) {
 			this.showLoginAuthTypeSelector();
 			return;
@@ -4873,10 +4843,10 @@ export class InteractiveMode {
 	private async startProviderLogin(providerOption: AuthSelectorProvider): Promise<void> {
 		if (providerOption.authType === "oauth") {
 			await this.showLoginDialog(providerOption.id, providerOption.name);
-		} else if (providerOption.id === BEDROCK_PROVIDER_ID) {
-			this.showBedrockSetupDialog(providerOption.id, providerOption.name);
-		} else {
+		} else if (providerOption.method?.login) {
 			await this.showApiKeyLoginDialog(providerOption.id, providerOption.name);
+		} else {
+			this.showAmbientAuthDialog(providerOption);
 		}
 	}
 
@@ -4951,7 +4921,6 @@ export class InteractiveMode {
 		this.showSelector((done) => {
 			const selector = new OAuthSelectorComponent(
 				"login",
-				this.session.modelRegistry.authStorage,
 				providerOptions,
 				async (providerId, selectedAuthType) => {
 					done();
@@ -4973,7 +4942,6 @@ export class InteractiveMode {
 						this.ui.requestRender();
 					}
 				},
-				(providerId) => this.session.modelRegistry.getProviderAuthStatus(providerId),
 				initialSearchInput,
 			);
 			return { component: selector, focus: selector };
@@ -4986,7 +4954,7 @@ export class InteractiveMode {
 			return;
 		}
 
-		const providerOptions = this.getLogoutProviderOptions();
+		const providerOptions = await this.getLogoutProviderOptions();
 		if (providerOptions.length === 0) {
 			this.showStatus(
 				"No stored credentials to remove. /logout only removes credentials saved by /login; environment variables and models.json config are unchanged.",
@@ -4997,7 +4965,6 @@ export class InteractiveMode {
 		this.showSelector((done) => {
 			const selector = new OAuthSelectorComponent(
 				mode,
-				this.session.modelRegistry.authStorage,
 				providerOptions,
 				async (providerId: string) => {
 					done();
@@ -5008,8 +4975,7 @@ export class InteractiveMode {
 					}
 
 					try {
-						this.session.modelRegistry.authStorage.logout(providerOption.id);
-						this.session.modelRegistry.refresh();
+						await this.session.modelRuntime.logout(providerOption.id);
 						await this.updateAvailableProviderCount();
 						const message =
 							providerOption.authType === "oauth"
@@ -5035,14 +5001,14 @@ export class InteractiveMode {
 		authType: "oauth" | "api_key",
 		previousModel: Model<any> | undefined,
 	): Promise<void> {
-		this.session.modelRegistry.refresh();
+		await this.session.modelRuntime.getAvailable();
 
 		const actionLabel = authType === "oauth" ? `Logged in to ${providerName}` : `Saved API key for ${providerName}`;
 
 		let selectedModel: Model<any> | undefined;
 		let selectionError: string | undefined;
 		if (isUnknownModel(previousModel)) {
-			const availableModels = this.session.modelRegistry.getAvailable();
+			const availableModels = await this.session.modelRuntime.getAvailable();
 			const providerModels = availableModels.filter((model) => model.provider === providerId);
 			if (!hasDefaultModelProvider(providerId)) {
 				selectionError = `${actionLabel}, but no default model is configured for provider "${providerId}". Use /model to select a model.`;
@@ -5082,7 +5048,7 @@ export class InteractiveMode {
 		}
 	}
 
-	private showBedrockSetupDialog(providerId: string, providerName: string): void {
+	private showAmbientAuthDialog(providerOption: AuthSelectorProvider): void {
 		const restoreEditor = () => {
 			this.editorContainer.clear();
 			this.editorContainer.addChild(this.editor);
@@ -5092,17 +5058,12 @@ export class InteractiveMode {
 
 		const dialog = new LoginDialogComponent(
 			this.ui,
-			providerId,
+			providerOption.id,
 			() => restoreEditor(),
-			providerName,
-			"Amazon Bedrock setup",
+			providerOption.name,
+			`${providerOption.name} setup`,
 		);
-		dialog.showInfo([
-			theme.fg("text", "Amazon Bedrock uses AWS credentials instead of a single API key."),
-			theme.fg("text", "Configure an AWS profile, IAM keys, bearer token, or role-based credentials."),
-			theme.fg("muted", "See:"),
-			theme.fg("accent", `  ${path.join(getDocsPath(), "providers.md")}`),
-		]);
+		dialog.showInfo(`${providerOption.method?.name ?? "Authentication"} is configured outside pi.`, [], true);
 
 		this.editorContainer.clear();
 		this.editorContainer.addChild(dialog);
@@ -5135,13 +5096,7 @@ export class InteractiveMode {
 		};
 
 		try {
-			const apiKey = (await dialog.showPrompt("Enter API key:")).trim();
-			if (!apiKey) {
-				throw new Error("API key cannot be empty.");
-			}
-
-			this.session.modelRegistry.authStorage.set(providerId, { type: "api_key", key: apiKey });
-
+			await this.loginProvider(dialog, providerId, "api_key");
 			restoreEditor();
 			await this.completeProviderAuthentication(providerId, providerName, "api_key", previousModel);
 		} catch (error: unknown) {
@@ -5153,8 +5108,11 @@ export class InteractiveMode {
 		}
 	}
 
-	private showOAuthLoginSelect(dialog: LoginDialogComponent, prompt: OAuthSelectPrompt): Promise<string | undefined> {
-		return new Promise((resolve) => {
+	private showAuthSelect(
+		dialog: LoginDialogComponent,
+		prompt: Extract<AuthPrompt, { type: "select" }>,
+	): Promise<string> {
+		return new Promise((resolve, reject) => {
 			const restoreDialog = () => {
 				this.editorContainer.clear();
 				this.editorContainer.addChild(dialog);
@@ -5167,11 +5125,13 @@ export class InteractiveMode {
 				labels,
 				(optionLabel) => {
 					restoreDialog();
-					resolve(prompt.options.find((option) => option.label === optionLabel)?.id);
+					const id = prompt.options.find((option) => option.label === optionLabel)?.id;
+					if (id) resolve(id);
+					else reject(new Error("Login cancelled"));
 				},
 				() => {
 					restoreDialog();
-					resolve(undefined);
+					reject(new Error("Login cancelled"));
 				},
 			);
 			this.editorContainer.clear();
@@ -5181,40 +5141,63 @@ export class InteractiveMode {
 		});
 	}
 
+	private async showAuthPrompt(dialog: LoginDialogComponent, prompt: AuthPrompt): Promise<string> {
+		let response: Promise<string>;
+		if (prompt.type === "select") {
+			response = this.showAuthSelect(dialog, prompt);
+		} else if (prompt.type === "manual_code") {
+			response = dialog.showManualInput(prompt.message);
+		} else {
+			response = dialog.showPrompt(prompt.message, prompt.placeholder);
+		}
+		if (!prompt.signal) return response;
+		if (prompt.signal.aborted) throw new Error("Login cancelled");
+		const signal = prompt.signal;
+		let onAbort: (() => void) | undefined;
+		const aborted = new Promise<string>((_resolve, reject) => {
+			onAbort = () => reject(new Error("Login cancelled"));
+			signal.addEventListener("abort", onAbort, { once: true });
+		});
+		try {
+			return await Promise.race([response, aborted]);
+		} finally {
+			if (onAbort) signal.removeEventListener("abort", onAbort);
+		}
+	}
+
+	private notifyAuthDialog(dialog: LoginDialogComponent, event: AuthEvent): void {
+		if (event.type === "auth_url") {
+			dialog.showAuth(event.url, event.instructions);
+		} else if (event.type === "device_code") {
+			dialog.showDeviceCode(event);
+			dialog.showWaiting("Waiting for authentication...");
+		} else if (event.type === "info") {
+			dialog.showInfo(event.message, event.links);
+		} else {
+			dialog.showProgress(event.message);
+		}
+	}
+
+	private async loginProvider(
+		dialog: LoginDialogComponent,
+		providerId: string,
+		method: "api_key" | "oauth",
+	): Promise<void> {
+		await this.session.modelRuntime.login(providerId, method, {
+			signal: dialog.signal,
+			prompt: (prompt) => this.showAuthPrompt(dialog, prompt),
+			notify: (event) => this.notifyAuthDialog(dialog, event),
+		});
+	}
+
 	private async showLoginDialog(providerId: string, providerName: string): Promise<void> {
-		const providerInfo = this.session.modelRegistry.authStorage
-			.getOAuthProviders()
-			.find((provider) => provider.id === providerId);
 		const previousModel = this.session.model;
-
-		// Providers that use callback servers (can paste redirect URL)
-		const usesCallbackServer = providerInfo?.usesCallbackServer ?? false;
-
-		// Create login dialog component
-		const dialog = new LoginDialogComponent(
-			this.ui,
-			providerId,
-			(_success, _message) => {
-				// Completion handled below
-			},
-			providerName,
-		);
-
-		// Show dialog in editor container
+		const dialog = new LoginDialogComponent(this.ui, providerId, (_success, _message) => {}, providerName);
 		this.editorContainer.clear();
 		this.editorContainer.addChild(dialog);
 		this.ui.setFocus(dialog);
 		this.ui.requestRender();
 
-		// Promise for manual code input (racing with callback server)
-		let manualCodeResolve: ((code: string) => void) | undefined;
-		let manualCodeReject: ((err: Error) => void) | undefined;
-		const manualCodePromise = new Promise<string>((resolve, reject) => {
-			manualCodeResolve = resolve;
-			manualCodeReject = reject;
-		});
-
-		// Restore editor helper
 		const restoreEditor = () => {
 			this.editorContainer.clear();
 			this.editorContainer.addChild(this.editor);
@@ -5223,51 +5206,7 @@ export class InteractiveMode {
 		};
 
 		try {
-			await this.session.modelRegistry.authStorage.login(providerId as OAuthProviderId, {
-				onAuth: (info: { url: string; instructions?: string }) => {
-					dialog.showAuth(info.url, info.instructions);
-
-					if (usesCallbackServer) {
-						// Show input for manual paste, racing with callback
-						dialog
-							.showManualInput("Paste redirect URL below, or complete login in browser:")
-							.then((value) => {
-								if (value && manualCodeResolve) {
-									manualCodeResolve(value);
-									manualCodeResolve = undefined;
-								}
-							})
-							.catch(() => {
-								if (manualCodeReject) {
-									manualCodeReject(new Error("Login cancelled"));
-									manualCodeReject = undefined;
-								}
-							});
-					}
-					// For Anthropic: onPrompt is called immediately after
-				},
-
-				onDeviceCode: (info) => {
-					dialog.showDeviceCode(info);
-					dialog.showWaiting("Waiting for authentication...");
-				},
-
-				onPrompt: async (prompt: { message: string; placeholder?: string }) => {
-					return dialog.showPrompt(prompt.message, prompt.placeholder);
-				},
-
-				onProgress: (message: string) => {
-					dialog.showProgress(message);
-				},
-
-				onSelect: (prompt: OAuthSelectPrompt) => this.showOAuthLoginSelect(dialog, prompt),
-
-				onManualCodeInput: () => manualCodePromise,
-
-				signal: dialog.signal,
-			});
-
-			// Success
+			await this.loginProvider(dialog, providerId, "oauth");
 			restoreEditor();
 			await this.completeProviderAuthentication(providerId, providerName, "oauth", previousModel);
 		} catch (error: unknown) {
@@ -5368,7 +5307,7 @@ export class InteractiveMode {
 				showDiagnosticsWhenQuiet: true,
 			});
 			const savedImplicitProjectTrust = this.maybeSaveImplicitProjectTrustAfterReload();
-			const modelsJsonError = this.session.modelRegistry.getError();
+			const modelsJsonError = this.session.modelRuntime.getError();
 			if (modelsJsonError) {
 				this.showError(`models.json error: ${modelsJsonError}`);
 			}
@@ -5613,7 +5552,7 @@ export class InteractiveMode {
 		const stats = this.session.getSessionStats();
 		const sessionName = this.sessionManager.getSessionName();
 		const entries = this.sessionManager.getEntries();
-		const cacheWaste = computeCacheWaste(entries, this.session.modelRegistry);
+		const cacheWaste = computeCacheWaste(entries, this.session.modelRuntime);
 
 		// Cost/token totals per provider/model actually used (e.g. OpenRouter `auto`
 		// resolves to a concrete responseModel), sorted by cost descending.

packages/coding-agent/src/modes/rpc/rpc-mode.ts

@@ -465,7 +465,7 @@ export async function runRpcMode(runtimeHost: AgentSessionRuntime): Promise<neve
 			// =================================================================
 
 			case "set_model": {
-				const models = await session.modelRegistry.getAvailable();
+				const models = await session.modelRuntime.getAvailable();
 				const model = models.find((m) => m.provider === command.provider && m.id === command.modelId);
 				if (!model) {
 					return error(id, "set_model", `Model not found: ${command.provider}/${command.modelId}`);
@@ -483,7 +483,7 @@ export async function runRpcMode(runtimeHost: AgentSessionRuntime): Promise<neve
 			}
 
 			case "get_available_models": {
-				const models = await session.modelRegistry.getAvailable();
+				const models = await session.modelRuntime.getAvailable();
 				return success(id, "get_available_models", { models });
 			}
 

packages/coding-agent/test/agent-session-auto-compaction-queue.test.ts

@@ -7,9 +7,9 @@ import { getModel } from "@earendil-works/pi-ai/compat";
 import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
 import { AgentSession } from "../src/core/agent-session.ts";
 import { AuthStorage } from "../src/core/auth-storage.ts";
-import { ModelRegistry } from "../src/core/model-registry.ts";
 import { SessionManager } from "../src/core/session-manager.ts";
 import { SettingsManager } from "../src/core/settings-manager.ts";
+import { createModelRegistry, getModelRuntime } from "./model-runtime-test-utils.ts";
 import { createTestResourceLoader } from "./utilities.ts";
 
 describe("AgentSession auto-compaction queue resume", () => {
@@ -18,7 +18,7 @@ describe("AgentSession auto-compaction queue resume", () => {
 	let settingsManager: SettingsManager;
 	let tempDir: string;
 
-	beforeEach(() => {
+	beforeEach(async () => {
 		tempDir = join(tmpdir(), `pi-auto-compaction-queue-${Date.now()}`);
 		mkdirSync(tempDir, { recursive: true });
 		vi.useFakeTimers();
@@ -35,15 +35,15 @@ describe("AgentSession auto-compaction queue resume", () => {
 		sessionManager = SessionManager.inMemory();
 		settingsManager = SettingsManager.create(tempDir, tempDir);
 		const authStorage = AuthStorage.create(join(tempDir, "auth.json"));
-		authStorage.setRuntimeApiKey("anthropic", "test-key");
-		const modelRegistry = ModelRegistry.create(authStorage, tempDir);
+		await authStorage.modify("anthropic", async () => ({ type: "api_key", key: "test-key" }));
+		const modelRegistry = await createModelRegistry(authStorage, tempDir);
 
 		session = new AgentSession({
 			agent,
 			sessionManager,
 			settingsManager,
 			cwd: tempDir,
-			modelRegistry,
+			modelRuntime: getModelRuntime(modelRegistry),
 			resourceLoader: createTestResourceLoader(),
 		});
 	});

packages/coding-agent/test/agent-session-branching.test.ts

@@ -48,7 +48,7 @@ describe.skipIf(!API_KEY)("AgentSession forking", () => {
 		const model = getModel("anthropic", "claude-sonnet-4-5")!;
 		sessionManager = noSession ? SessionManager.inMemory(tempDir) : SessionManager.create(tempDir);
 		const authStorage = AuthStorage.create(join(tempDir, "auth.json"));
-		authStorage.setRuntimeApiKey("anthropic", API_KEY!);
+		await authStorage.modify("anthropic", async () => ({ type: "api_key", key: API_KEY! }));
 
 		const servicesOptions = {
 			agentDir: tempDir,

packages/coding-agent/test/agent-session-compaction.test.ts

@@ -1,3 +1,4 @@
+import { createModelRegistry, getModelRuntime } from "./model-runtime-test-utils.ts";
 /**
  * E2E tests for AgentSession compaction behavior.
  *
@@ -15,7 +16,6 @@ import { getModel } from "@earendil-works/pi-ai/compat";
 import { afterEach, beforeEach, describe, expect, it } from "vitest";
 import { AgentSession, type AgentSessionEvent } from "../src/core/agent-session.ts";
 import { AuthStorage } from "../src/core/auth-storage.ts";
-import { ModelRegistry } from "../src/core/model-registry.ts";
 import { SessionManager } from "../src/core/session-manager.ts";
 import { SettingsManager } from "../src/core/settings-manager.ts";
 import { createCodingTools } from "../src/index.ts";
@@ -27,7 +27,7 @@ describe.skipIf(!API_KEY)("AgentSession compaction e2e", () => {
 	let sessionManager: SessionManager;
 	let events: AgentSessionEvent[];
 
-	beforeEach(() => {
+	beforeEach(async () => {
 		// Create temp directory for session files
 		tempDir = join(tmpdir(), `pi-compaction-test-${Date.now()}`);
 		mkdirSync(tempDir, { recursive: true });
@@ -45,7 +45,7 @@ describe.skipIf(!API_KEY)("AgentSession compaction e2e", () => {
 		}
 	});
 
-	function createSession(inMemory = false) {
+	async function createSession(inMemory = false) {
 		const model = getModel("anthropic", "claude-sonnet-4-5")!;
 		const agent = new Agent({
 			getApiKey: () => API_KEY,
@@ -61,14 +61,14 @@ describe.skipIf(!API_KEY)("AgentSession compaction e2e", () => {
 		// Use minimal keepRecentTokens so small test conversations have something to summarize
 		settingsManager.applyOverrides({ compaction: { keepRecentTokens: 1 } });
 		const authStorage = AuthStorage.create(join(tempDir, "auth.json"));
-		const modelRegistry = ModelRegistry.create(authStorage);
+		const modelRegistry = await createModelRegistry(authStorage);
 
 		session = new AgentSession({
 			agent,
 			sessionManager,
 			settingsManager,
 			cwd: tempDir,
-			modelRegistry,
+			modelRuntime: getModelRuntime(modelRegistry),
 			resourceLoader: createTestResourceLoader(),
 		});
 
@@ -81,7 +81,7 @@ describe.skipIf(!API_KEY)("AgentSession compaction e2e", () => {
 	}
 
 	it("should trigger manual compaction via compact()", async () => {
-		createSession();
+		await createSession();
 
 		// Send a few prompts to build up history
 		await session.prompt("What is 2+2? Reply with just the number.");
@@ -107,7 +107,7 @@ describe.skipIf(!API_KEY)("AgentSession compaction e2e", () => {
 	}, 120000);
 
 	it("should maintain valid session state after compaction", async () => {
-		createSession();
+		await createSession();
 
 		// Build up history
 		await session.prompt("What is the capital of France? One word answer.");
@@ -132,7 +132,7 @@ describe.skipIf(!API_KEY)("AgentSession compaction e2e", () => {
 	}, 180000);
 
 	it("should persist compaction to session file", async () => {
-		createSession();
+		await createSession();
 
 		await session.prompt("Say hello");
 		await session.agent.waitForIdle();
@@ -160,7 +160,7 @@ describe.skipIf(!API_KEY)("AgentSession compaction e2e", () => {
 	}, 120000);
 
 	it("should work with --no-session mode (in-memory only)", async () => {
-		createSession(true); // in-memory mode
+		await createSession(true); // in-memory mode
 
 		// Send prompts
 		await session.prompt("What is 2+2? Reply with just the number.");
@@ -182,7 +182,7 @@ describe.skipIf(!API_KEY)("AgentSession compaction e2e", () => {
 	}, 120000);
 
 	it("should emit compaction events during manual compaction", async () => {
-		createSession();
+		await createSession();
 
 		// Build some history
 		await session.prompt("Say hello");

packages/coding-agent/test/agent-session-concurrent.test.ts

@@ -1,3 +1,4 @@
+import { createModelRegistry, getModelRuntime } from "./model-runtime-test-utils.ts";
 /**
  * Tests for AgentSession concurrent prompt guard.
  */
@@ -18,7 +19,6 @@ import { Type } from "typebox";
 import { afterEach, beforeEach, describe, expect, it } from "vitest";
 import { AgentSession } from "../src/core/agent-session.ts";
 import { AuthStorage } from "../src/core/auth-storage.ts";
-import { ModelRegistry } from "../src/core/model-registry.ts";
 import { SessionManager } from "../src/core/session-manager.ts";
 import { SettingsManager } from "../src/core/settings-manager.ts";
 import type { BuildSystemPromptOptions } from "../src/core/system-prompt.ts";
@@ -62,7 +62,7 @@ describe("AgentSession concurrent prompt guard", () => {
 	let session: AgentSession;
 	let tempDir: string;
 
-	beforeEach(() => {
+	beforeEach(async () => {
 		tempDir = join(tmpdir(), `pi-concurrent-test-${Date.now()}`);
 		mkdirSync(tempDir, { recursive: true });
 	});
@@ -78,7 +78,7 @@ describe("AgentSession concurrent prompt guard", () => {
 		}
 	});
 
-	function createSession() {
+	async function createSession() {
 		const model = getModel("anthropic", "claude-sonnet-4-5")!;
 		let abortSignal: AbortSignal | undefined;
 
@@ -111,16 +111,16 @@ describe("AgentSession concurrent prompt guard", () => {
 		const sessionManager = SessionManager.inMemory();
 		const settingsManager = SettingsManager.create(tempDir, tempDir);
 		const authStorage = AuthStorage.create(join(tempDir, "auth.json"));
-		const modelRegistry = ModelRegistry.create(authStorage, tempDir);
+		const modelRegistry = await createModelRegistry(authStorage, tempDir);
 		// Set a runtime API key so validation passes
-		authStorage.setRuntimeApiKey("anthropic", "test-key");
+		await authStorage.modify("anthropic", async () => ({ type: "api_key", key: "test-key" }));
 
 		session = new AgentSession({
 			agent,
 			sessionManager,
 			settingsManager,
 			cwd: tempDir,
-			modelRegistry,
+			modelRuntime: getModelRuntime(modelRegistry),
 			resourceLoader: createTestResourceLoader(),
 		});
 
@@ -128,7 +128,7 @@ describe("AgentSession concurrent prompt guard", () => {
 	}
 
 	it("should throw when prompt() called while streaming", async () => {
-		createSession();
+		await createSession();
 
 		// Start first prompt (don't await, it will block until abort)
 		const firstPrompt = session.prompt("First message");
@@ -150,7 +150,7 @@ describe("AgentSession concurrent prompt guard", () => {
 	});
 
 	it("should allow steer() while streaming", async () => {
-		createSession();
+		await createSession();
 
 		// Start first prompt
 		const firstPrompt = session.prompt("First message");
@@ -166,7 +166,7 @@ describe("AgentSession concurrent prompt guard", () => {
 	});
 
 	it("should allow followUp() while streaming", async () => {
-		createSession();
+		await createSession();
 
 		// Start first prompt
 		const firstPrompt = session.prompt("First message");
@@ -236,8 +236,8 @@ describe("AgentSession concurrent prompt guard", () => {
 		const sessionManager = SessionManager.inMemory();
 		const settingsManager = SettingsManager.create(tempDir, tempDir);
 		const authStorage = AuthStorage.create(join(tempDir, "auth.json"));
-		const modelRegistry = ModelRegistry.create(authStorage, tempDir);
-		authStorage.setRuntimeApiKey("anthropic", "test-key");
+		const modelRegistry = await createModelRegistry(authStorage, tempDir);
+		await authStorage.modify("anthropic", async () => ({ type: "api_key", key: "test-key" }));
 
 		const extensionsResult = await createTestExtensionsResult([
 			(pi) => {
@@ -255,7 +255,7 @@ describe("AgentSession concurrent prompt guard", () => {
 			sessionManager,
 			settingsManager,
 			cwd: tempDir,
-			modelRegistry,
+			modelRuntime: getModelRuntime(modelRegistry),
 			resourceLoader: createTestResourceLoader({ extensionsResult }),
 		});
 		session.subscribe((event) => {
@@ -314,15 +314,15 @@ describe("AgentSession concurrent prompt guard", () => {
 		const sessionManager = SessionManager.inMemory();
 		const settingsManager = SettingsManager.create(tempDir, tempDir);
 		const authStorage = AuthStorage.create(join(tempDir, "auth.json"));
-		const modelRegistry = ModelRegistry.create(authStorage, tempDir);
-		authStorage.setRuntimeApiKey("anthropic", "test-key");
+		const modelRegistry = await createModelRegistry(authStorage, tempDir);
+		await authStorage.modify("anthropic", async () => ({ type: "api_key", key: "test-key" }));
 
 		session = new AgentSession({
 			agent,
 			sessionManager,
 			settingsManager,
 			cwd: tempDir,
-			modelRegistry,
+			modelRuntime: getModelRuntime(modelRegistry),
 			resourceLoader: createTestResourceLoader(),
 		});
 
@@ -420,15 +420,15 @@ describe("AgentSession concurrent prompt guard", () => {
 		const sessionManager = SessionManager.inMemory();
 		const settingsManager = SettingsManager.create(tempDir, tempDir);
 		const authStorage = AuthStorage.create(join(tempDir, "auth.json"));
-		const modelRegistry = ModelRegistry.create(authStorage, tempDir);
-		authStorage.setRuntimeApiKey("anthropic", "test-key");
+		const modelRegistry = await createModelRegistry(authStorage, tempDir);
+		await authStorage.modify("anthropic", async () => ({ type: "api_key", key: "test-key" }));
 
 		session = new AgentSession({
 			agent,
 			sessionManager,
 			settingsManager,
 			cwd: tempDir,
-			modelRegistry,
+			modelRuntime: getModelRuntime(modelRegistry),
 			resourceLoader: createTestResourceLoader(),
 			baseToolsOverride: { dummy: tool },
 		});
@@ -567,15 +567,15 @@ describe("AgentSession concurrent prompt guard", () => {
 		const sessionManager = SessionManager.inMemory();
 		const settingsManager = SettingsManager.create(tempDir, tempDir);
 		const authStorage = AuthStorage.create(join(tempDir, "auth.json"));
-		const modelRegistry = ModelRegistry.create(authStorage, tempDir);
-		authStorage.setRuntimeApiKey("anthropic", "test-key");
+		const modelRegistry = await createModelRegistry(authStorage, tempDir);
+		await authStorage.modify("anthropic", async () => ({ type: "api_key", key: "test-key" }));
 
 		session = new AgentSession({
 			agent,
 			sessionManager,
 			settingsManager,
 			cwd: tempDir,
-			modelRegistry,
+			modelRuntime: getModelRuntime(modelRegistry),
 			resourceLoader: createTestResourceLoader(),
 			baseToolsOverride: { dummy: tool },
 		});

packages/coding-agent/test/agent-session-dynamic-provider.test.ts

@@ -4,6 +4,7 @@ import { join } from "node:path";
 import { getModel } from "@earendil-works/pi-ai/compat";
 import { afterEach, beforeEach, describe, expect, it } from "vitest";
 import { AuthStorage } from "../src/core/auth-storage.ts";
+import { ModelRuntime } from "../src/core/model-runtime.ts";
 import { DefaultResourceLoader } from "../src/core/resource-loader.ts";
 import type { ExtensionFactory } from "../src/core/sdk.ts";
 import { createAgentSession } from "../src/core/sdk.ts";
@@ -30,7 +31,11 @@ describe("AgentSession dynamic provider registration", () => {
 		const settingsManager = SettingsManager.create(tempDir, agentDir);
 		const sessionManager = SessionManager.inMemory();
 		const authStorage = AuthStorage.create(join(agentDir, "auth.json"));
-		authStorage.setRuntimeApiKey("anthropic", "test-key");
+		await authStorage.modify("anthropic", async () => ({ type: "api_key", key: "test-key" }));
+		const modelRuntime = await ModelRuntime.create({
+			credentials: authStorage,
+			modelsPath: join(agentDir, "models.json"),
+		});
 		const resourceLoader = new DefaultResourceLoader({
 			cwd: tempDir,
 			agentDir,
@@ -45,7 +50,7 @@ describe("AgentSession dynamic provider registration", () => {
 			model: getModel("anthropic", "claude-sonnet-4-5")!,
 			settingsManager,
 			sessionManager,
-			authStorage,
+			modelRuntime,
 			resourceLoader,
 		});
 

packages/coding-agent/test/agent-session-retry.test.ts

@@ -7,9 +7,9 @@ import { Type } from "typebox";
 import { afterEach, beforeEach, describe, expect, it } from "vitest";
 import { AgentSession } from "../src/core/agent-session.ts";
 import { AuthStorage } from "../src/core/auth-storage.ts";
-import { ModelRegistry } from "../src/core/model-registry.ts";
 import { SessionManager } from "../src/core/session-manager.ts";
 import { SettingsManager } from "../src/core/settings-manager.ts";
+import { createModelRegistry, getModelRuntime } from "./model-runtime-test-utils.ts";
 import { createTestResourceLoader } from "./utilities.ts";
 
 class MockAssistantStream extends EventStream<AssistantMessageEvent, AssistantMessage> {
@@ -54,7 +54,7 @@ describe("AgentSession retry", () => {
 	let session: AgentSession;
 	let tempDir: string;
 
-	beforeEach(() => {
+	beforeEach(async () => {
 		tempDir = join(tmpdir(), `pi-retry-test-${Date.now()}`);
 		mkdirSync(tempDir, { recursive: true });
 	});
@@ -68,7 +68,11 @@ describe("AgentSession retry", () => {
 		}
 	});
 
-	function createSession(options?: { failCount?: number; maxRetries?: number; delayAssistantMessageEndMs?: number }) {
+	async function createSession(options?: {
+		failCount?: number;
+		maxRetries?: number;
+		delayAssistantMessageEndMs?: number;
+	}) {
 		const failCount = options?.failCount ?? 1;
 		const maxRetries = options?.maxRetries ?? 3;
 		const delayAssistantMessageEndMs = options?.delayAssistantMessageEndMs ?? 0;
@@ -102,8 +106,8 @@ describe("AgentSession retry", () => {
 		const sessionManager = SessionManager.inMemory();
 		const settingsManager = SettingsManager.create(tempDir, tempDir);
 		const authStorage = AuthStorage.create(join(tempDir, "auth.json"));
-		const modelRegistry = ModelRegistry.create(authStorage, tempDir);
-		authStorage.setRuntimeApiKey("anthropic", "test-key");
+		const modelRegistry = await createModelRegistry(authStorage, tempDir);
+		await authStorage.modify("anthropic", async () => ({ type: "api_key", key: "test-key" }));
 		settingsManager.applyOverrides({ retry: { enabled: true, maxRetries, baseDelayMs: 1 } });
 
 		session = new AgentSession({
@@ -111,7 +115,7 @@ describe("AgentSession retry", () => {
 			sessionManager,
 			settingsManager,
 			cwd: tempDir,
-			modelRegistry,
+			modelRuntime: getModelRuntime(modelRegistry),
 			resourceLoader: createTestResourceLoader(),
 		});
 
@@ -130,7 +134,7 @@ describe("AgentSession retry", () => {
 	}
 
 	it("retries after a transient error and succeeds", async () => {
-		const created = createSession({ failCount: 1 });
+		const created = await createSession({ failCount: 1 });
 		const events: string[] = [];
 		created.session.subscribe((event) => {
 			if (event.type === "auto_retry_start") events.push(`start:${event.attempt}`);
@@ -145,7 +149,7 @@ describe("AgentSession retry", () => {
 	});
 
 	it("exhausts max retries and emits failure", async () => {
-		const created = createSession({ failCount: 99, maxRetries: 2 });
+		const created = await createSession({ failCount: 99, maxRetries: 2 });
 		const events: string[] = [];
 		created.session.subscribe((event) => {
 			if (event.type === "auto_retry_start") events.push(`start:${event.attempt}`);
@@ -162,7 +166,7 @@ describe("AgentSession retry", () => {
 	});
 
 	it("prompt waits for retry completion even when assistant message_end handling is delayed", async () => {
-		const created = createSession({ failCount: 1, delayAssistantMessageEndMs: 40 });
+		const created = await createSession({ failCount: 1, delayAssistantMessageEndMs: 40 });
 
 		await created.session.prompt("Test");
 
@@ -171,7 +175,7 @@ describe("AgentSession retry", () => {
 	});
 
 	it("retries provider network_error failures", async () => {
-		const created = createSession({ failCount: 0 });
+		const created = await createSession({ failCount: 0 });
 		let callCount = 0;
 		const streamFn = () => {
 			callCount++;
@@ -204,15 +208,15 @@ describe("AgentSession retry", () => {
 		const sessionManager = SessionManager.inMemory();
 		const settingsManager = SettingsManager.create(tempDir, tempDir);
 		const authStorage = AuthStorage.create(join(tempDir, "auth.json"));
-		const modelRegistry = ModelRegistry.create(authStorage, tempDir);
-		authStorage.setRuntimeApiKey("anthropic", "test-key");
+		const modelRegistry = await createModelRegistry(authStorage, tempDir);
+		await authStorage.modify("anthropic", async () => ({ type: "api_key", key: "test-key" }));
 		settingsManager.applyOverrides({ retry: { enabled: true, maxRetries: 3, baseDelayMs: 1 } });
 		session = new AgentSession({
 			agent,
 			sessionManager,
 			settingsManager,
 			cwd: tempDir,
-			modelRegistry,
+			modelRuntime: getModelRuntime(modelRegistry),
 			resourceLoader: createTestResourceLoader(),
 		});
 
@@ -289,8 +293,8 @@ describe("AgentSession retry", () => {
 		const sessionManager = SessionManager.inMemory();
 		const settingsManager = SettingsManager.create(tempDir, tempDir);
 		const authStorage = AuthStorage.create(join(tempDir, "auth.json"));
-		const modelRegistry = ModelRegistry.create(authStorage, tempDir);
-		authStorage.setRuntimeApiKey("anthropic", "test-key");
+		const modelRegistry = await createModelRegistry(authStorage, tempDir);
+		await authStorage.modify("anthropic", async () => ({ type: "api_key", key: "test-key" }));
 		settingsManager.applyOverrides({ retry: { enabled: true, maxRetries: 3, baseDelayMs: 1 } });
 
 		session = new AgentSession({
@@ -298,7 +302,7 @@ describe("AgentSession retry", () => {
 			sessionManager,
 			settingsManager,
 			cwd: tempDir,
-			modelRegistry,
+			modelRuntime: getModelRuntime(modelRegistry),
 			resourceLoader: createTestResourceLoader(),
 			baseToolsOverride: { echo: echoTool },
 		});

packages/coding-agent/test/agent-session-runtime-events.test.ts

@@ -10,6 +10,7 @@ import {
 	createAgentSessionServices,
 } from "../src/core/agent-session-runtime.ts";
 import { AuthStorage } from "../src/core/auth-storage.ts";
+import { ModelRuntime } from "../src/core/model-runtime.ts";
 import { SessionManager } from "../src/core/session-manager.ts";
 import type {
 	ExtensionFactory,
@@ -42,11 +43,33 @@ describe("AgentSessionRuntime session lifecycle events", () => {
 		faux.setResponses([fauxAssistantMessage("one"), fauxAssistantMessage("two"), fauxAssistantMessage("three")]);
 
 		const authStorage = AuthStorage.inMemory();
-		authStorage.setRuntimeApiKey(faux.getModel().provider, "faux-key");
+		await authStorage.modify(faux.getModel().provider, async () => ({ type: "api_key", key: "faux-key" }));
+		const modelRuntime = await ModelRuntime.create({
+			credentials: authStorage,
+			modelsPath: join(tempDir, "models.json"),
+		});
+		const model = faux.getModel();
+		modelRuntime.registerProvider(model.provider, {
+			baseUrl: model.baseUrl,
+			api: model.api,
+			models: [
+				{
+					id: model.id,
+					name: model.name,
+					api: model.api,
+					reasoning: model.reasoning,
+					input: model.input,
+					cost: model.cost,
+					contextWindow: model.contextWindow,
+					maxTokens: model.maxTokens,
+					baseUrl: model.baseUrl,
+				},
+			],
+		});
 
 		const runtimeOptions = {
 			agentDir: tempDir,
-			authStorage,
+			modelRuntime,
 			model: faux.getModel(),
 			resourceLoaderOptions: {
 				extensionFactories: [extensionFactory],

packages/coding-agent/test/agent-session-stats.test.ts

@@ -3,9 +3,9 @@ import { type AssistantMessage, getModel, type Usage } from "@earendil-works/pi-
 import { describe, expect, it } from "vitest";
 import { AgentSession } from "../src/core/agent-session.ts";
 import { AuthStorage } from "../src/core/auth-storage.ts";
-import { ModelRegistry } from "../src/core/model-registry.ts";
 import { SessionManager } from "../src/core/session-manager.ts";
 import { SettingsManager } from "../src/core/settings-manager.ts";
+import { createInMemoryModelRegistry, getModelRuntime } from "./model-runtime-test-utils.ts";
 import { createTestResourceLoader } from "./utilities.ts";
 
 const model = getModel("anthropic", "claude-sonnet-4-5")!;
@@ -48,11 +48,11 @@ function createUserMessage(text: string, timestamp: number) {
 	};
 }
 
-function createSession() {
+async function createSession() {
 	const settingsManager = SettingsManager.inMemory();
 	const sessionManager = SessionManager.inMemory();
 	const authStorage = AuthStorage.inMemory();
-	authStorage.setRuntimeApiKey("anthropic", "test-key");
+	await authStorage.modify("anthropic", async () => ({ type: "api_key", key: "test-key" }));
 	const session = new AgentSession({
 		agent: new Agent({
 			getApiKey: () => "test-key",
@@ -66,7 +66,7 @@ function createSession() {
 		sessionManager,
 		settingsManager,
 		cwd: process.cwd(),
-		modelRegistry: ModelRegistry.inMemory(authStorage),
+		modelRuntime: getModelRuntime(await createInMemoryModelRegistry(authStorage)),
 		resourceLoader: createTestResourceLoader(),
 	});
 
@@ -78,8 +78,8 @@ function syncAgentMessages(session: AgentSession, sessionManager: SessionManager
 }
 
 describe("AgentSession.getSessionStats", () => {
-	it("exposes the current context usage alongside token totals", () => {
-		const { session, sessionManager } = createSession();
+	it("exposes the current context usage alongside token totals", async () => {
+		const { session, sessionManager } = await createSession();
 
 		try {
 			sessionManager.appendMessage(createUserMessage("hello", 1));
@@ -96,8 +96,8 @@ describe("AgentSession.getSessionStats", () => {
 		}
 	});
 
-	it("reports unknown current context usage immediately after compaction", () => {
-		const { session, sessionManager } = createSession();
+	it("reports unknown current context usage immediately after compaction", async () => {
+		const { session, sessionManager } = await createSession();
 
 		try {
 			sessionManager.appendMessage(createUserMessage("first", 1));
@@ -119,8 +119,8 @@ describe("AgentSession.getSessionStats", () => {
 		}
 	});
 
-	it("uses post-compaction usage for current context instead of stale kept usage", () => {
-		const { session, sessionManager } = createSession();
+	it("uses post-compaction usage for current context instead of stale kept usage", async () => {
+		const { session, sessionManager } = await createSession();
 
 		try {
 			sessionManager.appendMessage(createUserMessage("first", 1));
@@ -143,8 +143,8 @@ describe("AgentSession.getSessionStats", () => {
 		}
 	});
 
-	it("ignores zero-usage messages when checking for post-compaction context usage", () => {
-		const { session, sessionManager } = createSession();
+	it("ignores zero-usage messages when checking for post-compaction context usage", async () => {
+		const { session, sessionManager } = await createSession();
 
 		try {
 			sessionManager.appendMessage(createUserMessage("first", 1));

packages/coding-agent/test/agent-session-tree-navigation.test.ts

@@ -15,8 +15,8 @@ import { API_KEY, createTestSession, type TestSessionContext } from "./utilities
 describe.skipIf(!API_KEY)("AgentSession tree navigation e2e", () => {
 	let ctx: TestSessionContext;
 
-	beforeEach(() => {
-		ctx = createTestSession({
+	beforeEach(async () => {
+		ctx = await createTestSession({
 			systemPrompt: "You are a helpful assistant. Reply with just a few words.",
 			settingsOverrides: { compaction: { keepRecentTokens: 1 } },
 		});
@@ -279,8 +279,8 @@ describe.skipIf(!API_KEY)("AgentSession tree navigation e2e", () => {
 describe.skipIf(!API_KEY)("AgentSession tree navigation - branch scenarios", () => {
 	let ctx: TestSessionContext;
 
-	beforeEach(() => {
-		ctx = createTestSession({
+	beforeEach(async () => {
+		ctx = await createTestSession({
 			systemPrompt: "You are a helpful assistant. Reply with just a few words.",
 		});
 	});

packages/coding-agent/test/auth-storage.test.ts

@@ -1,17 +1,14 @@
 import { existsSync, mkdirSync, readFileSync, rmSync, writeFileSync } from "node:fs";
 import { tmpdir } from "node:os";
 import { join } from "node:path";
-import { registerOAuthProvider } from "@earendil-works/pi-ai/oauth";
+import { createModels, type Provider } from "@earendil-works/pi-ai";
 import lockfile from "proper-lockfile";
 import { afterEach, beforeEach, describe, expect, test, vi } from "vitest";
 import { AuthStorage } from "../src/core/auth-storage.ts";
-import { clearConfigValueCache, resolveConfigValueUncached } from "../src/core/resolve-config-value.ts";
-import * as shellModule from "../src/utils/shell.ts";
 
 describe("AuthStorage", () => {
 	let tempDir: string;
 	let authJsonPath: string;
-	let authStorage: AuthStorage;
 
 	beforeEach(() => {
 		tempDir = join(tmpdir(), `pi-test-auth-storage-${Date.now()}-${Math.random().toString(36).slice(2)}`);
@@ -20,680 +17,201 @@ describe("AuthStorage", () => {
 	});
 
 	afterEach(() => {
-		if (tempDir && existsSync(tempDir)) {
-			rmSync(tempDir, { recursive: true });
-		}
-		clearConfigValueCache();
+		if (existsSync(tempDir)) rmSync(tempDir, { recursive: true });
 		vi.restoreAllMocks();
 	});
 
-	function writeAuthJson(data: Record<string, unknown>) {
+	function writeAuthJson(data: Record<string, unknown>): void {
 		writeFileSync(authJsonPath, JSON.stringify(data));
 	}
 
-	function toShPath(value: string): string {
-		return value.replace(/\\/g, "/").replace(/"/g, '\\"');
-	}
+	test("reads and resolves stored API-key credentials", async () => {
+		const original = process.env.TEST_AUTH_STORAGE_KEY;
+		process.env.TEST_AUTH_STORAGE_KEY = "environment-key";
+		try {
+			writeAuthJson({ anthropic: { type: "api_key", key: "$TEST_AUTH_STORAGE_KEY" } });
+			const storage = AuthStorage.create(authJsonPath);
+			expect(await storage.read("anthropic")).toEqual({ type: "api_key", key: "environment-key" });
+		} finally {
+			if (original === undefined) delete process.env.TEST_AUTH_STORAGE_KEY;
+			else process.env.TEST_AUTH_STORAGE_KEY = original;
+		}
+	});
 
-	describe("API key resolution", () => {
-		test("literal API key is returned directly", async () => {
-			writeAuthJson({
-				anthropic: { type: "api_key", key: "sk-ant-literal-key" },
-			});
+	test("resolves command-backed API-key credentials", async () => {
+		writeAuthJson({ anthropic: { type: "api_key", key: "!printf 'command-key'" } });
+		const storage = AuthStorage.create(authJsonPath);
+		expect(await storage.read("anthropic")).toEqual({ type: "api_key", key: "command-key" });
+	});
 
-			authStorage = AuthStorage.create(authJsonPath);
-			const apiKey = await authStorage.getApiKey("anthropic");
+	test("returns OAuth credentials unchanged", async () => {
+		const credential = {
+			type: "oauth" as const,
+			access: "access-token",
+			refresh: "refresh-token",
+			expires: Date.now() + 60_000,
+		};
+		const storage = AuthStorage.inMemory({ anthropic: credential });
+		expect(await storage.read("anthropic")).toEqual(credential);
+	});
 
-			expect(apiKey).toBe("sk-ant-literal-key");
+	test("credential-scoped env takes precedence and remains inspectable", async () => {
+		writeAuthJson({
+			anthropic: {
+				type: "api_key",
+				key: "$SCOPED_KEY",
+				env: { SCOPED_KEY: "scoped-value", REGION: "test-region" },
+			},
+		});
+		const storage = AuthStorage.create(authJsonPath);
+		expect(await storage.read("anthropic")).toMatchObject({
+			key: "scoped-value",
+			env: { SCOPED_KEY: "scoped-value", REGION: "test-region" },
+		});
+	});
+
+	test("modify persists a credential while preserving unrelated external edits", async () => {
+		writeAuthJson({ anthropic: { type: "api_key", key: "old" } });
+		const storage = AuthStorage.create(authJsonPath);
+		writeAuthJson({
+			anthropic: { type: "api_key", key: "old" },
+			openai: { type: "api_key", key: "external" },
 		});
 
-		test("apiKey with ! prefix executes command and uses stdout", async () => {
-			writeAuthJson({
-				anthropic: { type: "api_key", key: "!echo test-api-key-from-command" },
-			});
+		await storage.modify("anthropic", async () => ({ type: "api_key", key: "new" }));
 
-			authStorage = AuthStorage.create(authJsonPath);
-			const apiKey = await authStorage.getApiKey("anthropic");
+		expect(JSON.parse(readFileSync(authJsonPath, "utf8"))).toEqual({
+			anthropic: { type: "api_key", key: "new" },
+			openai: { type: "api_key", key: "external" },
+		});
+	});
 
-			expect(apiKey).toBe("test-api-key-from-command");
+	test("modify with undefined leaves the current credential unchanged", async () => {
+		writeAuthJson({ anthropic: { type: "api_key", key: "stored" } });
+		const storage = AuthStorage.create(authJsonPath);
+		expect(await storage.modify("anthropic", async () => undefined)).toEqual({ type: "api_key", key: "stored" });
+		expect(await storage.read("anthropic")).toEqual({ type: "api_key", key: "stored" });
+	});
+
+	test("serializes concurrent modifications", async () => {
+		writeAuthJson({});
+		const first = AuthStorage.create(authJsonPath);
+		const second = AuthStorage.create(authJsonPath);
+		await Promise.all([
+			first.modify("anthropic", async () => ({ type: "api_key", key: "anthropic-key" })),
+			second.modify("openai", async () => ({ type: "api_key", key: "openai-key" })),
+		]);
+		expect(JSON.parse(readFileSync(authJsonPath, "utf8"))).toEqual({
+			anthropic: { type: "api_key", key: "anthropic-key" },
+			openai: { type: "api_key", key: "openai-key" },
+		});
+	});
+
+	test("delete removes one credential while preserving others", async () => {
+		writeAuthJson({
+			anthropic: { type: "api_key", key: "anthropic-key" },
+			openai: { type: "api_key", key: "openai-key" },
+		});
+		const storage = AuthStorage.create(authJsonPath);
+		writeAuthJson({
+			anthropic: { type: "api_key", key: "anthropic-key" },
+			openai: { type: "api_key", key: "openai-key" },
+			google: { type: "api_key", key: "external-key" },
+		});
+		await storage.delete("anthropic");
+		await expect(storage.list()).resolves.toEqual([
+			{ providerId: "openai", type: "api_key" },
+			{ providerId: "google", type: "api_key" },
+		]);
+		expect(await storage.read("anthropic")).toBeUndefined();
+		expect(await storage.read("openai")).toEqual({ type: "api_key", key: "openai-key" });
+		expect(await storage.read("google")).toEqual({ type: "api_key", key: "external-key" });
+	});
+
+	test("in-memory storage implements the same credential-store behavior", async () => {
+		const storage = AuthStorage.inMemory({ anthropic: { type: "api_key", key: "initial" } });
+		expect(await storage.read("anthropic")).toEqual({ type: "api_key", key: "initial" });
+		await storage.modify("anthropic", async () => ({ type: "api_key", key: "updated" }));
+		expect(await storage.read("anthropic")).toEqual({ type: "api_key", key: "updated" });
+		await storage.delete("anthropic");
+		await expect(storage.list()).resolves.toEqual([]);
+	});
+
+	test("does not write after lock acquisition failure and recovers on retry", async () => {
+		writeAuthJson({ anthropic: { type: "api_key", key: "stored" } });
+		const storage = AuthStorage.create(authJsonPath);
+		const lockSpy = vi.spyOn(lockfile, "lock").mockRejectedValueOnce(new Error("lock unavailable"));
+
+		await expect(storage.modify("openai", async () => ({ type: "api_key", key: "new" }))).rejects.toThrow(
+			"lock unavailable",
+		);
+		expect(JSON.parse(readFileSync(authJsonPath, "utf8"))).toEqual({
+			anthropic: { type: "api_key", key: "stored" },
 		});
 
-		test("apiKey with ! prefix trims whitespace from command output", async () => {
-			writeAuthJson({
-				anthropic: { type: "api_key", key: "!echo '  spaced-key  '" },
-			});
-
-			authStorage = AuthStorage.create(authJsonPath);
-			const apiKey = await authStorage.getApiKey("anthropic");
-
-			expect(apiKey).toBe("spaced-key");
+		lockSpy.mockRestore();
+		await storage.modify("openai", async () => ({ type: "api_key", key: "new" }));
+		expect(JSON.parse(readFileSync(authJsonPath, "utf8"))).toEqual({
+			anthropic: { type: "api_key", key: "stored" },
+			openai: { type: "api_key", key: "new" },
 		});
+	});
 
-		test("apiKey with ! prefix handles multiline output (uses trimmed result)", async () => {
-			writeAuthJson({
-				anthropic: { type: "api_key", key: "!printf 'line1\\nline2'" },
-			});
-
-			authStorage = AuthStorage.create(authJsonPath);
-			const apiKey = await authStorage.getApiKey("anthropic");
-
-			expect(apiKey).toBe("line1\nline2");
+	test("surfaces a compromised OAuth refresh lock and allows a later retry", async () => {
+		const providerId = "oauth-provider";
+		writeAuthJson({
+			[providerId]: {
+				type: "oauth",
+				access: "expired-access",
+				refresh: "refresh-token",
+				expires: 0,
+			},
 		});
-
-		test("apiKey with ! prefix returns undefined on command failure", async () => {
-			writeAuthJson({
-				anthropic: { type: "api_key", key: "!exit 1" },
-			});
-
-			authStorage = AuthStorage.create(authJsonPath);
-			const apiKey = await authStorage.getApiKey("anthropic");
-
-			expect(apiKey).toBeUndefined();
-		});
-
-		test("apiKey with ! prefix returns undefined on nonexistent command", async () => {
-			writeAuthJson({
-				anthropic: { type: "api_key", key: "!nonexistent-command-12345" },
-			});
-
-			authStorage = AuthStorage.create(authJsonPath);
-			const apiKey = await authStorage.getApiKey("anthropic");
-
-			expect(apiKey).toBeUndefined();
-		});
-
-		test("apiKey with ! prefix returns undefined on empty output", async () => {
-			writeAuthJson({
-				anthropic: { type: "api_key", key: "!printf ''" },
-			});
-
-			authStorage = AuthStorage.create(authJsonPath);
-			const apiKey = await authStorage.getApiKey("anthropic");
-
-			expect(apiKey).toBeUndefined();
-		});
-
-		test("apiKey with $ prefix resolves to env value", async () => {
-			const originalEnv = process.env.TEST_AUTH_API_KEY_12345;
-			process.env.TEST_AUTH_API_KEY_12345 = "env-api-key-value";
-
-			try {
-				writeAuthJson({
-					anthropic: { type: "api_key", key: "$TEST_AUTH_API_KEY_12345" },
-				});
-
-				authStorage = AuthStorage.create(authJsonPath);
-				const apiKey = await authStorage.getApiKey("anthropic");
-
-				expect(apiKey).toBe("env-api-key-value");
-			} finally {
-				if (originalEnv === undefined) {
-					delete process.env.TEST_AUTH_API_KEY_12345;
-				} else {
-					process.env.TEST_AUTH_API_KEY_12345 = originalEnv;
-				}
-			}
-		});
-
-		test("apiKey env bag takes precedence over process.env", async () => {
-			const originalEnv = process.env.TEST_AUTH_SCOPED_API_KEY_12345;
-			process.env.TEST_AUTH_SCOPED_API_KEY_12345 = "process-env-value";
-
-			try {
-				writeAuthJson({
-					anthropic: {
-						type: "api_key",
-						key: "$TEST_AUTH_SCOPED_API_KEY_12345",
-						env: { TEST_AUTH_SCOPED_API_KEY_12345: "credential-env-value" },
+		const storage = AuthStorage.create(authJsonPath);
+		const provider: Provider = {
+			id: providerId,
+			name: "OAuth Provider",
+			auth: {
+				oauth: {
+					name: "OAuth",
+					login: async () => {
+						throw new Error("not used");
 					},
-				});
-
-				authStorage = AuthStorage.create(authJsonPath);
-
-				expect(await authStorage.getApiKey("anthropic")).toBe("credential-env-value");
-				expect(authStorage.getProviderEnv("anthropic")).toEqual({
-					TEST_AUTH_SCOPED_API_KEY_12345: "credential-env-value",
-				});
-			} finally {
-				if (originalEnv === undefined) {
-					delete process.env.TEST_AUTH_SCOPED_API_KEY_12345;
-				} else {
-					process.env.TEST_AUTH_SCOPED_API_KEY_12345 = originalEnv;
-				}
-			}
-		});
-
-		test("apiKey with braced env syntax resolves to env value", async () => {
-			const originalEnv = process.env.TEST_AUTH_BRACED_API_KEY_12345;
-			process.env.TEST_AUTH_BRACED_API_KEY_12345 = "braced-env-api-key-value";
-			const bracedKey = "$" + "{TEST_AUTH_BRACED_API_KEY_12345}";
-
-			try {
-				writeAuthJson({
-					anthropic: { type: "api_key", key: bracedKey },
-				});
-
-				authStorage = AuthStorage.create(authJsonPath);
-				const apiKey = await authStorage.getApiKey("anthropic");
-
-				expect(apiKey).toBe("braced-env-api-key-value");
-			} finally {
-				if (originalEnv === undefined) {
-					delete process.env.TEST_AUTH_BRACED_API_KEY_12345;
-				} else {
-					process.env.TEST_AUTH_BRACED_API_KEY_12345 = originalEnv;
-				}
-			}
-		});
-
-		test("apiKey interpolates braced env references inside literals", async () => {
-			const originalPartA = process.env.TEST_AUTH_INTERPOLATED_PART_A_12345;
-			const originalPartB = process.env.TEST_AUTH_INTERPOLATED_PART_B_12345;
-			process.env.TEST_AUTH_INTERPOLATED_PART_A_12345 = "left";
-			process.env.TEST_AUTH_INTERPOLATED_PART_B_12345 = "right";
-			const interpolatedKey = [
-				"$",
-				"{TEST_AUTH_INTERPOLATED_PART_A_12345}_$",
-				"{TEST_AUTH_INTERPOLATED_PART_B_12345}",
-			].join("");
-
-			try {
-				writeAuthJson({
-					anthropic: { type: "api_key", key: interpolatedKey },
-				});
-
-				authStorage = AuthStorage.create(authJsonPath);
-				const apiKey = await authStorage.getApiKey("anthropic");
-
-				expect(apiKey).toBe("left_right");
-			} finally {
-				if (originalPartA === undefined) {
-					delete process.env.TEST_AUTH_INTERPOLATED_PART_A_12345;
-				} else {
-					process.env.TEST_AUTH_INTERPOLATED_PART_A_12345 = originalPartA;
-				}
-				if (originalPartB === undefined) {
-					delete process.env.TEST_AUTH_INTERPOLATED_PART_B_12345;
-				} else {
-					process.env.TEST_AUTH_INTERPOLATED_PART_B_12345 = originalPartB;
-				}
-			}
-		});
-
-		test("apiKey with $$ prefix escapes a leading dollar", async () => {
-			writeAuthJson({
-				anthropic: { type: "api_key", key: "$$TEST_AUTH_API_KEY_12345" },
-			});
-
-			authStorage = AuthStorage.create(authJsonPath);
-			const apiKey = await authStorage.getApiKey("anthropic");
-
-			expect(apiKey).toBe("$TEST_AUTH_API_KEY_12345");
-		});
-
-		test("apiKey with $! escapes a literal bang and still interpolates later env refs", async () => {
-			const originalEnv = process.env.TEST_AUTH_API_KEY_12345;
-			process.env.TEST_AUTH_API_KEY_12345 = "env-api-key-value";
-
-			try {
-				writeAuthJson({
-					anthropic: { type: "api_key", key: "$!literal-$TEST_AUTH_API_KEY_12345" },
-				});
-
-				authStorage = AuthStorage.create(authJsonPath);
-				const apiKey = await authStorage.getApiKey("anthropic");
-
-				expect(apiKey).toBe("!literal-env-api-key-value");
-			} finally {
-				if (originalEnv === undefined) {
-					delete process.env.TEST_AUTH_API_KEY_12345;
-				} else {
-					process.env.TEST_AUTH_API_KEY_12345 = originalEnv;
-				}
-			}
-		});
-
-		test("plain API key is used directly even when it matches an env var", async () => {
-			const originalEnv = process.env.TEST_AUTH_API_KEY_12345;
-			process.env.TEST_AUTH_API_KEY_12345 = "env-api-key-value";
-
-			try {
-				writeAuthJson({
-					anthropic: { type: "api_key", key: "TEST_AUTH_API_KEY_12345" },
-				});
-
-				authStorage = AuthStorage.create(authJsonPath);
-				const apiKey = await authStorage.getApiKey("anthropic");
-
-				expect(apiKey).toBe("TEST_AUTH_API_KEY_12345");
-			} finally {
-				if (originalEnv === undefined) {
-					delete process.env.TEST_AUTH_API_KEY_12345;
-				} else {
-					process.env.TEST_AUTH_API_KEY_12345 = originalEnv;
-				}
-			}
-		});
-
-		test("literal public API key is not corrupted by the Windows PUBLIC env var", async () => {
-			const originalPublic = process.env.PUBLIC;
-			process.env.PUBLIC = "C:\\Users\\Public";
-
-			try {
-				writeAuthJson({
-					opencode: { type: "api_key", key: "public" },
-				});
-
-				authStorage = AuthStorage.create(authJsonPath);
-				const apiKey = await authStorage.getApiKey("opencode");
-
-				expect(apiKey).toBe("public");
-			} finally {
-				if (originalPublic === undefined) {
-					delete process.env.PUBLIC;
-				} else {
-					process.env.PUBLIC = originalPublic;
-				}
-			}
-		});
-
-		test("apiKey as literal value is used directly when not an env var", async () => {
-			// Make sure this isn't an env var
-			delete process.env.literal_api_key_value;
-
-			writeAuthJson({
-				anthropic: { type: "api_key", key: "literal_api_key_value" },
-			});
-
-			authStorage = AuthStorage.create(authJsonPath);
-			const apiKey = await authStorage.getApiKey("anthropic");
-
-			expect(apiKey).toBe("literal_api_key_value");
-		});
-
-		test("apiKey command can use shell features like pipes", async () => {
-			writeAuthJson({
-				anthropic: { type: "api_key", key: "!echo 'hello world' | tr ' ' '-'" },
-			});
-
-			authStorage = AuthStorage.create(authJsonPath);
-			const apiKey = await authStorage.getApiKey("anthropic");
-
-			expect(apiKey).toBe("hello-world");
-		});
-
-		test("command config uses stdin when configured shell requires it", () => {
-			if (process.platform === "win32") return;
-			const platformDescriptor = Object.getOwnPropertyDescriptor(process, "platform");
-			vi.spyOn(shellModule, "getShellConfig").mockReturnValue({
-				shell: "/bin/bash",
-				args: ["-s"],
-				commandTransport: "stdin",
-			});
-
-			try {
-				Object.defineProperty(process, "platform", {
-					configurable: true,
-					value: "win32",
-				});
-				const nameExpansion = "$" + "{name}";
-
-				expect(resolveConfigValueUncached(`!name='World'; echo "Hello, ${nameExpansion}!"`)).toBe("Hello, World!");
-			} finally {
-				if (platformDescriptor) {
-					Object.defineProperty(process, "platform", platformDescriptor);
-				}
-			}
-		});
-
-		describe("caching", () => {
-			test("command is only executed once per process", async () => {
-				// Use a command that writes to a file to count invocations
-				const counterFile = join(tempDir, "counter");
-				writeFileSync(counterFile, "0");
-
-				const counterPath = toShPath(counterFile);
-				const command = `!sh -c 'count=$(cat "${counterPath}"); echo $((count + 1)) > "${counterPath}"; echo "key-value"'`;
-				writeAuthJson({
-					anthropic: { type: "api_key", key: command },
-				});
-
-				authStorage = AuthStorage.create(authJsonPath);
-
-				// Call multiple times
-				await authStorage.getApiKey("anthropic");
-				await authStorage.getApiKey("anthropic");
-				await authStorage.getApiKey("anthropic");
-
-				// Command should have only run once
-				const count = parseInt(readFileSync(counterFile, "utf-8").trim(), 10);
-				expect(count).toBe(1);
-			});
-
-			test("cache persists across AuthStorage instances", async () => {
-				const counterFile = join(tempDir, "counter");
-				writeFileSync(counterFile, "0");
-
-				const counterPath = toShPath(counterFile);
-				const command = `!sh -c 'count=$(cat "${counterPath}"); echo $((count + 1)) > "${counterPath}"; echo "key-value"'`;
-				writeAuthJson({
-					anthropic: { type: "api_key", key: command },
-				});
-
-				// Create multiple AuthStorage instances
-				const storage1 = AuthStorage.create(authJsonPath);
-				await storage1.getApiKey("anthropic");
-
-				const storage2 = AuthStorage.create(authJsonPath);
-				await storage2.getApiKey("anthropic");
-
-				// Command should still have only run once
-				const count = parseInt(readFileSync(counterFile, "utf-8").trim(), 10);
-				expect(count).toBe(1);
-			});
-
-			test("clearConfigValueCache allows command to run again", async () => {
-				const counterFile = join(tempDir, "counter");
-				writeFileSync(counterFile, "0");
-
-				const counterPath = toShPath(counterFile);
-				const command = `!sh -c 'count=$(cat "${counterPath}"); echo $((count + 1)) > "${counterPath}"; echo "key-value"'`;
-				writeAuthJson({
-					anthropic: { type: "api_key", key: command },
-				});
-
-				authStorage = AuthStorage.create(authJsonPath);
-				await authStorage.getApiKey("anthropic");
-
-				// Clear cache and call again
-				clearConfigValueCache();
-				await authStorage.getApiKey("anthropic");
-
-				// Command should have run twice
-				const count = parseInt(readFileSync(counterFile, "utf-8").trim(), 10);
-				expect(count).toBe(2);
-			});
-
-			test("different commands are cached separately", async () => {
-				writeAuthJson({
-					anthropic: { type: "api_key", key: "!echo key-anthropic" },
-					openai: { type: "api_key", key: "!echo key-openai" },
-				});
-
-				authStorage = AuthStorage.create(authJsonPath);
-
-				const keyA = await authStorage.getApiKey("anthropic");
-				const keyB = await authStorage.getApiKey("openai");
-
-				expect(keyA).toBe("key-anthropic");
-				expect(keyB).toBe("key-openai");
-			});
-
-			test("failed commands are cached (not retried)", async () => {
-				const counterFile = join(tempDir, "counter");
-				writeFileSync(counterFile, "0");
-
-				const counterPath = toShPath(counterFile);
-				const command = `!sh -c 'count=$(cat "${counterPath}"); echo $((count + 1)) > "${counterPath}"; exit 1'`;
-				writeAuthJson({
-					anthropic: { type: "api_key", key: command },
-				});
-
-				authStorage = AuthStorage.create(authJsonPath);
-
-				// Call multiple times - all should return undefined
-				const key1 = await authStorage.getApiKey("anthropic");
-				const key2 = await authStorage.getApiKey("anthropic");
-
-				expect(key1).toBeUndefined();
-				expect(key2).toBeUndefined();
-
-				// Command should have only run once despite failures
-				const count = parseInt(readFileSync(counterFile, "utf-8").trim(), 10);
-				expect(count).toBe(1);
-			});
-
-			test("environment variables are not cached (changes are picked up)", async () => {
-				const envVarName = "TEST_AUTH_KEY_CACHE_TEST_98765";
-				const originalEnv = process.env[envVarName];
-
-				try {
-					process.env[envVarName] = "first-value";
-
-					writeAuthJson({
-						anthropic: { type: "api_key", key: `$${envVarName}` },
-					});
-
-					authStorage = AuthStorage.create(authJsonPath);
-
-					const key1 = await authStorage.getApiKey("anthropic");
-					expect(key1).toBe("first-value");
-
-					// Change env var
-					process.env[envVarName] = "second-value";
-
-					const key2 = await authStorage.getApiKey("anthropic");
-					expect(key2).toBe("second-value");
-				} finally {
-					if (originalEnv === undefined) {
-						delete process.env[envVarName];
-					} else {
-						process.env[envVarName] = originalEnv;
-					}
-				}
-			});
-		});
-	});
-
-	describe("oauth lock compromise handling", () => {
-		test("returns undefined on compromised lock and allows a later retry", async () => {
-			const providerId = `test-oauth-provider-${Date.now()}-${Math.random().toString(36).slice(2)}`;
-			registerOAuthProvider({
-				id: providerId,
-				name: "Test OAuth Provider",
-				async login() {
-					throw new Error("Not used in this test");
-				},
-				async refreshToken(credentials) {
-					return {
-						...credentials,
-						access: "refreshed-access-token",
+					refresh: async (credential) => ({
+						...credential,
+						access: "refreshed-access",
 						expires: Date.now() + 60_000,
-					};
+					}),
+					toAuth: async (credential) => ({ apiKey: credential.access }),
 				},
-				getApiKey(credentials) {
-					return `Bearer ${credentials.access}`;
-				},
-			});
+			},
+			getModels: () => [],
+			stream: () => {
+				throw new Error("not used");
+			},
+			streamSimple: () => {
+				throw new Error("not used");
+			},
+		};
+		const models = createModels({ credentials: storage });
+		models.setProvider(provider);
 
-			writeAuthJson({
-				[providerId]: {
-					type: "oauth",
-					refresh: "refresh-token",
-					access: "expired-access-token",
-					expires: Date.now() - 10_000,
-				},
-			});
-
-			authStorage = AuthStorage.create(authJsonPath);
-
-			const realLock = lockfile.lock.bind(lockfile);
-			const lockSpy = vi.spyOn(lockfile, "lock");
-			lockSpy.mockImplementationOnce(async (file, options) => {
-				options?.onCompromised?.(new Error("Unable to update lock within the stale threshold"));
-				return realLock(file, options);
-			});
-
-			const firstTry = await authStorage.getApiKey(providerId);
-			expect(firstTry).toBeUndefined();
-
-			lockSpy.mockRestore();
-
-			const secondTry = await authStorage.getApiKey(providerId);
-			expect(secondTry).toBe("Bearer refreshed-access-token");
+		const realLock = lockfile.lock.bind(lockfile);
+		const lockSpy = vi.spyOn(lockfile, "lock").mockImplementationOnce(async (file, options) => {
+			options?.onCompromised?.(new Error("lock compromised"));
+			return realLock(file, options);
 		});
+		await expect(models.getAuth(providerId)).rejects.toMatchObject({ code: "auth" });
+
+		lockSpy.mockRestore();
+		await expect(models.getAuth(providerId)).resolves.toMatchObject({ auth: { apiKey: "refreshed-access" } });
 	});
 
-	describe("persistence semantics", () => {
-		test("set preserves unrelated external edits", () => {
-			writeAuthJson({
-				anthropic: { type: "api_key", key: "old-anthropic" },
-				openai: { type: "api_key", key: "openai-key" },
-			});
-
-			authStorage = AuthStorage.create(authJsonPath);
-
-			// Simulate external edit while process is running
-			writeAuthJson({
-				anthropic: { type: "api_key", key: "old-anthropic" },
-				openai: { type: "api_key", key: "openai-key" },
-				google: { type: "api_key", key: "google-key" },
-			});
-
-			authStorage.set("anthropic", { type: "api_key", key: "new-anthropic" });
-
-			const updated = JSON.parse(readFileSync(authJsonPath, "utf-8")) as Record<string, { key: string }>;
-			expect(updated.anthropic.key).toBe("new-anthropic");
-			expect(updated.openai.key).toBe("openai-key");
-			expect(updated.google.key).toBe("google-key");
-		});
-
-		test("remove preserves unrelated external edits", () => {
-			writeAuthJson({
-				anthropic: { type: "api_key", key: "anthropic-key" },
-				openai: { type: "api_key", key: "openai-key" },
-			});
-
-			authStorage = AuthStorage.create(authJsonPath);
-
-			// Simulate external edit while process is running
-			writeAuthJson({
-				anthropic: { type: "api_key", key: "anthropic-key" },
-				openai: { type: "api_key", key: "openai-key" },
-				google: { type: "api_key", key: "google-key" },
-			});
-
-			authStorage.remove("anthropic");
-
-			const updated = JSON.parse(readFileSync(authJsonPath, "utf-8")) as Record<string, { key: string }>;
-			expect(updated.anthropic).toBeUndefined();
-			expect(updated.openai.key).toBe("openai-key");
-			expect(updated.google.key).toBe("google-key");
-		});
-
-		test("throws and does not overwrite malformed auth file after load error", () => {
-			writeAuthJson({
-				anthropic: { type: "api_key", key: "anthropic-key" },
-			});
-
-			authStorage = AuthStorage.create(authJsonPath);
-			writeFileSync(authJsonPath, "{invalid-json", "utf-8");
-
-			authStorage.reload();
-			expect(() => authStorage.set("openai", { type: "api_key", key: "openai-key" })).toThrow(
-				"Cannot update auth storage because it could not be loaded",
-			);
-
-			const raw = readFileSync(authJsonPath, "utf-8");
-			expect(raw).toBe("{invalid-json");
-			expect(authStorage.has("openai")).toBe(false);
-		});
-
-		test("throws when a stale auth lock prevents persistence", () => {
-			writeAuthJson({});
-			writeFileSync(`${authJsonPath}.lock`, "", "utf-8");
-
-			authStorage = AuthStorage.create(authJsonPath);
-			expect(() => authStorage.set("github-copilot", { type: "api_key", key: "copilot-key" })).toThrow(
-				"Cannot update auth storage because it could not be loaded",
-			);
-
-			expect(readFileSync(authJsonPath, "utf-8")).toBe("{}");
-			expect(authStorage.has("github-copilot")).toBe(false);
-		});
-
-		test("recovers from an earlier load error before persisting", () => {
-			writeAuthJson({});
-			const lockPath = `${authJsonPath}.lock`;
-			writeFileSync(lockPath, "", "utf-8");
-
-			authStorage = AuthStorage.create(authJsonPath);
-			rmSync(lockPath);
-			authStorage.set("github-copilot", { type: "api_key", key: "copilot-key" });
-
-			const updated = JSON.parse(readFileSync(authJsonPath, "utf-8")) as Record<string, { key: string }>;
-			expect(updated["github-copilot"].key).toBe("copilot-key");
-			expect(authStorage.has("github-copilot")).toBe(true);
-		});
-
-		test("reload records parse errors and drainErrors clears buffer", () => {
-			writeAuthJson({
-				anthropic: { type: "api_key", key: "anthropic-key" },
-			});
-
-			authStorage = AuthStorage.create(authJsonPath);
-			writeFileSync(authJsonPath, "{invalid-json", "utf-8");
-
-			authStorage.reload();
-
-			// Keeps previous in-memory data on reload failure
-			expect(authStorage.get("anthropic")).toEqual({ type: "api_key", key: "anthropic-key" });
-
-			const firstDrain = authStorage.drainErrors();
-			expect(firstDrain.length).toBeGreaterThan(0);
-			expect(firstDrain[0]).toBeInstanceOf(Error);
-
-			const secondDrain = authStorage.drainErrors();
-			expect(secondDrain).toHaveLength(0);
-		});
-	});
-
-	describe("auth status", () => {
-		test("does not expose stored API keys or OAuth tokens", () => {
-			authStorage = AuthStorage.inMemory({
-				anthropic: { type: "api_key", key: "secret-api-key" },
-				openai: {
-					type: "oauth",
-					access: "secret-access-token",
-					refresh: "secret-refresh-token",
-					expires: Date.now() + 1000,
-				},
-			});
-
-			expect(authStorage.getAuthStatus("anthropic")).toEqual({ configured: true, source: "stored" });
-			expect(authStorage.getAuthStatus("openai")).toEqual({ configured: true, source: "stored" });
-			expect(JSON.stringify(authStorage.getAuthStatus("anthropic"))).not.toContain("secret-api-key");
-			expect(JSON.stringify(authStorage.getAuthStatus("openai"))).not.toContain("secret-access-token");
-			expect(JSON.stringify(authStorage.getAuthStatus("openai"))).not.toContain("secret-refresh-token");
-		});
-	});
-
-	describe("runtime overrides", () => {
-		test("runtime override takes priority over auth.json", async () => {
-			writeAuthJson({
-				anthropic: { type: "api_key", key: "!echo stored-key" },
-			});
-
-			authStorage = AuthStorage.create(authJsonPath);
-			authStorage.setRuntimeApiKey("anthropic", "runtime-key");
-
-			const apiKey = await authStorage.getApiKey("anthropic");
-
-			expect(apiKey).toBe("runtime-key");
-		});
-
-		test("removing runtime override falls back to auth.json", async () => {
-			writeAuthJson({
-				anthropic: { type: "api_key", key: "!echo stored-key" },
-			});
-
-			authStorage = AuthStorage.create(authJsonPath);
-			authStorage.setRuntimeApiKey("anthropic", "runtime-key");
-			authStorage.removeRuntimeApiKey("anthropic");
-
-			const apiKey = await authStorage.getApiKey("anthropic");
-
-			expect(apiKey).toBe("stored-key");
-		});
+	test("does not overwrite malformed auth files", async () => {
+		writeAuthJson({ anthropic: { type: "api_key", key: "stored" } });
+		const storage = AuthStorage.create(authJsonPath);
+		writeFileSync(authJsonPath, "{invalid-json", "utf8");
+		await expect(storage.modify("openai", async () => ({ type: "api_key", key: "new" }))).rejects.toThrow();
+		expect(readFileSync(authJsonPath, "utf8")).toBe("{invalid-json");
 	});
 });

packages/coding-agent/test/cache-stats.test.ts

@@ -12,7 +12,7 @@ const zeroCost = { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 };
 
 const models: ModelPriceSource = {
 	// $/million tokens; used as cache-read price fallback on full-miss turns
-	find: () => ({ cost: { cacheRead: 0.3 } }),
+	getModel: () => ({ cost: { cacheRead: 0.3 } }),
 };
 
 function assistant(options: {

packages/coding-agent/test/compaction-extensions.test.ts

@@ -1,3 +1,4 @@
+import { createModelRegistry, getModelRuntime } from "./model-runtime-test-utils.ts";
 /**
  * Tests for compaction extension events (before_compact / compact).
  */
@@ -17,7 +18,6 @@ import {
 	type SessionCompactEvent,
 	type SessionEvent,
 } from "../src/core/extensions/index.ts";
-import { ModelRegistry } from "../src/core/model-registry.ts";
 import { SessionManager } from "../src/core/session-manager.ts";
 import { SettingsManager } from "../src/core/settings-manager.ts";
 import { createSyntheticSourceInfo } from "../src/core/source-info.ts";
@@ -31,7 +31,7 @@ describe.skipIf(!API_KEY)("Compaction extensions", () => {
 	let tempDir: string;
 	let capturedEvents: SessionEvent[];
 
-	beforeEach(() => {
+	beforeEach(async () => {
 		tempDir = join(tmpdir(), `pi-compaction-extensions-test-${Date.now()}`);
 		mkdirSync(tempDir, { recursive: true });
 		capturedEvents = [];
@@ -85,7 +85,7 @@ describe.skipIf(!API_KEY)("Compaction extensions", () => {
 		};
 	}
 
-	function createSession(extensions: Extension[]) {
+	async function createSession(extensions: Extension[]) {
 		const model = getModel("anthropic", "claude-sonnet-4-5")!;
 		const agent = new Agent({
 			getApiKey: () => API_KEY,
@@ -100,7 +100,7 @@ describe.skipIf(!API_KEY)("Compaction extensions", () => {
 		const settingsManager = SettingsManager.create(tempDir, tempDir);
 		settingsManager.applyOverrides({ compaction: { keepRecentTokens: 1 } });
 		const authStorage = AuthStorage.create(join(tempDir, "auth.json"));
-		const modelRegistry = ModelRegistry.create(authStorage);
+		const modelRegistry = await createModelRegistry(authStorage);
 
 		const runtime = createExtensionRuntime();
 		const resourceLoader = {
@@ -113,7 +113,7 @@ describe.skipIf(!API_KEY)("Compaction extensions", () => {
 			sessionManager,
 			settingsManager,
 			cwd: tempDir,
-			modelRegistry,
+			modelRuntime: getModelRuntime(modelRegistry),
 			resourceLoader,
 		});
 
@@ -122,7 +122,7 @@ describe.skipIf(!API_KEY)("Compaction extensions", () => {
 
 	it("should emit before_compact and compact events", async () => {
 		const extension = createExtension();
-		createSession([extension]);
+		await createSession([extension]);
 
 		await session.prompt("What is 2+2? Reply with just the number.");
 		await session.agent.waitForIdle();
@@ -158,7 +158,7 @@ describe.skipIf(!API_KEY)("Compaction extensions", () => {
 
 	it("should allow extensions to cancel compaction", async () => {
 		const extension = createExtension(() => ({ cancel: true }));
-		createSession([extension]);
+		await createSession([extension]);
 
 		await session.prompt("What is 2+2? Reply with just the number.");
 		await session.agent.waitForIdle();
@@ -184,7 +184,7 @@ describe.skipIf(!API_KEY)("Compaction extensions", () => {
 			}
 			return undefined;
 		});
-		createSession([extension]);
+		await createSession([extension]);
 
 		await session.prompt("What is 2+2? Reply with just the number.");
 		await session.agent.waitForIdle();
@@ -208,7 +208,7 @@ describe.skipIf(!API_KEY)("Compaction extensions", () => {
 
 	it("should include entries in compact event after compaction is saved", async () => {
 		const extension = createExtension();
-		createSession([extension]);
+		await createSession([extension]);
 
 		await session.prompt("What is 2+2? Reply with just the number.");
 		await session.agent.waitForIdle();
@@ -259,7 +259,7 @@ describe.skipIf(!API_KEY)("Compaction extensions", () => {
 			shortcuts: new Map(),
 		};
 
-		createSession([throwingExtension]);
+		await createSession([throwingExtension]);
 
 		await session.prompt("What is 2+2? Reply with just the number.");
 		await session.agent.waitForIdle();
@@ -339,7 +339,7 @@ describe.skipIf(!API_KEY)("Compaction extensions", () => {
 			shortcuts: new Map(),
 		};
 
-		createSession([extension1, extension2]);
+		await createSession([extension1, extension2]);
 
 		await session.prompt("What is 2+2? Reply with just the number.");
 		await session.agent.waitForIdle();
@@ -356,7 +356,7 @@ describe.skipIf(!API_KEY)("Compaction extensions", () => {
 			capturedBeforeEvent = event;
 			return undefined;
 		});
-		createSession([extension]);
+		await createSession([extension]);
 
 		await session.prompt("What is 2+2? Reply with just the number.");
 		await session.agent.waitForIdle();
@@ -378,10 +378,9 @@ describe.skipIf(!API_KEY)("Compaction extensions", () => {
 
 		expect(Array.isArray(event.branchEntries)).toBe(true);
 
-		// sessionManager, modelRegistry, and model are now on ctx, not event
-		// Verify they're accessible via session
+		// sessionManager and model runtime remain available on the session.
 		expect(typeof session.sessionManager.getEntries).toBe("function");
-		expect(typeof session.modelRegistry.getApiKeyAndHeaders).toBe("function");
+		expect(typeof session.modelRuntime.getAuth).toBe("function");
 
 		const entries = session.sessionManager.getEntries();
 		expect(Array.isArray(entries)).toBe(true);
@@ -403,7 +402,7 @@ describe.skipIf(!API_KEY)("Compaction extensions", () => {
 			}
 			return undefined;
 		});
-		createSession([extension]);
+		await createSession([extension]);
 
 		await session.prompt("What is 2+2? Reply with just the number.");
 		await session.agent.waitForIdle();