Merge main into model-registry

.github/APPROVED_CONTRIBUTORS

@@ -237,3 +237,9 @@ davidlifschitz pr
 vdxz pr
 
 dangooddd pr
+
+Mearman pr
+
+dodiego pr
+
+any-victor pr

.github/ISSUE_TEMPLATE/bug.yml

@@ -11,6 +11,8 @@ body:
 
         Keep this short. If it doesn't fit on one screen, it's too long. Write in your own voice.
 
+        **Important:** before reporting an issue in core, please validate first with `pi -ne` that this is not caused by an extension you loaded.
+
   - type: textarea
     id: description
     attributes:

.github/ISSUE_TEMPLATE/package-report.yml

@@ -0,0 +1,49 @@
+name: Package Report
+description: Report a problematic Pi package listed on pi.dev
+labels: ["package-report"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Use this form to report a package listed on pi.dev. For Pi core bugs, use the bug report template instead.
+
+        New issues from new contributors are auto-closed by default. Maintainers review auto-closed issues daily. Issues that do not meet the quality bar in [CONTRIBUTING.md](https://github.com/earendil-works/pi-mono/blob/main/CONTRIBUTING.md) will not be reopened or receive a reply.
+
+        Keep this short. If it doesn't fit on one screen, it's too long. Write in your own voice.
+
+  - type: input
+    id: package-name
+    attributes:
+      label: Package name
+      description: The npm package name from pi.dev.
+      placeholder: "@scope/package"
+    validations:
+      required: true
+
+  - type: input
+    id: package-version
+    attributes:
+      label: Version
+      description: The package version shown on pi.dev.
+      placeholder: "0.1.0"
+    validations:
+      required: false
+
+  - type: dropdown
+    id: report-type
+    attributes:
+      label: What are you reporting?
+      options:
+        - Malicious or unsafe behavior
+        - Impersonation
+        - Trademark / TOS Violations
+    validations:
+      required: true
+
+  - type: textarea
+    id: details
+    attributes:
+      label: Details
+      description: Describe the concern and include links, logs, or screenshots if helpful.
+    validations:
+      required: true

.github/workflows/build-binaries.yml

@@ -59,28 +59,27 @@ jobs:
         run: |
           cd packages/coding-agent/binaries
 
+          release_assets=(
+            pi-darwin-arm64.tar.gz
+            pi-darwin-x64.tar.gz
+            pi-linux-x64.tar.gz
+            pi-linux-arm64.tar.gz
+            pi-windows-x64.zip
+            pi-windows-arm64.zip
+          )
+          sha256sum "${release_assets[@]}" > SHA256SUMS
+          release_assets+=(SHA256SUMS)
+
           if gh release view "${RELEASE_TAG}" >/dev/null 2>&1; then
             gh release edit "${RELEASE_TAG}" \
               --title "${RELEASE_TAG}" \
               --notes-file /tmp/release-notes.md
-            gh release upload "${RELEASE_TAG}" \
-              pi-darwin-arm64.tar.gz \
-              pi-darwin-x64.tar.gz \
-              pi-linux-x64.tar.gz \
-              pi-linux-arm64.tar.gz \
-              pi-windows-x64.zip \
-              pi-windows-arm64.zip \
-              --clobber
+            gh release upload "${RELEASE_TAG}" "${release_assets[@]}" --clobber
           else
             gh release create "${RELEASE_TAG}" \
               --title "${RELEASE_TAG}" \
               --notes-file /tmp/release-notes.md \
-              pi-darwin-arm64.tar.gz \
-              pi-darwin-x64.tar.gz \
-              pi-linux-x64.tar.gz \
-              pi-linux-arm64.tar.gz \
-              pi-windows-x64.zip \
-              pi-windows-arm64.zip
+              "${release_assets[@]}"
           fi
 
   publish-npm:

.github/workflows/issue-gate.yml

@@ -111,9 +111,17 @@ jobs:
               body: message,
             });
 
+            await github.rest.issues.addLabels({
+              owner: context.repo.owner,
+              repo: context.repo.repo,
+              issue_number: context.issue.number,
+              labels: ['untriaged'],
+            });
+
             await github.rest.issues.update({
               owner: context.repo.owner,
               repo: context.repo.repo,
               issue_number: context.issue.number,
               state: 'closed',
+              state_reason: 'not_planned',
             });

.github/workflows/issue-triage-labels.yml

@@ -0,0 +1,142 @@
+name: Issue Triage Labels
+
+on:
+  issues:
+    types: [reopened, labeled]
+
+jobs:
+  update-labels:
+    runs-on: ubuntu-latest
+    permissions:
+      issues: write
+    steps:
+      - name: Update triage labels
+        uses: actions/github-script@v7
+        with:
+          script: |
+            const UNTRIAGED_LABEL = 'untriaged';
+            const NO_ACTION_LABEL = 'no-action';
+            const LAST_READ_LABEL = 'last-read';
+            const TO_DISCUSS_LABEL = 'to-discuss';
+            const INPROGRESS_LABEL = 'inprogress';
+
+            function issueHasLabel(issue, labelName) {
+              return (issue.labels ?? []).some((label) => label.name === labelName);
+            }
+
+            async function removeLabelIfPresent(issueNumber, issue, labelName) {
+              if (!issueHasLabel(issue, labelName)) {
+                console.log(`Issue #${issueNumber} does not have ${labelName}`);
+                return;
+              }
+
+              try {
+                await github.rest.issues.removeLabel({
+                  owner: context.repo.owner,
+                  repo: context.repo.repo,
+                  issue_number: issueNumber,
+                  name: labelName,
+                });
+                console.log(`Removed ${labelName} from #${issueNumber}`);
+              } catch (error) {
+                if (error.status === 404) {
+                  console.log(`Label ${labelName} was already absent from #${issueNumber}`);
+                  return;
+                }
+                throw error;
+              }
+            }
+
+            if (context.payload.action === 'reopened') {
+              await removeLabelIfPresent(context.issue.number, context.payload.issue, UNTRIAGED_LABEL);
+              await removeLabelIfPresent(context.issue.number, context.payload.issue, NO_ACTION_LABEL);
+              return;
+            }
+
+            if (context.payload.action === 'labeled' && context.payload.label?.name === NO_ACTION_LABEL) {
+              await removeLabelIfPresent(context.issue.number, context.payload.issue, UNTRIAGED_LABEL);
+              return;
+            }
+
+            if (context.payload.action !== 'labeled' || context.payload.label?.name !== LAST_READ_LABEL) {
+              console.log('Not a last-read label event');
+              return;
+            }
+
+            const currentIssueNumber = context.issue.number;
+            const lastReadIssues = await github.paginate(github.rest.issues.listForRepo, {
+              owner: context.repo.owner,
+              repo: context.repo.repo,
+              state: 'all',
+              labels: LAST_READ_LABEL,
+              per_page: 100,
+            });
+
+            const previousIssueNumbers = lastReadIssues
+              .filter((issue) => !issue.pull_request)
+              .map((issue) => issue.number)
+              .filter((issueNumber) => issueNumber !== currentIssueNumber);
+
+            if (previousIssueNumbers.length === 0) {
+              console.log('No previous last-read issue found');
+              return;
+            }
+
+            const previousIssueNumber = Math.max(...previousIssueNumbers);
+            if (currentIssueNumber <= previousIssueNumber) {
+              console.log(
+                `Last-read was added to old issue #${currentIssueNumber}; latest last-read is #${previousIssueNumber}`,
+              );
+              return;
+            }
+
+            const untriagedIssues = await github.paginate(github.rest.issues.listForRepo, {
+              owner: context.repo.owner,
+              repo: context.repo.repo,
+              state: 'all',
+              labels: UNTRIAGED_LABEL,
+              per_page: 100,
+            });
+
+            const issuesToMark = untriagedIssues
+              .filter((issue) => !issue.pull_request)
+              .filter((issue) => issue.number >= previousIssueNumber && issue.number <= currentIssueNumber)
+              .sort((a, b) => a.number - b.number);
+
+            if (issuesToMark.length === 0) {
+              console.log(`No untriaged issues found from #${previousIssueNumber} to #${currentIssueNumber}`);
+              return;
+            }
+
+            for (const issue of issuesToMark) {
+              if (issueHasLabel(issue, TO_DISCUSS_LABEL)) {
+                console.log(`Skipped ${NO_ACTION_LABEL} for #${issue.number} because it has ${TO_DISCUSS_LABEL}`);
+              } else {
+                await github.rest.issues.addLabels({
+                  owner: context.repo.owner,
+                  repo: context.repo.repo,
+                  issue_number: issue.number,
+                  labels: [NO_ACTION_LABEL],
+                });
+                console.log(`Added ${NO_ACTION_LABEL} to #${issue.number}`);
+              }
+
+              await github.rest.issues.update({
+                owner: context.repo.owner,
+                repo: context.repo.repo,
+                issue_number: issue.number,
+                state: 'closed',
+                state_reason: 'not_planned',
+              });
+              console.log(`Closed #${issue.number} as not planned`);
+
+              await removeLabelIfPresent(issue.number, issue, INPROGRESS_LABEL);
+
+              await github.rest.issues.removeLabel({
+                owner: context.repo.owner,
+                repo: context.repo.repo,
+                issue_number: issue.number,
+                name: UNTRIAGED_LABEL,
+              });
+              console.log(`Removed ${UNTRIAGED_LABEL} from #${issue.number}`);
+            }

.github/workflows/remove-inprogress-on-close.yml

@@ -0,0 +1,31 @@
+name: Remove In Progress Label On Close
+
+on:
+  issues:
+    types: [closed]
+
+jobs:
+  remove-label:
+    runs-on: ubuntu-latest
+    permissions:
+      issues: write
+    steps:
+      - name: Remove inprogress label
+        uses: actions/github-script@v7
+        with:
+          script: |
+            const labelName = 'inprogress';
+            const labels = context.payload.issue.labels ?? [];
+            const hasLabel = labels.some((label) => label.name === labelName);
+
+            if (!hasLabel) {
+              console.log(`Issue does not have ${labelName} label`);
+              return;
+            }
+
+            await github.rest.issues.removeLabel({
+              owner: context.repo.owner,
+              repo: context.repo.repo,
+              issue_number: context.issue.number,
+              name: labelName,
+            });

CONTRIBUTING.md

@@ -2,19 +2,27 @@
 
 This guide exists to save both sides time.
 
+## Philosophy
+
+First things first: **pi's core is minimal**.
+
+If your feature does not belong in the core, it should be an extension. PRs that bloat the core will likely be rejected.
+
+Pi's core exists to be minimal and to be extensible so that it can be influenced and manipulated by extensions.  Even hook points for extensions however should be well considered and discussed to avoid adding unmaintainable bloat and complex interactions.
+
 ## The One Rule
 
 **You must understand your code.** If you cannot explain what your changes do and how they interact with the rest of the system, your PR will be closed.
 
 Using AI to write code is fine. Submitting AI-generated slop without understanding it is not.
 
-If you use an agent, run it from the `pi-mono` root directory so it picks up `AGENTS.md` automatically. Your agent must follow the rules and guidelines in that file.
+If you use an agent, run it from the `pi` root directory so it picks up `AGENTS.md` automatically. Your agent must follow the rules and guidelines in that file.
 
 ## Contribution Gate
 
 All issues and PRs from new contributors are auto-closed by default.
 
-Issues submitted Friday through Sunday are not reviewed. If something is urgent, ask on Discord: https://discord.com/invite/3cU7Bz4UPx
+Issues submitted Friday through Sunday are not guaranteed to be reviewed.  If something is urgent, ask on Discord: https://discord.com/invite/3cU7Bz4UPx
 
 Maintainers review auto-closed issues daily and reopen worthwhile ones. Issues that do not meet the quality bar below will not be reopened or receive a reply.
 
@@ -32,7 +40,7 @@ If you open an issue, you must use one of the two GitHub issue templates.
 If you open an issue, keep it short, concrete, and worth reading.
 
 - Keep it concise. If it does not fit on one screen, it is too long.
-- Write in your own voice.
+- Write in your own voice (do not use an LLM to generate text, if you must, follow up with a clearly AI labeled comment).
 - State the bug or request clearly.
 - Explain why it matters.
 - If you want to implement the change yourself, say so.
@@ -62,10 +70,6 @@ Do not edit `CHANGELOG.md`. Changelog entries are added by maintainers.
 
 If you are adding a new provider to `packages/ai`, see `AGENTS.md` for required tests.
 
-## Philosophy
-
-pi's core is minimal. If your feature does not belong in the core, it should be an extension. PRs that bloat the core will likely be rejected.
-
 ## Questions?
 
 Ask on [Discord](https://discord.com/invite/nKXTsAcmbT).
@@ -76,9 +80,9 @@ Ask on [Discord](https://discord.com/invite/nKXTsAcmbT).
 
 pi receives more issues than the maintainers can responsibly review in real time. Many reports do not meet the quality bar in this guide or do not follow CONTRIBUTING.md. Some are slung at the repository mindlessly via an agent instead of being reviewed and shaped by the person submitting them. Auto-closing creates a buffer so maintainers can review the tracker on their own schedule and reopen the issues that meet the quality bar.
 
-### Why are weekend issues not reviewed?
+### Why are weekend issues lower priority?
 
-Maintainers need uninterrupted time away from the issue tracker. Issues submitted Friday through Sunday are auto-closed and are not part of the Monday review queue. If a problem is urgent, ask on Discord and include the short version, a repro, and the relevant logs.
+We triage the tracker during working hours. That means more issues can accumulate over the weekend. Anything submitted Friday through Sunday may be missed or given lower priority in the Monday review queue. If a problem is urgent, ask on Discord and include the short version, a repro, and the relevant logs.
 
 ### Why do some issues get no reply?
 
@@ -91,3 +95,8 @@ AI can help group duplicates, summarize reports, and spot missing information. I
 ### Is this hostile to contributors?
 
 No. It is a guardrail against burnout and tracker spam. Short, concrete, reproducible issues are welcome. Thoughtful contributions are welcome. Automated slop, entitlement, and large volumes of low-effort reports are not.
+
+## Where can I learn about plans?
+
+Earendil uses RFCs to discuss larger changes.  Not all of them are public, but
+quite a few are.  They can be found at [rfc.earendil.com](https://rfc.earendil.com/keyword/pi/).

README.md

@@ -5,46 +5,24 @@
 </p>
 <p align="center">
   <a href="https://discord.com/invite/3cU7Bz4UPx"><img alt="Discord" src="https://img.shields.io/badge/discord-community-5865F2?style=flat-square&logo=discord&logoColor=white" /></a>
-</p>
-<p align="center">
-  <a href="https://pi.dev">pi.dev</a> domain graciously donated by
-  <br /><br />
-  <a href="https://exe.dev"><img src="packages/coding-agent/docs/images/exy.png" alt="Exy mascot" width="48" /><br />exe.dev</a>
+  <a href="https://www.npmjs.com/package/@earendil-works/pi-coding-agent"><img alt="npm" src="https://img.shields.io/npm/v/@earendil-works/pi-coding-agent?style=flat-square" /></a>
 </p>
 
 > New issues and PRs from new contributors are auto-closed by default. Maintainers review auto-closed issues daily. See [CONTRIBUTING.md](CONTRIBUTING.md).
 
----
+# Pi Agent Harness
 
-# Pi Agent Harness Mono Repo
-
-This is the home of the pi agent harness project including our self extensible coding agent.
+This is the home of the Pi agent harness project including our self extensible coding agent.
 
 * **[@earendil-works/pi-coding-agent](packages/coding-agent)**: Interactive coding agent CLI
 * **[@earendil-works/pi-agent-core](packages/agent)**: Agent runtime with tool calling and state management
 * **[@earendil-works/pi-ai](packages/ai)**: Unified multi-provider LLM API (OpenAI, Anthropic, Google, …)
 
-To learn more about pi:
+To learn more about Pi:
 
 * [Visit pi.dev](https://pi.dev), the project website with demos
 * [Read the documentation](https://pi.dev/docs/latest), but you can also ask the agent to explain itself
 
-## Share your OSS coding agent sessions
-
-If you use pi or other coding agents for open source work, please share your sessions.
-
-Public OSS session data helps improve coding agents with real-world tasks, tool use, failures, and fixes instead of toy benchmarks.
-
-For the full explanation, see [this post on X](https://x.com/badlogicgames/status/2037811643774652911).
-
-To publish sessions, use [`badlogic/pi-share-hf`](https://github.com/badlogic/pi-share-hf). Read its README.md for setup instructions. All you need is a Hugging Face account, the Hugging Face CLI, and `pi-share-hf`.
-
-You can also watch [this video](https://x.com/badlogicgames/status/2041151967695634619), where I show how I publish my `pi-mono` sessions.
-
-I regularly publish my own `pi-mono` work sessions here:
-
-- [badlogicgames/pi-mono on Hugging Face](https://huggingface.co/datasets/badlogicgames/pi-mono)
-
 ## All Packages
 
 | Package | Description |
@@ -62,13 +40,13 @@ Pi does not include a built-in permission system for restricting filesystem, pro
 
 If you need stronger boundaries, containerize or sandbox Pi. See [packages/coding-agent/docs/containerization.md](packages/coding-agent/docs/containerization.md) for three patterns:
 
-- **OpenShell**: run the whole `pi` process in a policy-controlled sandbox.
 - **Gondolin extension**: keep `pi` and provider auth on the host while routing built-in tools and `!` commands into a local Linux micro-VM.
 - **Plain Docker**: run the whole `pi` process in a local container for simple isolation.
+- **OpenShell**: run the whole `pi` process in a policy-controlled sandbox.
 
 ## Contributing
 
-See [CONTRIBUTING.md](CONTRIBUTING.md) for contribution guidelines and [AGENTS.md](AGENTS.md) for project-specific rules (for both humans and agents).
+See [CONTRIBUTING.md](CONTRIBUTING.md) for contribution guidelines and [AGENTS.md](AGENTS.md) for project-specific rules (for both humans and agents).  Longer term plans for Pi can also be found in [RFCs](https://rfc.earendil.com/keyword/pi/).
 
 ## Development
 
@@ -94,6 +72,28 @@ We treat npm dependency changes as reviewed code changes.
 - CI installs with `npm ci --ignore-scripts`, and a scheduled GitHub workflow runs `npm audit --omit=dev` plus `npm audit signatures --omit=dev`.
 - Shrinkwrap generation has an explicit allowlist for dependency lifecycle scripts; new lifecycle-script deps fail checks until reviewed.
 
+## Share your OSS coding agent sessions
+
+If you use Pi or other coding agents for open source work, please share your sessions.
+
+Public OSS session data helps improve coding agents with real-world tasks, tool use, failures, and fixes instead of toy benchmarks.
+
+For the full explanation, see [this post on X](https://x.com/badlogicgames/status/2037811643774652911).
+
+To publish sessions, use [`badlogic/pi-share-hf`](https://github.com/badlogic/pi-share-hf). Read its README.md for setup instructions. All you need is a Hugging Face account, the Hugging Face CLI, and `pi-share-hf`.
+
+You can also watch [this video](https://x.com/badlogicgames/status/2041151967695634619), where I show how I publish my `pi-mono` sessions.
+
+I regularly publish my own `pi-mono` work sessions here:
+
+- [badlogicgames/pi-mono on Hugging Face](https://huggingface.co/datasets/badlogicgames/pi-mono)
+
 ## License
 
 MIT
+
+<p align="center">
+  <a href="https://pi.dev">pi.dev</a> domain graciously donated by
+  <br /><br />
+  <a href="https://exe.dev"><img src="packages/coding-agent/docs/images/exy.png" alt="Exy mascot" width="48" /><br />exe.dev</a>
+</p>

package.json

@@ -41,7 +41,7 @@
 		"@biomejs/biome": "2.3.5",
 		"@types/node": "22.19.19",
 		"@typescript/native-preview": "7.0.0-dev.20260120.1",
-		"esbuild": "0.28.0",
+		"esbuild": "0.28.1",
 		"husky": "9.1.7",
 		"jiti": "2.7.0",
 		"shx": "0.4.0",

packages/agent/CHANGELOG.md

@@ -8,6 +8,36 @@
 - `compact()`, `generateSummary()`, and `generateBranchSummary()` take a `Models` parameter and no longer accept explicit `apiKey`/`headers`.
 - `StreamFn` is defined structurally (`(model, context, options?) => AssistantMessageEventStream | Promise<...>`); `Models.streamSimple` satisfies it.
 
+## [0.79.10] - 2026-06-22
+
+## [0.79.9] - 2026-06-20
+
+### Fixed
+
+- Fixed Node execution environment commands through legacy WSL `bash.exe` to pass scripts over stdin so shell variables expand in the target bash ([#5893](https://github.com/earendil-works/pi/issues/5893)).
+
+## [0.79.8] - 2026-06-19
+
+### Added
+
+- Added `@earendil-works/pi-agent-core/base` for bundlers that want to pair the agent core with selective `@earendil-works/pi-ai/base` provider registration ([#5348](https://github.com/earendil-works/pi/pull/5348) by [@FredKSchott](https://github.com/FredKSchott)).
+
+## [0.79.7] - 2026-06-18
+
+## [0.79.6] - 2026-06-16
+
+## [0.79.5] - 2026-06-16
+
+## [0.79.4] - 2026-06-15
+
+## [0.79.3] - 2026-06-13
+
+## [0.79.2] - 2026-06-12
+
+### Fixed
+
+- Fixed late tool progress callbacks after tool settlement to be ignored instead of emitting stale `tool_execution_update` events ([#5573](https://github.com/earendil-works/pi/issues/5573)).
+
 ## [0.79.1] - 2026-06-09
 
 ## [0.79.0] - 2026-06-08

packages/agent/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@earendil-works/pi-agent-core",
-	"version": "0.79.1",
+	"version": "0.79.10",
 	"description": "General-purpose agent with transport abstraction, state management, and attachment support",
 	"type": "module",
 	"main": "./dist/index.js",
@@ -29,7 +29,7 @@
 		"prepublishOnly": "npm run clean && npm run build"
 	},
 	"dependencies": {
-		"@earendil-works/pi-ai": "^0.79.1",
+		"@earendil-works/pi-ai": "^0.79.10",
 		"ignore": "7.0.5",
 		"typebox": "1.1.38",
 		"yaml": "2.9.0"
@@ -53,8 +53,8 @@
 	},
 	"devDependencies": {
 		"@types/node": "24.12.4",
-		"@vitest/coverage-v8": "3.2.4",
+		"@vitest/coverage-v8": "4.1.9",
 		"typescript": "5.9.3",
-		"vitest": "3.2.4"
+		"vitest": "4.1.9"
 	}
 }

packages/agent/src/agent-loop.ts

@@ -631,6 +631,7 @@ async function executePreparedToolCall(
 	emit: AgentEventSink,
 ): Promise<ExecutedToolCallOutcome> {
 	const updateEvents: Promise<void>[] = [];
+	let acceptingUpdates = true;
 
 	try {
 		const result = await prepared.tool.execute(
@@ -638,6 +639,7 @@ async function executePreparedToolCall(
 			prepared.args as never,
 			signal,
 			(partialResult) => {
+				if (!acceptingUpdates) return;
 				updateEvents.push(
 					Promise.resolve(
 						emit({
@@ -651,14 +653,18 @@ async function executePreparedToolCall(
 				);
 			},
 		);
+		acceptingUpdates = false;
 		await Promise.all(updateEvents);
 		return { result, isError: false };
 	} catch (error) {
+		acceptingUpdates = false;
 		await Promise.all(updateEvents);
 		return {
 			result: createErrorToolResult(error instanceof Error ? error.message : String(error)),
 			isError: true,
 		};
+	} finally {
+		acceptingUpdates = false;
 	}
 }
 

packages/agent/src/harness/env/nodejs.ts

@@ -144,12 +144,25 @@ async function findBashOnPath(): Promise<string | null> {
 	return firstMatch && (await pathExists(firstMatch)) ? firstMatch : null;
 }
 
-async function getShellConfig(
-	customShellPath?: string,
-): Promise<Result<{ shell: string; args: string[] }, ExecutionError>> {
+interface ShellConfig {
+	shell: string;
+	args: string[];
+	commandTransport?: "argv" | "stdin";
+}
+
+function isLegacyWslBashPath(path: string): boolean {
+	const normalized = path.replace(/\//g, "\\").toLowerCase();
+	return /^[a-z]:\\windows\\(?:system32|sysnative)\\bash\.exe$/.test(normalized);
+}
+
+function getBashShellConfig(shell: string): ShellConfig {
+	return isLegacyWslBashPath(shell) ? { shell, args: ["-s"], commandTransport: "stdin" } : { shell, args: ["-c"] };
+}
+
+async function getShellConfig(customShellPath?: string): Promise<Result<ShellConfig, ExecutionError>> {
 	if (customShellPath) {
 		if (await pathExists(customShellPath)) {
-			return ok({ shell: customShellPath, args: ["-c"] });
+			return ok(getBashShellConfig(customShellPath));
 		}
 		return err(new ExecutionError("shell_unavailable", `Custom shell path not found: ${customShellPath}`));
 	}
@@ -161,22 +174,22 @@ async function getShellConfig(
 		if (programFilesX86) candidates.push(`${programFilesX86}\\Git\\bin\\bash.exe`);
 		for (const candidate of candidates) {
 			if (await pathExists(candidate)) {
-				return ok({ shell: candidate, args: ["-c"] });
+				return ok(getBashShellConfig(candidate));
 			}
 		}
 		const bashOnPath = await findBashOnPath();
 		if (bashOnPath) {
-			return ok({ shell: bashOnPath, args: ["-c"] });
+			return ok(getBashShellConfig(bashOnPath));
 		}
 		return err(new ExecutionError("shell_unavailable", "No bash shell found"));
 	}
 
 	if (await pathExists("/bin/bash")) {
-		return ok({ shell: "/bin/bash", args: ["-c"] });
+		return ok(getBashShellConfig("/bin/bash"));
 	}
 	const bashOnPath = await findBashOnPath();
 	if (bashOnPath) {
-		return ok({ shell: bashOnPath, args: ["-c"] });
+		return ok(getBashShellConfig(bashOnPath));
 	}
 	return ok({ shell: "sh", args: ["-c"] });
 }
@@ -274,13 +287,22 @@ export class NodeExecutionEnv implements ExecutionEnv {
 			};
 
 			try {
-				child = spawn(shellConfig.value.shell, [...shellConfig.value.args, command], {
+				const commandFromStdin = shellConfig.value.commandTransport === "stdin";
+				child = spawn(
+					shellConfig.value.shell,
+					commandFromStdin ? shellConfig.value.args : [...shellConfig.value.args, command],
+					{
 						cwd,
 						detached: process.platform !== "win32",
 						env: getShellEnv(this.shellEnv, options?.env),
-					stdio: ["ignore", "pipe", "pipe"],
+						stdio: [commandFromStdin ? "pipe" : "ignore", "pipe", "pipe"],
 						windowsHide: true,
-				});
+					},
+				);
+				if (commandFromStdin) {
+					child.stdin?.on("error", () => {});
+					child.stdin?.end(command);
+				}
 			} catch (error) {
 				const cause = toError(error);
 				settle(err(new ExecutionError("spawn_error", cause.message, cause)));

packages/agent/src/types.ts

@@ -359,7 +359,12 @@ export interface AgentToolResult<T> {
 	terminate?: boolean;
 }
 
-/** Callback used by tools to stream partial execution updates. */
+/**
+ * Callback used by tools to stream partial execution updates.
+ *
+ * The callback is scoped to the current `execute()` invocation. Calls made after
+ * the tool promise settles are ignored.
+ */
 export type AgentToolUpdateCallback<T = any> = (partialResult: AgentToolResult<T>) => void;
 
 /** Tool definition used by the agent runtime. */

packages/agent/test/agent.test.ts

@@ -1,6 +1,7 @@
 import { type AssistantMessage, type AssistantMessageEvent, EventStream, getModel } from "@earendil-works/pi-ai/compat";
+import { Type } from "typebox";
 import { describe, expect, it } from "vitest";
-import { Agent } from "../src/index.ts";
+import { Agent, type AgentEvent, type AgentTool, type AgentToolUpdateCallback } from "../src/index.ts";
 
 // Mock stream that mimics AssistantMessageEventStream
 class MockAssistantStream extends EventStream<AssistantMessageEvent, AssistantMessage> {
@@ -36,6 +37,28 @@ function createAssistantMessage(text: string): AssistantMessage {
 	};
 }
 
+type ToolCallContent = Extract<AssistantMessage["content"][number], { type: "toolCall" }>;
+
+function createAssistantToolUseMessage(content: ToolCallContent[]): AssistantMessage {
+	return {
+		role: "assistant",
+		content,
+		api: "openai-responses",
+		provider: "openai",
+		model: "mock",
+		usage: {
+			input: 0,
+			output: 0,
+			cacheRead: 0,
+			cacheWrite: 0,
+			totalTokens: 0,
+			cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
+		},
+		stopReason: "toolUse",
+		timestamp: Date.now(),
+	};
+}
+
 function createDeferred(): {
 	promise: Promise<void>;
 	resolve: () => void;
@@ -242,6 +265,147 @@ describe("Agent", () => {
 		expect(receivedSignal?.aborted).toBe(true);
 	});
 
+	it("should ignore tool updates after the tool execution settles", async () => {
+		const toolSchema = Type.Object({});
+		let delayedUpdate: AgentToolUpdateCallback<{ status: string }> | undefined;
+		const events: AgentEvent[] = [];
+		const unhandledRejections: unknown[] = [];
+		const onUnhandledRejection = (error: unknown) => {
+			unhandledRejections.push(error);
+		};
+		const tool: AgentTool<typeof toolSchema, { status: string }> = {
+			name: "delayed_tool",
+			label: "Delayed Tool",
+			description: "Captures progress callbacks",
+			parameters: toolSchema,
+			async execute(_toolCallId, _params, _signal, onUpdate) {
+				delayedUpdate = onUpdate;
+				onUpdate?.({
+					content: [{ type: "text", text: "running" }],
+					details: { status: "running" },
+				});
+				return {
+					content: [{ type: "text", text: "ok" }],
+					details: { status: "done" },
+					terminate: true,
+				};
+			},
+		};
+		const agent = new Agent({
+			initialState: { tools: [tool] },
+			streamFn: () => {
+				const stream = new MockAssistantStream();
+				queueMicrotask(() => {
+					stream.push({
+						type: "done",
+						reason: "toolUse",
+						message: createAssistantToolUseMessage([
+							{ type: "toolCall", id: "call-1", name: "delayed_tool", arguments: {} },
+						]),
+					});
+				});
+				return stream;
+			},
+		});
+		agent.subscribe((event) => {
+			events.push(event);
+		});
+
+		process.on("unhandledRejection", onUnhandledRejection);
+		try {
+			await agent.prompt("run tool");
+			const eventCountAfterPrompt = events.length;
+
+			delayedUpdate?.({
+				content: [{ type: "text", text: "late" }],
+				details: { status: "late" },
+			});
+			await new Promise((resolve) => setTimeout(resolve, 0));
+
+			expect(events.filter((event) => event.type === "tool_execution_update")).toHaveLength(1);
+			expect(events).toHaveLength(eventCountAfterPrompt);
+			expect(unhandledRejections).toEqual([]);
+		} finally {
+			process.off("unhandledRejection", onUnhandledRejection);
+		}
+	});
+
+	it("should ignore a settled parallel tool update while another tool is still running", async () => {
+		const toolSchema = Type.Object({});
+		const slowStarted = createDeferred();
+		const settledToolEnded = createDeferred();
+		const releaseSlow = createDeferred();
+		let settledToolUpdate: AgentToolUpdateCallback<{ status: string }> | undefined;
+		const events: AgentEvent[] = [];
+		const settledTool: AgentTool<typeof toolSchema, { status: string }> = {
+			name: "settled_tool",
+			label: "Settled Tool",
+			description: "Captures progress callbacks",
+			parameters: toolSchema,
+			async execute(_toolCallId, _params, _signal, onUpdate) {
+				settledToolUpdate = onUpdate;
+				return {
+					content: [{ type: "text", text: "done" }],
+					details: { status: "done" },
+					terminate: true,
+				};
+			},
+		};
+		const slowTool: AgentTool<typeof toolSchema, { status: string }> = {
+			name: "slow_tool",
+			label: "Slow Tool",
+			description: "Keeps the agent run active",
+			parameters: toolSchema,
+			async execute() {
+				slowStarted.resolve();
+				await releaseSlow.promise;
+				return {
+					content: [{ type: "text", text: "done" }],
+					details: { status: "done" },
+					terminate: true,
+				};
+			},
+		};
+		const agent = new Agent({
+			initialState: { tools: [settledTool, slowTool] },
+			streamFn: () => {
+				const stream = new MockAssistantStream();
+				queueMicrotask(() => {
+					stream.push({
+						type: "done",
+						reason: "toolUse",
+						message: createAssistantToolUseMessage([
+							{ type: "toolCall", id: "call-1", name: "settled_tool", arguments: {} },
+							{ type: "toolCall", id: "call-2", name: "slow_tool", arguments: {} },
+						]),
+					});
+				});
+				return stream;
+			},
+		});
+		agent.subscribe((event) => {
+			events.push(event);
+			if (event.type === "tool_execution_end" && event.toolCallId === "call-1") {
+				settledToolEnded.resolve();
+			}
+		});
+
+		const promptPromise = agent.prompt("run tools");
+		await Promise.all([slowStarted.promise, settledToolEnded.promise]);
+		const eventCountBeforeLateUpdate = events.length;
+
+		settledToolUpdate?.({
+			content: [{ type: "text", text: "late" }],
+			details: { status: "late" },
+		});
+		await new Promise((resolve) => setTimeout(resolve, 0));
+		expect(events).toHaveLength(eventCountBeforeLateUpdate);
+
+		releaseSlow.resolve();
+		await promptPromise;
+		expect(events.filter((event) => event.type === "tool_execution_update")).toHaveLength(0);
+	});
+
 	it("should update state with mutators", () => {
 		const agent = new Agent();
 

packages/agent/test/harness/nodejs-env.test.ts

@@ -1,5 +1,5 @@
 import { access, chmod, realpath, symlink } from "node:fs/promises";
-import { join } from "node:path";
+import { delimiter, join } from "node:path";
 import { afterEach, describe, expect, it } from "vitest";
 import { NodeExecutionEnv } from "../../src/harness/env/nodejs.ts";
 import { FileError, getOrThrow } from "../../src/harness/types.ts";
@@ -201,6 +201,39 @@ describe("NodeExecutionEnv", () => {
 		expect(result).toEqual({ stdout: `${await realpath(root)}:ok`, stderr: "", exitCode: 0 });
 	});
 
+	it("uses stdin command transport for legacy WSL bash paths", async () => {
+		if (process.platform === "win32") return;
+		const root = createTempDir();
+		const shellPath = "C:\\Windows\\System32\\bash.exe";
+		const env = new NodeExecutionEnv({ cwd: root });
+		getOrThrow(await env.writeFile(shellPath, '#!/bin/sh\nprintf \'args:%s\\n\' "$*" >&2\nexec /bin/bash "$@"\n'));
+		await chmod(join(root, shellPath), 0o755);
+
+		const originalCwd = process.cwd();
+		const originalPath = process.env.PATH;
+		const platformDescriptor = Object.getOwnPropertyDescriptor(process, "platform");
+		try {
+			process.chdir(root);
+			process.env.PATH = `${root}${delimiter}${originalPath ?? ""}`;
+			Object.defineProperty(process, "platform", {
+				configurable: true,
+				value: "win32",
+			});
+
+			const wslEnv = new NodeExecutionEnv({ cwd: root, shellPath });
+			const nameExpansion = "$" + "{name}";
+			const result = getOrThrow(await wslEnv.exec(`name='World'; echo "Hello, ${nameExpansion}!"`));
+
+			expect(result).toEqual({ stdout: "Hello, World!\n", stderr: "args:-s\n", exitCode: 0 });
+		} finally {
+			process.chdir(originalCwd);
+			process.env.PATH = originalPath;
+			if (platformDescriptor) {
+				Object.defineProperty(process, "platform", platformDescriptor);
+			}
+		}
+	});
+
 	it("streams stdout and stderr chunks", async () => {
 		const root = createTempDir();
 		const env = new NodeExecutionEnv({ cwd: root });

packages/ai/CHANGELOG.md

@@ -23,6 +23,86 @@
 
 - Fixed Claude Fable 5 thinking-off requests to omit Anthropic's unsupported `thinking.type: "disabled"` payload ([#5567](https://github.com/earendil-works/pi/pull/5567) by [@tmustier](https://github.com/tmustier)).
 
+## [0.79.10] - 2026-06-22
+
+### Fixed
+
+- Fixed OpenAI-compatible streaming to preserve encrypted `reasoning_details` that arrive before matching tool call deltas ([#5114](https://github.com/earendil-works/pi/issues/5114)).
+
+## [0.79.9] - 2026-06-20
+
+### Added
+
+- Added configurable `chat-template` thinking support for OpenAI-compatible providers that use `chat_template_kwargs`, such as DeepSeek models behind vLLM ([#5673](https://github.com/earendil-works/pi/issues/5673)).
+
+### Fixed
+
+- Fixed Fireworks GLM-5.2 metadata to use the OpenAI-compatible Chat Completions endpoint with `reasoning_effort` support ([#5923](https://github.com/earendil-works/pi/issues/5923)).
+- Fixed OpenRouter GLM-5.2 metadata to expose `xhigh` reasoning and send OpenRouter's native `xhigh` effort ([#5770](https://github.com/earendil-works/pi/issues/5770)).
+- Fixed GitHub Copilot OAuth model availability to use the authenticated account's model picker catalog ([#5897](https://github.com/earendil-works/pi/issues/5897)).
+
+## [0.79.8] - 2026-06-19
+
+### Added
+
+- Added `@earendil-works/pi-ai/base` and direct provider registration exports for bundlers that want selective provider transports without root built-in registration ([#5348](https://github.com/earendil-works/pi/pull/5348) by [@FredKSchott](https://github.com/FredKSchott)).
+- Added prompt caching for Mistral requests using the pi session ID as `prompt_cache_key`, including cached-token usage and cost accounting ([#5854](https://github.com/earendil-works/pi/issues/5854)).
+- Added the OpenRouter Fusion alias as `openrouter/fusion` ([#5866](https://github.com/earendil-works/pi/pull/5866) by [@dannote](https://github.com/dannote)).
+
+## [0.79.7] - 2026-06-18
+
+### Added
+
+- Added GLM-5.2 model to the OpenCode Go subscription model catalog ([#5860](https://github.com/earendil-works/pi/issues/5860)).
+
+## [0.79.6] - 2026-06-16
+
+### Fixed
+
+- Fixed OpenCode Go DeepSeek V4 thinking-off requests to send the provider's `thinking: { type: "disabled" }` compatibility parameter.
+
+## [0.79.5] - 2026-06-16
+
+### Added
+
+- Added provider-scoped `StreamOptions.env` overrides for provider configuration, including Cloudflare endpoint placeholders, Azure OpenAI, Google Vertex, Amazon Bedrock, cache retention, and proxy environment lookups ([#5728](https://github.com/earendil-works/pi/issues/5728)).
+
+### Fixed
+
+- Fixed OpenAI Responses streaming to tolerate null message content from OpenAI-compatible servers before tool calls ([#5819](https://github.com/earendil-works/pi/issues/5819)).
+- Fixed OpenCode DeepSeek V4 thinking requests to avoid sending both `thinking` and `reasoning_effort` ([#5818](https://github.com/earendil-works/pi/issues/5818)).
+- Fixed Z.AI GLM-5.2 thinking requests to send `reasoning_effort` with the provider's `high`/`max` effort mapping ([#5770](https://github.com/earendil-works/pi/issues/5770)).
+- Fixed Google and `google-vertex` Gemini model metadata to map `latest` aliases to the current models, add Gemini 3.5 Flash for Vertex, correct Gemini 2.5 Flash Vertex cache pricing, and remove shut-down Vertex preview models ([#5761](https://github.com/earendil-works/pi/issues/5761)).
+- Fixed Moonshot AI China model metadata to include Kimi K2.7 Code, and omitted unsupported thinking-off payloads for Kimi K2.7 Code models ([#5760](https://github.com/earendil-works/pi/issues/5760)).
+
+## [0.79.4] - 2026-06-15
+
+### Fixed
+
+- Fixed Anthropic 1-hour prompt-cache write cost accounting to price 1-hour cache writes at 2x input instead of the 5-minute cache-write rate ([#5738](https://github.com/earendil-works/pi/pull/5738) by [@theBucky](https://github.com/theBucky)).
+- Fixed GitHub Copilot Claude adaptive-thinking effort metadata to match manually checked Copilot model capabilities ([#4637](https://github.com/earendil-works/pi/issues/4637)).
+- Fixed OpenCode/OpenCode Go completion models that reject `prompt_cache_retention` to omit long-retention cache fields when `cacheRetention` is `long` ([#5702](https://github.com/earendil-works/pi/issues/5702)).
+
+## [0.79.3] - 2026-06-13
+
+### Fixed
+
+- Restored OpenAI GPT-5.4/GPT-5.5 and OpenAI Codex GPT-5.4/GPT-5.4 mini/GPT-5.5 context window metadata to the observed 272k-token Codex backend limit, avoiding a billing hazard from sending prompts above Codex's accepted limit (reported by [@trethore](https://github.com/trethore)).
+
+## [0.79.2] - 2026-06-12
+
+### Added
+
+- Added AWS data retention documentation links to Amazon Bedrock unsupported data retention mode validation errors ([#5561](https://github.com/earendil-works/pi/pull/5561) by [@unexge](https://github.com/unexge)).
+
+### Fixed
+
+- Fixed OpenAI-compatible context overflow detection for parenthesized `maximum context length (N)` errors ([#5677](https://github.com/earendil-works/pi/issues/5677)).
+- Fixed OpenAI GPT-5.4/GPT-5.5 and OpenAI Codex GPT-5.4/GPT-5.4 mini/GPT-5.5 context window metadata to match current OpenAI limits ([#5644](https://github.com/earendil-works/pi/issues/5644)).
+- Increased the OpenAI Codex Responses SSE response-header timeout to 20 seconds to reduce false-positive stalls while retaining the bounded wait introduced for zero-event hangs ([#4945](https://github.com/earendil-works/pi/issues/4945)).
+- Fixed Anthropic refusal stops to preserve provider `stop_details` explanations in error messages ([#5666](https://github.com/earendil-works/pi/pull/5666) by [@rwachtler](https://github.com/rwachtler)).
+- Fixed Claude Fable 5 thinking-off requests to omit Anthropic's unsupported `thinking.type: "disabled"` payload ([#5567](https://github.com/earendil-works/pi/pull/5567) by [@tmustier](https://github.com/tmustier)).
+
 ## [0.79.1] - 2026-06-09
 
 ### Added

packages/ai/README.md

@@ -1055,7 +1055,8 @@ interface OpenAICompletionsCompat {
   requiresAssistantAfterToolResult?: boolean; // Whether tool results must be followed by an assistant message (default: false)
   requiresThinkingAsText?: boolean;  // Whether thinking blocks must be converted to text (default: false)
   requiresReasoningContentOnAssistantMessages?: boolean; // Whether all replayed assistant messages must include empty reasoning_content when reasoning is enabled (default: auto-detected for DeepSeek)
-  thinkingFormat?: 'openai' | 'openrouter' | 'deepseek' | 'together' | 'zai' | 'qwen' | 'qwen-chat-template' | 'string-thinking' | 'ant-ling'; // Format for reasoning param: 'openai' uses reasoning_effort, 'openrouter' uses reasoning: { effort }, 'deepseek' uses thinking: { type } plus reasoning_effort when supported, 'together' uses reasoning: { enabled } plus reasoning_effort when supported, 'zai' uses enable_thinking, 'qwen' uses enable_thinking, 'qwen-chat-template' uses chat_template_kwargs.enable_thinking, 'string-thinking' uses top-level thinking, 'ant-ling' uses reasoning: { effort } only for mapped efforts (default: openai)
+  thinkingFormat?: 'openai' | 'openrouter' | 'deepseek' | 'together' | 'zai' | 'qwen' | 'chat-template' | 'qwen-chat-template' | 'string-thinking' | 'ant-ling'; // Format for reasoning param: 'openai' uses reasoning_effort, 'openrouter' uses reasoning: { effort }, 'deepseek' uses thinking: { type } plus reasoning_effort when supported, 'together' uses reasoning: { enabled } plus reasoning_effort when supported, 'zai' uses thinking: { type }, 'qwen' uses enable_thinking, 'chat-template' uses configurable chat_template_kwargs, 'qwen-chat-template' uses chat_template_kwargs.enable_thinking and preserve_thinking, 'string-thinking' uses top-level thinking, 'ant-ling' uses reasoning: { effort } only for mapped efforts (default: openai)
+  chatTemplateKwargs?: Record<string, string | number | boolean | null | { '$var': 'thinking.enabled' | 'thinking.effort'; omitWhenOff?: boolean }>; // chat_template_kwargs values; use $var for pi-controlled thinking values
   cacheControlFormat?: 'anthropic';  // Anthropic-style cache_control on system prompt, last tool, and last user/assistant text content
   openRouterRouting?: OpenRouterRouting; // OpenRouter routing preferences (default: {})
   vercelGatewayRouting?: VercelGatewayRouting; // Vercel AI Gateway routing preferences (default: {})
@@ -1265,6 +1266,25 @@ Browser compatibility notes:
 - OAuth login flows are Node-only. They are lazy-loaded behind bundler-opaque imports, so registering an OAuth-capable provider does not pull Node-only code into a browser bundle — only actually logging in would.
 - Use a server-side proxy or backend service if you need Bedrock or OAuth-based auth from a web app.
 
+### Provider-Scoped Environment Overrides
+
+Pass `env` in stream options to scope provider configuration to a request. Values in `env` are used before process environment variables for provider auth and configuration such as Cloudflare account IDs, Azure OpenAI settings, Vertex project/location, Bedrock settings, `PI_CACHE_RETENTION`, and `HTTP_PROXY`/`HTTPS_PROXY`.
+
+```typescript
+const models = builtinModels();
+const model = models.getModel('cloudflare-ai-gateway', 'workers-ai/@cf/moonshotai/kimi-k2.6')!;
+
+const response = await models.complete(model, context, {
+  env: {
+    CLOUDFLARE_API_KEY: '...',
+    CLOUDFLARE_ACCOUNT_ID: 'account-id',
+    CLOUDFLARE_GATEWAY_ID: 'gateway-id'
+  }
+});
+```
+
+Use this when one process needs different provider settings per request, or when ambient environment variables should not leak into a provider call.
+
 ## OAuth Providers
 
 Several providers support OAuth authentication instead of static API keys:

packages/ai/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@earendil-works/pi-ai",
-	"version": "0.79.1",
+	"version": "0.79.10",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"type": "module",
 	"main": "./dist/index.js",
@@ -86,9 +86,10 @@
 	"dependencies": {
 		"@anthropic-ai/sdk": "0.91.1",
 		"@aws-sdk/client-bedrock-runtime": "3.1048.0",
-		"@smithy/node-http-handler": "4.7.3",
 		"@google/genai": "1.52.0",
-		"@mistralai/mistralai": "2.2.1",
+		"@mistralai/mistralai": "2.2.6",
+		"@opentelemetry/api": "1.9.0",
+		"@smithy/node-http-handler": "4.7.3",
 		"http-proxy-agent": "7.0.2",
 		"https-proxy-agent": "7.0.6",
 		"openai": "6.26.0",
@@ -118,6 +119,6 @@
 	"devDependencies": {
 		"@types/node": "24.12.4",
 		"canvas": "3.2.3",
-		"vitest": "3.2.4"
+		"vitest": "4.1.9"
 	}
 }

packages/ai/scripts/generate-models.ts

@@ -68,6 +68,8 @@ const KIMI_STATIC_HEADERS = {
 	"User-Agent": "KimiCLI/1.5",
 } as const;
 
+const MOONSHOT_CN_MIRRORED_MODEL_IDS = new Set(["kimi-k2.7-code", "kimi-k2.7-code-highspeed"]);
+
 const TOGETHER_BASE_URL = "https://api.together.ai/v1";
 const TOGETHER_BASE_COMPAT: OpenAICompletionsCompat = {
 	supportsStore: false,
@@ -121,6 +123,7 @@ const TOGETHER_TOGGLE_REASONING_LEVEL_MAP = {
 
 const AI_GATEWAY_MODELS_URL = "https://ai-gateway.vercel.sh/v1";
 const AI_GATEWAY_BASE_URL = "https://ai-gateway.vercel.sh";
+const VERTEX_BASE_URL = "https://{location}-aiplatform.googleapis.com";
 const NVIDIA_BASE_URL = "https://integrate.api.nvidia.com/v1";
 const NVIDIA_HEADERS = {
 	"NVCF-POLL-SECONDS": "3600",
@@ -154,6 +157,13 @@ const NVIDIA_NIM_UNSUPPORTED_MODELS = new Set([
 	"upstage/solar-10.7b-instruct",
 ]);
 const ZAI_TOOL_STREAM_UNSUPPORTED_MODELS = new Set(["glm-4.5", "glm-4.5-air", "glm-4.5-flash", "glm-4.5v"]);
+const ZAI_GLM52_THINKING_LEVEL_MAP = {
+	minimal: null,
+	low: "high",
+	medium: "high",
+	high: "high",
+	xhigh: "max",
+} as const;
 const EAGER_TOOL_INPUT_STREAMING_UNSUPPORTED_ANTHROPIC_MODELS = new Set([
 	"github-copilot:claude-haiku-4.5",
 	"github-copilot:claude-sonnet-4",
@@ -187,6 +197,23 @@ const OPENAI_RESPONSES_NONE_REASONING_MODELS = new Set([
 	"gpt-5.5",
 ]);
 
+const OPENCODE_OPENAI_COMPLETIONS_LONG_CACHE_RETENTION_UNSUPPORTED_MODELS = new Set([
+	"opencode:deepseek-v4-flash",
+	"opencode:deepseek-v4-pro",
+	"opencode:kimi-k2.5",
+	"opencode:kimi-k2.6",
+	"opencode:minimax-m2.7",
+	"opencode-go:kimi-k2.6",
+]);
+
+// Checked manually against the authenticated GitHub Copilot /models endpoint on 2026-06-15.
+// Keep this to narrow corrections over models.dev metadata instead of snapshotting Copilot's catalog.
+const GITHUB_COPILOT_THINKING_LEVEL_OVERRIDES = {
+	"claude-opus-4.7": { minimal: "low" },
+	"claude-opus-4.8": { minimal: "low" },
+	"claude-sonnet-4.6": { minimal: "low", xhigh: "max" },
+} satisfies Record<string, NonNullable<Model<Api>["thinkingLevelMap"]>>;
+
 function mergeThinkingLevelMap(model: Model<any>, map: NonNullable<Model<any>["thinkingLevelMap"]>): void {
 	model.thinkingLevelMap = { ...model.thinkingLevelMap, ...map };
 }
@@ -251,7 +278,8 @@ function isGemini3ProModel(modelId: string): boolean {
 }
 
 function isGemini3FlashModel(modelId: string): boolean {
-	return /gemini-3(?:\.\d+)?-flash/.test(modelId.toLowerCase());
+	const id = modelId.toLowerCase();
+	return /gemini-3(?:\.\d+)?-flash/.test(id) || id === "gemini-flash-latest" || id === "gemini-flash-lite-latest";
 }
 
 function isGemma4Model(modelId: string): boolean {
@@ -330,6 +358,15 @@ function applyThinkingLevelMetadata(model: Model<any>): void {
 	if (model.provider === "openai-codex" && supportsOpenAiXhigh(model.id)) {
 		mergeThinkingLevelMap(model, { minimal: "low" });
 	}
+	if (
+		(model.provider === "moonshotai" || model.provider === "moonshotai-cn") &&
+		(model.id === "kimi-k2.7-code" || model.id === "kimi-k2.7-code-highspeed")
+	) {
+		// Kimi K2.7 Code is always-thinking. Official docs say
+		// `thinking: { type: "disabled" }` is rejected, and callers can omit
+		// the thinking parameter to use the enabled default.
+		mergeThinkingLevelMap(model, { off: null });
+	}
 	if (model.provider === "openrouter" && model.id.startsWith("inception/mercury-2")) {
 		// Mercury 2 in instant mode (reasoning_effort: "none") disables tool calling.
 		// Mark "off" unsupported so the openai-completions provider omits the reasoning param
@@ -337,6 +374,12 @@ function applyThinkingLevelMetadata(model: Model<any>): void {
 		// Pi's low/medium/high pass through verbatim; OpenRouter normalizes to Mercury's vocabulary.
 		mergeThinkingLevelMap(model, { off: null });
 	}
+	if (model.provider === "openrouter" && model.id === "z-ai/glm-5.2") {
+		mergeThinkingLevelMap(model, { xhigh: "xhigh" });
+	}
+	if (model.provider === "fireworks" && model.id === "accounts/fireworks/models/glm-5p2") {
+		mergeThinkingLevelMap(model, { off: "none", minimal: null, low: "high", medium: "high", xhigh: "max" });
+	}
 	if (model.provider === "opencode-go" && model.id === "kimi-k2.6") {
 		// OpenCode Go exposes Kimi K2.6 thinking as on/off, not distinct effort tiers.
 		mergeThinkingLevelMap(model, { minimal: null, low: null, medium: null });
@@ -349,6 +392,12 @@ function applyThinkingLevelMetadata(model: Model<any>): void {
 		// Ring reasons by default. Only high/xhigh have documented explicit effort controls.
 		mergeThinkingLevelMap(model, ANT_LING_RING_THINKING_LEVEL_MAP);
 	}
+	if (model.provider === "github-copilot") {
+		const override = GITHUB_COPILOT_THINKING_LEVEL_OVERRIDES[model.id];
+		if (override) {
+			mergeThinkingLevelMap(model, override);
+		}
+	}
 }
 
 function getAnthropicMessagesCompat(provider: string, modelId: string): AnthropicMessagesCompat | undefined {
@@ -372,6 +421,10 @@ function normalizeNvidiaModelId(modelId: string): string {
 	return modelId.toLowerCase().replaceAll("_", ".");
 }
 
+function roundCost(value: number): number {
+	return Number(value.toFixed(6));
+}
+
 async function fetchNvidiaNimModelIds(): Promise<Map<string, string>> {
 	try {
 		console.log("Fetching models from NVIDIA NIM API...");
@@ -417,10 +470,10 @@ async function fetchOpenRouterModels(): Promise<Model<any>[]> {
 			}
 
 			// Convert pricing from $/token to $/million tokens
-			const inputCost = parseFloat(model.pricing?.prompt || "0") * 1_000_000;
-			const outputCost = parseFloat(model.pricing?.completion || "0") * 1_000_000;
-			const cacheReadCost = parseFloat(model.pricing?.input_cache_read || "0") * 1_000_000;
-			const cacheWriteCost = parseFloat(model.pricing?.input_cache_write || "0") * 1_000_000;
+			const inputCost = roundCost(parseFloat(model.pricing?.prompt || "0") * 1_000_000);
+			const outputCost = roundCost(parseFloat(model.pricing?.completion || "0") * 1_000_000);
+			const cacheReadCost = roundCost(parseFloat(model.pricing?.input_cache_read || "0") * 1_000_000);
+			const cacheWriteCost = roundCost(parseFloat(model.pricing?.input_cache_write || "0") * 1_000_000);
 
 			const normalizedModel: Model<any> = {
 				id: modelKey,
@@ -476,10 +529,10 @@ async function fetchAiGatewayModels(): Promise<Model<any>[]> {
 				input.push("image");
 			}
 
-			const inputCost = toNumber(model.pricing?.input) * 1_000_000;
-			const outputCost = toNumber(model.pricing?.output) * 1_000_000;
-			const cacheReadCost = toNumber(model.pricing?.input_cache_read) * 1_000_000;
-			const cacheWriteCost = toNumber(model.pricing?.input_cache_write) * 1_000_000;
+			const inputCost = roundCost(toNumber(model.pricing?.input) * 1_000_000);
+			const outputCost = roundCost(toNumber(model.pricing?.output) * 1_000_000);
+			const cacheReadCost = roundCost(toNumber(model.pricing?.input_cache_read) * 1_000_000);
+			const cacheWriteCost = roundCost(toNumber(model.pricing?.input_cache_write) * 1_000_000);
 
 			models.push({
 				id: model.id,
@@ -586,6 +639,13 @@ async function loadModelsDevData(): Promise<Model<any>[]> {
 			for (const [modelId, model] of Object.entries(data.google.models)) {
 				const m = model as ModelsDevModel;
 				if (m.tool_call !== true) continue;
+				let source = m;
+				if (modelId === "gemini-flash-latest") {
+					source = (data.google.models["gemini-3.5-flash"] as ModelsDevModel | undefined) ?? m;
+				}
+				if (modelId === "gemini-flash-lite-latest") {
+					source = (data.google.models["gemini-3.1-flash-lite"] as ModelsDevModel | undefined) ?? m;
+				}
 
 				models.push({
 					id: modelId,
@@ -593,16 +653,57 @@ async function loadModelsDevData(): Promise<Model<any>[]> {
 					api: "google-generative-ai",
 					provider: "google",
 					baseUrl: "https://generativelanguage.googleapis.com/v1beta",
-					reasoning: m.reasoning === true,
-					input: m.modalities?.input?.includes("image") ? ["text", "image"] : ["text"],
+					reasoning: source.reasoning === true,
+					input: source.modalities?.input?.includes("image") ? ["text", "image"] : ["text"],
 					cost: {
-						input: m.cost?.input || 0,
-						output: m.cost?.output || 0,
-						cacheRead: m.cost?.cache_read || 0,
-						cacheWrite: m.cost?.cache_write || 0,
+						input: source.cost?.input || 0,
+						output: source.cost?.output || 0,
+						cacheRead: source.cost?.cache_read || 0,
+						cacheWrite: source.cost?.cache_write || 0,
 					},
-					contextWindow: m.limit?.context || 4096,
-					maxTokens: m.limit?.output || 4096,
+					contextWindow: source.limit?.context || 4096,
+					maxTokens: source.limit?.output || 4096,
+				});
+			}
+		}
+
+		// Process Google Vertex Gemini models. The google-vertex models.dev catalog also includes
+		// Claude, OpenAI, and other MaaS models that do not use the @google/genai Gemini streaming
+		// path implemented by our google-vertex provider.
+		if (data["google-vertex"]?.models) {
+			for (const [modelId, model] of Object.entries(data["google-vertex"].models)) {
+				const m = model as ModelsDevModel;
+				if (m.tool_call !== true) continue;
+				if (!modelId.startsWith("gemini-")) continue;
+				if (modelId === "gemini-3.1-flash-lite-preview") continue;
+				let source = m;
+				if (modelId === "gemini-flash-latest") {
+					source = (data["google-vertex"].models["gemini-3.5-flash"] as ModelsDevModel | undefined) ?? m;
+				}
+				if (modelId === "gemini-flash-lite-latest") {
+					source = (data["google-vertex"].models["gemini-3.1-flash-lite"] as ModelsDevModel | undefined) ?? m;
+				}
+
+				// models.dev reports Vertex cache_read/cache_write values for Gemini 2.5 Flash that
+				// do not match the official Gemini API standard pricing table. pi only accounts
+				// cachedContentTokenCount as cacheRead.
+				const cacheRead = modelId === "gemini-2.5-flash" ? 0.03 : source.cost?.cache_read || 0;
+				models.push({
+					id: modelId,
+					name: m.name || modelId,
+					api: "google-vertex",
+					provider: "google-vertex",
+					baseUrl: VERTEX_BASE_URL,
+					reasoning: source.reasoning === true,
+					input: source.modalities?.input?.includes("image") ? ["text", "image"] : ["text"],
+					cost: {
+						input: source.cost?.input || 0,
+						output: source.cost?.output || 0,
+						cacheRead,
+						cacheWrite: 0,
+					},
+					contextWindow: source.limit?.context || 4096,
+					maxTokens: source.limit?.output || 4096,
 				});
 			}
 		}
@@ -806,6 +907,8 @@ async function loadModelsDevData(): Promise<Model<any>[]> {
 					if (m.tool_call !== true) continue;
 					const supportsImage = m.modalities?.input?.includes("image");
 
+					const isGlm52 = modelId === "glm-5.2";
+
 					models.push({
 						id: modelId,
 						name: m.name || modelId,
@@ -813,6 +916,7 @@ async function loadModelsDevData(): Promise<Model<any>[]> {
 						provider,
 						baseUrl,
 						reasoning: m.reasoning === true,
+						...(isGlm52 ? { thinkingLevelMap: ZAI_GLM52_THINKING_LEVEL_MAP } : {}),
 						input: supportsImage ? ["text", "image"] : ["text"],
 						cost: {
 							input: m.cost?.input || 0,
@@ -823,6 +927,7 @@ async function loadModelsDevData(): Promise<Model<any>[]> {
 						compat: {
 							supportsDeveloperRole: false,
 							thinkingFormat: "zai",
+							...(isGlm52 ? { supportsReasoningEffort: true } : {}),
 							...(!ZAI_TOOL_STREAM_UNSUPPORTED_MODELS.has(modelId) ? { zaiToolStream: true } : {}),
 						},
 						contextWindow: m.limit?.context || 4096,
@@ -849,7 +954,7 @@ async function loadModelsDevData(): Promise<Model<any>[]> {
 					cost: {
 						input: m.cost?.input || 0,
 						output: m.cost?.output || 0,
-						cacheRead: m.cost?.cache_read || 0,
+						cacheRead: m.cost?.cache_read ?? (m.cost?.input ? roundCost(m.cost.input * 0.1) : 0),
 						cacheWrite: m.cost?.cache_write || 0,
 					},
 					contextWindow: m.limit?.context || 4096,
@@ -1066,6 +1171,13 @@ async function loadModelsDevData(): Promise<Model<any>[]> {
 
 				if (api === "openai-completions") {
 					compat = { ...(compat ?? {}), maxTokensField: "max_tokens" };
+					if (
+						OPENCODE_OPENAI_COMPLETIONS_LONG_CACHE_RETENTION_UNSUPPORTED_MODELS.has(
+							`${variant.provider}:${modelId}`,
+						)
+					) {
+						compat = { ...compat, supportsLongCacheRetention: false };
+					}
 				}
 
 				models.push({
@@ -1228,12 +1340,27 @@ async function loadModelsDevData(): Promise<Model<any>[]> {
 			supportsStrictMode: false,
 			thinkingFormat: "deepseek",
 		};
+		const getMoonshotProviderModels = (key: "moonshotai" | "moonshotai-cn"): Record<string, ModelsDevModel> => {
+			const providerModels = data[key]?.models as Record<string, ModelsDevModel> | undefined;
+			return providerModels ? { ...providerModels } : {};
+		};
+		const moonshotModels = {
+			moonshotai: getMoonshotProviderModels("moonshotai"),
+			"moonshotai-cn": getMoonshotProviderModels("moonshotai-cn"),
+		};
+
+		// models.dev can lag the CN catalog while the global Moonshot catalog already
+		// has the model. Mirror selected current model IDs into moonshotai-cn until
+		// upstream CN metadata catches up.
+		for (const modelId of MOONSHOT_CN_MIRRORED_MODEL_IDS) {
+			const model = moonshotModels.moonshotai[modelId];
+			if (model && !moonshotModels["moonshotai-cn"][modelId]) {
+				moonshotModels["moonshotai-cn"][modelId] = model;
+			}
+		}
 
 		for (const { key, provider, baseUrl } of moonshotVariants) {
-			if (!data[key]?.models) continue;
-
-			for (const [modelId, model] of Object.entries(data[key].models)) {
-				const m = model as ModelsDevModel;
+			for (const [modelId, m] of Object.entries(moonshotModels[key])) {
 				if (m.tool_call !== true) continue;
 
 				models.push({
@@ -1390,7 +1517,11 @@ async function generateModels() {
 			candidate.cost.output = 1.9;
 			candidate.cost.cacheRead = 0.119;
 		}
-
+		if (candidate.provider === "fireworks" && candidate.id === "accounts/fireworks/models/glm-5p2") {
+			candidate.api = "openai-completions";
+			candidate.baseUrl = "https://api.fireworks.ai/inference/v1";
+			candidate.compat = { supportsStore: false, supportsDeveloperRole: false };
+		}
 	}
 
 
@@ -1731,9 +1862,10 @@ async function generateModels() {
 
 	for (const candidate of allModels) {
 		if (candidate.api === "openai-completions" && candidate.id.includes("deepseek-v4")) {
+			const preservesNativeReasoningEffort = candidate.provider === "openrouter" || candidate.provider === "opencode";
 			candidate.compat = {
 				...candidate.compat,
-				...(candidate.provider === "openrouter"
+				...(preservesNativeReasoningEffort
 					? {
 							requiresReasoningContentOnAssistantMessages:
 								deepseekCompat.requiresReasoningContentOnAssistantMessages,
@@ -1908,166 +2040,31 @@ async function generateModels() {
 		});
 	}
 
-	const VERTEX_BASE_URL = "https://{location}-aiplatform.googleapis.com";
-	const vertexModels: Model<"google-vertex">[] = [
-		{
-			id: "gemini-3-pro-preview",
-			name: "Gemini 3 Pro Preview (Vertex)",
-			api: "google-vertex",
-			provider: "google-vertex",
-			baseUrl: VERTEX_BASE_URL,
+	// Add "fusion" alias for openrouter/fusion. OpenRouter exposes Fusion as a
+	// router alias/plugin entry point; its model metadata does not advertise
+	// tools, but the alias resolves to a concrete model that can invoke caller
+	// tools and has the openrouter:fusion server tool auto-injected.
+	if (!allModels.some(m => m.provider === "openrouter" && m.id === "openrouter/fusion")) {
+		allModels.push({
+			id: "openrouter/fusion",
+			name: "OpenRouter: Fusion",
+			api: "openai-completions",
+			provider: "openrouter",
+			baseUrl: "https://openrouter.ai/api/v1",
 			reasoning: true,
-			input: ["text", "image"],
-			cost: { input: 2, output: 12, cacheRead: 0.2, cacheWrite: 0 },
+			input: ["text"],
+			cost: {
+				// we dont know about the costs because Fusion routes to multiple models
+				// and then charges you for the underlying used models
+				input: 0,
+				output: 0,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
 			contextWindow: 1000000,
-			maxTokens: 64000,
-		},
-		{
-			id: "gemini-3.1-pro-preview",
-			name: "Gemini 3.1 Pro Preview (Vertex)",
-			api: "google-vertex",
-			provider: "google-vertex",
-			baseUrl: VERTEX_BASE_URL,
-			reasoning: true,
-			input: ["text", "image"],
-			cost: { input: 2, output: 12, cacheRead: 0.2, cacheWrite: 0 },
-			contextWindow: 1048576,
-			maxTokens: 65536,
-		},
-		{
-			id: "gemini-3.1-pro-preview-customtools",
-			name: "Gemini 3.1 Pro Preview Custom Tools (Vertex)",
-			api: "google-vertex",
-			provider: "google-vertex",
-			baseUrl: VERTEX_BASE_URL,
-			reasoning: true,
-			input: ["text", "image"],
-			cost: { input: 2, output: 12, cacheRead: 0.2, cacheWrite: 0 },
-			contextWindow: 1048576,
-			maxTokens: 65536,
-		},
-		{
-			id: "gemini-3-flash-preview",
-			name: "Gemini 3 Flash Preview (Vertex)",
-			api: "google-vertex",
-			provider: "google-vertex",
-			baseUrl: VERTEX_BASE_URL,
-			reasoning: true,
-			input: ["text", "image"],
-			cost: { input: 0.5, output: 3, cacheRead: 0.05, cacheWrite: 0 },
-			contextWindow: 1048576,
-			maxTokens: 65536,
-		},
-		{
-			id: "gemini-2.0-flash",
-			name: "Gemini 2.0 Flash (Vertex)",
-			api: "google-vertex",
-			provider: "google-vertex",
-			baseUrl: VERTEX_BASE_URL,
-			reasoning: false,
-			input: ["text", "image"],
-			cost: { input: 0.15, output: 0.6, cacheRead: 0.0375, cacheWrite: 0 },
-			contextWindow: 1048576,
-			maxTokens: 8192,
-		},
-		{
-			id: "gemini-2.0-flash-lite",
-			name: "Gemini 2.0 Flash Lite (Vertex)",
-			api: "google-vertex",
-			provider: "google-vertex",
-			baseUrl: VERTEX_BASE_URL,
-			reasoning: true,
-			input: ["text", "image"],
-			cost: { input: 0.075, output: 0.3, cacheRead: 0.01875, cacheWrite: 0 },
-			contextWindow: 1048576,
-			maxTokens: 65536,
-		},
-		{
-			id: "gemini-2.5-pro",
-			name: "Gemini 2.5 Pro (Vertex)",
-			api: "google-vertex",
-			provider: "google-vertex",
-			baseUrl: VERTEX_BASE_URL,
-			reasoning: true,
-			input: ["text", "image"],
-			cost: { input: 1.25, output: 10, cacheRead: 0.125, cacheWrite: 0 },
-			contextWindow: 1048576,
-			maxTokens: 65536,
-		},
-		{
-			id: "gemini-2.5-flash",
-			name: "Gemini 2.5 Flash (Vertex)",
-			api: "google-vertex",
-			provider: "google-vertex",
-			baseUrl: VERTEX_BASE_URL,
-			reasoning: true,
-			input: ["text", "image"],
-			cost: { input: 0.3, output: 2.5, cacheRead: 0.03, cacheWrite: 0 },
-			contextWindow: 1048576,
-			maxTokens: 65536,
-		},
-		{
-			id: "gemini-2.5-flash-lite-preview-09-2025",
-			name: "Gemini 2.5 Flash Lite Preview 09-25 (Vertex)",
-			api: "google-vertex",
-			provider: "google-vertex",
-			baseUrl: VERTEX_BASE_URL,
-			reasoning: true,
-			input: ["text", "image"],
-			cost: { input: 0.1, output: 0.4, cacheRead: 0.01, cacheWrite: 0 },
-			contextWindow: 1048576,
-			maxTokens: 65536,
-		},
-		{
-			id: "gemini-2.5-flash-lite",
-			name: "Gemini 2.5 Flash Lite (Vertex)",
-			api: "google-vertex",
-			provider: "google-vertex",
-			baseUrl: VERTEX_BASE_URL,
-			reasoning: true,
-			input: ["text", "image"],
-			cost: { input: 0.1, output: 0.4, cacheRead: 0.01, cacheWrite: 0 },
-			contextWindow: 1048576,
-			maxTokens: 65536,
-		},
-		{
-			id: "gemini-1.5-pro",
-			name: "Gemini 1.5 Pro (Vertex)",
-			api: "google-vertex",
-			provider: "google-vertex",
-			baseUrl: VERTEX_BASE_URL,
-			reasoning: false,
-			input: ["text", "image"],
-			cost: { input: 1.25, output: 5, cacheRead: 0.3125, cacheWrite: 0 },
-			contextWindow: 1000000,
-			maxTokens: 8192,
-		},
-		{
-			id: "gemini-1.5-flash",
-			name: "Gemini 1.5 Flash (Vertex)",
-			api: "google-vertex",
-			provider: "google-vertex",
-			baseUrl: VERTEX_BASE_URL,
-			reasoning: false,
-			input: ["text", "image"],
-			cost: { input: 0.075, output: 0.3, cacheRead: 0.01875, cacheWrite: 0 },
-			contextWindow: 1000000,
-			maxTokens: 8192,
-		},
-		{
-			id: "gemini-1.5-flash-8b",
-			name: "Gemini 1.5 Flash-8B (Vertex)",
-			api: "google-vertex",
-			provider: "google-vertex",
-			baseUrl: VERTEX_BASE_URL,
-			reasoning: false,
-			input: ["text", "image"],
-			cost: { input: 0.0375, output: 0.15, cacheRead: 0.01, cacheWrite: 0 },
-			contextWindow: 1000000,
-			maxTokens: 8192,
-		},
-	];
-	allModels.push(...vertexModels);
+			maxTokens: 30000,
+		});
+	}
 
 	// Azure Foundry deploys these with larger context windows than OpenAI's own API,
 	// which caps gpt-5.4/gpt-5.5 at 272k. See models-sold-directly-by-azure docs.

packages/ai/src/api/anthropic-messages.ts

@@ -5,6 +5,7 @@ import type {
 	MessageCreateParamsStreaming,
 	MessageParam,
 	RawMessageStreamEvent,
+	RefusalStopDetails,
 } from "@anthropic-ai/sdk/resources/messages.js";
 import { calculateCost } from "../models.ts";
 import type {
@@ -16,6 +17,7 @@ import type {
 	ImageContent,
 	Message,
 	Model,
+	ProviderEnv,
 	SimpleStreamOptions,
 	StopReason,
 	StreamFunction,
@@ -29,6 +31,7 @@ import type {
 import { AssistantMessageEventStream } from "../utils/event-stream.ts";
 import { headersToRecord } from "../utils/headers.ts";
 import { parseJsonWithRepair, parseStreamingJson } from "../utils/json-parse.ts";
+import { getProviderEnvValue } from "../utils/provider-env.ts";
 import { sanitizeSurrogates } from "../utils/sanitize-unicode.ts";
 
 import { resolveCloudflareBaseUrl } from "./cloudflare.ts";
@@ -40,11 +43,11 @@ import { transformMessages } from "./transform-messages.ts";
  * Resolve cache retention preference.
  * Defaults to "short" and uses PI_CACHE_RETENTION for backward compatibility.
  */
-function resolveCacheRetention(cacheRetention?: CacheRetention): CacheRetention {
+function resolveCacheRetention(cacheRetention?: CacheRetention, env?: ProviderEnv): CacheRetention {
 	if (cacheRetention) {
 		return cacheRetention;
 	}
-	if (typeof process !== "undefined" && process.env.PI_CACHE_RETENTION === "long") {
+	if (getProviderEnvValue("PI_CACHE_RETENTION", env) === "long") {
 		return "long";
 	}
 	return "short";
@@ -53,8 +56,9 @@ function resolveCacheRetention(cacheRetention?: CacheRetention): CacheRetention
 function getCacheControl(
 	model: Model<"anthropic-messages">,
 	cacheRetention?: CacheRetention,
+	env?: ProviderEnv,
 ): { retention: CacheRetention; cacheControl?: CacheControlEphemeral } {
-	const retention = resolveCacheRetention(cacheRetention);
+	const retention = resolveCacheRetention(cacheRetention, env);
 	if (retention === "none") {
 		return { retention };
 	}
@@ -493,7 +497,7 @@ export const stream: StreamFunction<"anthropic-messages", AnthropicOptions> = (
 					});
 				}
 
-				const cacheRetention = options?.cacheRetention ?? resolveCacheRetention();
+				const cacheRetention = resolveCacheRetention(options?.cacheRetention, options?.env);
 				const cacheSessionId = cacheRetention === "none" ? undefined : options?.sessionId;
 
 				const created = createClient(
@@ -504,6 +508,7 @@ export const stream: StreamFunction<"anthropic-messages", AnthropicOptions> = (
 					options?.headers,
 					copilotDynamicHeaders,
 					cacheSessionId,
+					options?.env,
 				);
 				client = created.client;
 				isOAuth = created.isOAuthToken;
@@ -534,6 +539,7 @@ export const stream: StreamFunction<"anthropic-messages", AnthropicOptions> = (
 					output.usage.output = event.message.usage.output_tokens || 0;
 					output.usage.cacheRead = event.message.usage.cache_read_input_tokens || 0;
 					output.usage.cacheWrite = event.message.usage.cache_creation_input_tokens || 0;
+					output.usage.cacheWrite1h = event.message.usage.cache_creation?.ephemeral_1h_input_tokens || 0;
 					// Anthropic doesn't provide total_tokens, compute from components
 					output.usage.totalTokens =
 						output.usage.input + output.usage.output + output.usage.cacheRead + output.usage.cacheWrite;
@@ -660,7 +666,11 @@ export const stream: StreamFunction<"anthropic-messages", AnthropicOptions> = (
 					}
 				} else if (event.type === "message_delta") {
 					if (event.delta.stop_reason) {
-						output.stopReason = mapStopReason(event.delta.stop_reason);
+						const stopReasonResult = mapStopReason(event.delta.stop_reason, event.delta.stop_details);
+						output.stopReason = stopReasonResult.stopReason;
+						if (stopReasonResult.errorMessage) {
+							output.errorMessage = stopReasonResult.errorMessage;
+						}
 					}
 					// Only update usage fields if present (not null).
 					// Preserves input_tokens from message_start when proxies omit it in message_delta.
@@ -688,7 +698,7 @@ export const stream: StreamFunction<"anthropic-messages", AnthropicOptions> = (
 			}
 
 			if (output.stopReason === "aborted" || output.stopReason === "error") {
-				throw new Error("An unknown error occurred");
+				throw new Error(output.errorMessage || "An unknown error occurred");
 			}
 
 			stream.push({ type: "done", reason: output.stopReason, message: output });
@@ -788,6 +798,7 @@ function createClient(
 	optionsHeaders?: Record<string, string>,
 	dynamicHeaders?: Record<string, string>,
 	sessionId?: string,
+	env?: ProviderEnv,
 ): { client: Anthropic; isOAuthToken: boolean } {
 	// Adaptive thinking models have interleaved thinking built in, so skip the beta header.
 	const needsInterleavedBeta = interleavedThinking && model.compat?.forceAdaptiveThinking !== true;
@@ -803,7 +814,7 @@ function createClient(
 		const client = new Anthropic({
 			apiKey: null,
 			authToken: null,
-			baseURL: resolveCloudflareBaseUrl(model),
+			baseURL: resolveCloudflareBaseUrl(model, env),
 			dangerouslyAllowBrowser: true,
 			defaultHeaders: mergeHeaders(
 				{
@@ -896,7 +907,7 @@ function buildParams(
 	isOAuthToken: boolean,
 	options?: AnthropicOptions,
 ): MessageCreateParamsStreaming {
-	const { cacheControl } = getCacheControl(model, options?.cacheRetention);
+	const { cacheControl } = getCacheControl(model, options?.cacheRetention, options?.env);
 	const compat = getAnthropicCompat(model);
 	const params: MessageCreateParamsStreaming = {
 		model: model.id,
@@ -1202,22 +1213,28 @@ function convertTools(
 	});
 }
 
-function mapStopReason(reason: Anthropic.Messages.StopReason | string): StopReason {
+function mapStopReason(
+	reason: Anthropic.Messages.StopReason | string,
+	stopDetails?: RefusalStopDetails | null,
+): { stopReason: StopReason; errorMessage?: string } {
 	switch (reason) {
 		case "end_turn":
-			return "stop";
+			return { stopReason: "stop" };
 		case "max_tokens":
-			return "length";
+			return { stopReason: "length" };
 		case "tool_use":
-			return "toolUse";
+			return { stopReason: "toolUse" };
 		case "refusal":
-			return "error";
+			return {
+				stopReason: "error",
+				errorMessage: stopDetails?.explanation || `The model refused to complete the request`,
+			};
 		case "pause_turn": // Stop is good enough -> resubmit
-			return "stop";
+			return { stopReason: "stop" };
 		case "stop_sequence":
-			return "stop"; // We don't supply stop sequences, so this should never happen
+			return { stopReason: "stop" }; // We don't supply stop sequences, so this should never happen
 		case "sensitive": // Content flagged by safety filters (not yet in SDK types)
-			return "error";
+			return { stopReason: "error" };
 		default:
 			// Handle unknown stop reasons gracefully (API may add new values)
 			throw new Error(`Unhandled stop reason: ${reason}`);

packages/ai/src/api/azure-openai-responses.ts

@@ -12,6 +12,7 @@ import type {
 } from "../types.ts";
 import { AssistantMessageEventStream } from "../utils/event-stream.ts";
 import { headersToRecord } from "../utils/headers.ts";
+import { getProviderEnvValue } from "../utils/provider-env.ts";
 import { clampOpenAIPromptCacheKey } from "./openai-prompt-cache.ts";
 import { convertResponsesMessages, convertResponsesTools, processResponsesStream } from "./openai-responses-shared.ts";
 import { buildBaseOptions } from "./simple-options.ts";
@@ -36,7 +37,9 @@ function resolveDeploymentName(model: Model<"azure-openai-responses">, options?:
 	if (options?.azureDeploymentName) {
 		return options.azureDeploymentName;
 	}
-	const mappedDeployment = parseDeploymentNameMap(process.env.AZURE_OPENAI_DEPLOYMENT_NAME_MAP).get(model.id);
+	const mappedDeployment = parseDeploymentNameMap(
+		getProviderEnvValue("AZURE_OPENAI_DEPLOYMENT_NAME_MAP", options?.env),
+	).get(model.id);
 	return mappedDeployment || model.id;
 }
 
@@ -198,10 +201,14 @@ function resolveAzureConfig(
 	model: Model<"azure-openai-responses">,
 	options?: AzureOpenAIResponsesOptions,
 ): { baseUrl: string; apiVersion: string } {
-	const apiVersion = options?.azureApiVersion || process.env.AZURE_OPENAI_API_VERSION || DEFAULT_AZURE_API_VERSION;
+	const apiVersion =
+		options?.azureApiVersion ||
+		getProviderEnvValue("AZURE_OPENAI_API_VERSION", options?.env) ||
+		DEFAULT_AZURE_API_VERSION;
 
-	const baseUrl = options?.azureBaseUrl?.trim() || process.env.AZURE_OPENAI_BASE_URL?.trim() || undefined;
-	const resourceName = options?.azureResourceName || process.env.AZURE_OPENAI_RESOURCE_NAME;
+	const baseUrl =
+		options?.azureBaseUrl?.trim() || getProviderEnvValue("AZURE_OPENAI_BASE_URL", options?.env)?.trim() || undefined;
+	const resourceName = options?.azureResourceName || getProviderEnvValue("AZURE_OPENAI_RESOURCE_NAME", options?.env);
 
 	let resolvedBaseUrl = baseUrl;
 

packages/ai/src/api/bedrock-converse-stream.ts

@@ -1,3 +1,4 @@
+import type { Agent as HttpsAgent } from "node:https";
 import {
 	BedrockRuntimeClient,
 	type BedrockRuntimeClientConfig,
@@ -23,6 +24,8 @@ import {
 } from "@aws-sdk/client-bedrock-runtime";
 import { NodeHttpHandler } from "@smithy/node-http-handler";
 import type { BuildMiddleware, DocumentType, MetadataBearer } from "@smithy/types";
+import { HttpProxyAgent } from "http-proxy-agent";
+import { HttpsProxyAgent } from "https-proxy-agent";
 import { calculateCost } from "../models.ts";
 import type {
 	Api,
@@ -31,6 +34,7 @@ import type {
 	Context,
 	ImageContent,
 	Model,
+	ProviderEnv,
 	SimpleStreamOptions,
 	StopReason,
 	StreamFunction,
@@ -45,7 +49,8 @@ import type {
 } from "../types.ts";
 import { AssistantMessageEventStream } from "../utils/event-stream.ts";
 import { parseStreamingJson } from "../utils/json-parse.ts";
-import { createHttpProxyAgentsForTarget } from "../utils/node-http-proxy.ts";
+import { resolveHttpProxyUrlForTarget } from "../utils/node-http-proxy.ts";
+import { getProviderEnvValue } from "../utils/provider-env.ts";
 import { sanitizeSurrogates } from "../utils/sanitize-unicode.ts";
 import { adjustMaxTokensForThinking, buildBaseOptions, clampReasoning } from "./simple-options.ts";
 import { transformMessages } from "./transform-messages.ts";
@@ -119,18 +124,18 @@ export const stream: StreamFunction<"bedrock-converse-stream", BedrockOptions> =
 		const blocks = output.content as Block[];
 
 		const config: BedrockRuntimeClientConfig = {
-			profile: options.profile,
+			profile: options.profile || getProviderEnvValue("AWS_PROFILE", options.env),
 		};
 		const configuredRegion = getConfiguredBedrockRegion(options);
-		const hasConfiguredProfile = hasConfiguredBedrockProfile();
+		const hasAmbientConfiguredProfile = Boolean(getProviderEnvValue("AWS_PROFILE"));
 		const endpointRegion = getStandardBedrockEndpointRegion(model.baseUrl);
 		const useExplicitEndpoint = shouldUseExplicitBedrockEndpoint(
 			model.baseUrl,
 			configuredRegion,
-			hasConfiguredProfile,
+			hasAmbientConfiguredProfile,
 		);
 
-		// Only pin standard AWS Bedrock runtime endpoints when no region/profile is configured.
+		// Only pin standard AWS Bedrock runtime endpoints when no region or ambient AWS_PROFILE is configured.
 		// This preserves custom endpoints (VPC/proxy) from #3402 without forcing built-in
 		// catalog defaults such as us-east-1 to override AWS_REGION/AWS_PROFILE.
 		if (useExplicitEndpoint) {
@@ -138,8 +143,10 @@ export const stream: StreamFunction<"bedrock-converse-stream", BedrockOptions> =
 		}
 
 		// Resolve bearer token for Bedrock API key auth.
-		const bearerToken = options.bearerToken || process.env.AWS_BEARER_TOKEN_BEDROCK || undefined;
-		const useBearerToken = bearerToken !== undefined && process.env.AWS_BEDROCK_SKIP_AUTH !== "1";
+		const skipAuth = getProviderEnvValue("AWS_BEDROCK_SKIP_AUTH", options.env) === "1";
+		const bearerToken =
+			options.bearerToken || getProviderEnvValue("AWS_BEARER_TOKEN_BEDROCK", options.env) || undefined;
+		const useBearerToken = bearerToken !== undefined && !skipAuth;
 
 		// in Node.js/Bun environment only
 		if (typeof process !== "undefined" && (process.versions?.node || process.versions?.bun)) {
@@ -153,25 +160,33 @@ export const stream: StreamFunction<"bedrock-converse-stream", BedrockOptions> =
 				config.region = configuredRegion;
 			} else if (endpointRegion && useExplicitEndpoint) {
 				config.region = endpointRegion;
-			} else if (!hasConfiguredProfile) {
+			} else if (!hasAmbientConfiguredProfile) {
 				config.region = "us-east-1";
 			}
 
 			// Support proxies that don't need authentication
-			if (process.env.AWS_BEDROCK_SKIP_AUTH === "1") {
+			if (skipAuth) {
 				config.credentials = {
 					accessKeyId: "dummy-access-key",
 					secretAccessKey: "dummy-secret-key",
 				};
 			}
 
-			const proxyAgents = createHttpProxyAgentsForTarget(model.baseUrl);
-			if (proxyAgents) {
+			const credentials = getConfiguredBedrockCredentials(options.env);
+			if (!skipAuth && credentials) {
+				config.credentials = credentials;
+			}
+
+			const proxyUrl = resolveHttpProxyUrlForTarget(model.baseUrl, options.env);
+			if (proxyUrl) {
 				// Bedrock runtime uses NodeHttp2Handler by default since v3.798.0, which is based
 				// on `http2` module and has no support for http agent.
 				// Use NodeHttpHandler to support HTTP(S) proxy agents.
-				config.requestHandler = new NodeHttpHandler(proxyAgents);
-			} else if (process.env.AWS_BEDROCK_FORCE_HTTP1 === "1") {
+				config.requestHandler = new NodeHttpHandler({
+					httpAgent: new HttpProxyAgent(proxyUrl),
+					httpsAgent: new HttpsProxyAgent(proxyUrl) as unknown as HttpsAgent,
+				});
+			} else if (getProviderEnvValue("AWS_BEDROCK_FORCE_HTTP1", options.env) === "1") {
 				// Some custom endpoints require HTTP/1.1 instead of HTTP/2
 				config.requestHandler = new NodeHttpHandler();
 			}
@@ -192,12 +207,12 @@ export const stream: StreamFunction<"bedrock-converse-stream", BedrockOptions> =
 			if (options.headers && Object.keys(options.headers).length > 0) {
 				addCustomHeadersMiddleware(client, options.headers);
 			}
-			const cacheRetention = resolveCacheRetention(options.cacheRetention);
+			const cacheRetention = resolveCacheRetention(options.cacheRetention, options.env);
 			const inferenceMaxTokens = options.maxTokens ?? (isAnthropicClaudeModel(model) ? model.maxTokens : undefined);
 			let commandInput = {
 				modelId: model.id,
-				messages: convertMessages(context, model, cacheRetention),
-				system: buildSystemPrompt(context.systemPrompt, model, cacheRetention),
+				messages: convertMessages(context, model, cacheRetention, options.env),
+				system: buildSystemPrompt(context.systemPrompt, model, cacheRetention, options.env),
 				inferenceConfig: {
 					...(inferenceMaxTokens !== undefined && { maxTokens: inferenceMaxTokens }),
 					...(options.temperature !== undefined && { temperature: options.temperature }),
@@ -578,11 +593,11 @@ function mapThinkingLevelToEffort(
  * Resolve cache retention preference.
  * Defaults to "short" and uses PI_CACHE_RETENTION for backward compatibility.
  */
-function resolveCacheRetention(cacheRetention?: CacheRetention): CacheRetention {
+function resolveCacheRetention(cacheRetention?: CacheRetention, env?: ProviderEnv): CacheRetention {
 	if (cacheRetention) {
 		return cacheRetention;
 	}
-	if (typeof process !== "undefined" && process.env.PI_CACHE_RETENTION === "long") {
+	if (getProviderEnvValue("PI_CACHE_RETENTION", env) === "long") {
 		return "long";
 	}
 	return "short";
@@ -617,14 +632,14 @@ function isAnthropicClaudeModel(model: Model<"bedrock-converse-stream">): boolea
  * As a last resort, set AWS_BEDROCK_FORCE_CACHE=1 to enable cache points.
  * Amazon Nova models have automatic caching and don't need explicit cache points.
  */
-function supportsPromptCaching(model: Model<"bedrock-converse-stream">): boolean {
+function supportsPromptCaching(model: Model<"bedrock-converse-stream">, env?: ProviderEnv): boolean {
 	const candidates = getModelMatchCandidates(model.id, model.name);
 
 	const hasClaudeRef = candidates.some((s) => s.includes("claude"));
 	if (!hasClaudeRef) {
 		// Application inference profiles don't contain the model name in the ARN.
 		// Allow users to force cache points via environment variable.
-		if (typeof process !== "undefined" && process.env.AWS_BEDROCK_FORCE_CACHE === "1") return true;
+		if (getProviderEnvValue("AWS_BEDROCK_FORCE_CACHE", env) === "1") return true;
 		return false;
 	}
 	// Claude 4.x models (opus-4, sonnet-4, haiku-4)
@@ -652,13 +667,14 @@ function buildSystemPrompt(
 	systemPrompt: string | undefined,
 	model: Model<"bedrock-converse-stream">,
 	cacheRetention: CacheRetention,
+	env?: ProviderEnv,
 ): SystemContentBlock[] | undefined {
 	if (!systemPrompt) return undefined;
 
 	const blocks: SystemContentBlock[] = [{ text: sanitizeSurrogates(systemPrompt) }];
 
 	// Add cache point for supported Claude models when caching is enabled
-	if (cacheRetention !== "none" && supportsPromptCaching(model)) {
+	if (cacheRetention !== "none" && supportsPromptCaching(model, env)) {
 		blocks.push({
 			cachePoint: { type: CachePointType.DEFAULT, ...(cacheRetention === "long" ? { ttl: CacheTTL.ONE_HOUR } : {}) },
 		});
@@ -699,6 +715,7 @@ function convertMessages(
 	context: Context,
 	model: Model<"bedrock-converse-stream">,
 	cacheRetention: CacheRetention,
+	env?: ProviderEnv,
 ): Message[] {
 	const result: Message[] = [];
 	const transformedMessages = transformMessages(context.messages, model, normalizeToolCallId);
@@ -844,7 +861,7 @@ function convertMessages(
 	}
 
 	// Add cache point to the last user message for supported Claude models when caching is enabled
-	if (cacheRetention !== "none" && supportsPromptCaching(model) && result.length > 0) {
+	if (cacheRetention !== "none" && supportsPromptCaching(model, env) && result.length > 0) {
 		const lastMessage = result[result.length - 1];
 		if (lastMessage.role === ConversationRole.USER && lastMessage.content) {
 			(lastMessage.content as ContentBlock[]).push({
@@ -906,19 +923,26 @@ function mapStopReason(reason: string | undefined): StopReason {
 }
 
 function getConfiguredBedrockRegion(options: BedrockOptions): string | undefined {
-	if (typeof process === "undefined") {
-		return options.region;
-	}
-
-	return options.region || process.env.AWS_REGION || process.env.AWS_DEFAULT_REGION || undefined;
+	return (
+		options.region ||
+		getProviderEnvValue("AWS_REGION", options.env) ||
+		getProviderEnvValue("AWS_DEFAULT_REGION", options.env) ||
+		undefined
+	);
 }
 
-function hasConfiguredBedrockProfile(): boolean {
-	if (typeof process === "undefined") {
-		return false;
+function getConfiguredBedrockCredentials(env?: ProviderEnv): BedrockRuntimeClientConfig["credentials"] | undefined {
+	const accessKeyId = getProviderEnvValue("AWS_ACCESS_KEY_ID", env);
+	const secretAccessKey = getProviderEnvValue("AWS_SECRET_ACCESS_KEY", env);
+	if (!accessKeyId || !secretAccessKey) {
+		return undefined;
 	}
-
-	return Boolean(process.env.AWS_PROFILE);
+	const sessionToken = getProviderEnvValue("AWS_SESSION_TOKEN", env);
+	return {
+		accessKeyId,
+		secretAccessKey,
+		...(sessionToken ? { sessionToken } : {}),
+	};
 }
 
 function getStandardBedrockEndpointRegion(baseUrl: string | undefined): string | undefined {
@@ -938,14 +962,14 @@ function getStandardBedrockEndpointRegion(baseUrl: string | undefined): string |
 function shouldUseExplicitBedrockEndpoint(
 	baseUrl: string,
 	configuredRegion: string | undefined,
-	hasConfiguredProfile: boolean,
+	hasAmbientConfiguredProfile: boolean,
 ): boolean {
 	const endpointRegion = getStandardBedrockEndpointRegion(baseUrl);
 	if (!endpointRegion) {
 		return true;
 	}
 
-	return !configuredRegion && !hasConfiguredProfile;
+	return !configuredRegion && !hasAmbientConfiguredProfile;
 }
 
 function isGovCloudBedrockTarget(model: Model<"bedrock-converse-stream">, options: BedrockOptions): boolean {

packages/ai/src/api/cloudflare.ts

@@ -1,4 +1,5 @@
-import type { Api, Model } from "../types.ts";
+import type { Api, Model, ProviderEnv } from "../types.ts";
+import { getProviderEnvValue } from "../utils/provider-env.ts";
 
 /** Workers AI direct endpoint. */
 export const CLOUDFLARE_WORKERS_AI_BASE_URL =
@@ -20,12 +21,12 @@ export function isCloudflareProvider(provider: string): boolean {
 	return provider === "cloudflare-workers-ai" || provider === "cloudflare-ai-gateway";
 }
 
-/** Substitute `{VAR}` placeholders in a Cloudflare baseUrl from process.env. */
-export function resolveCloudflareBaseUrl(model: Model<Api>): string {
+/** Substitute `{VAR}` placeholders in a Cloudflare baseUrl from provider env or process.env. */
+export function resolveCloudflareBaseUrl(model: Model<Api>, env?: ProviderEnv): string {
 	const url = model.baseUrl;
 	if (!url.includes("{")) return url;
 	const baseUrl = url.replace(/\{([A-Z_][A-Z0-9_]*)\}/g, (_match, name: string) => {
-		const value = process.env[name];
+		const value = getProviderEnvValue(name, env);
 		if (!value) {
 			throw new Error(`${name} is required for provider ${model.provider} but is not set.`);
 		}

packages/ai/src/api/google-generative-ai.ts

@@ -406,7 +406,8 @@ function isGemini3ProModel(model: Model<"google-generative-ai">): boolean {
 }
 
 function isGemini3FlashModel(model: Model<"google-generative-ai">): boolean {
-	return /gemini-3(?:\.\d+)?-flash/.test(model.id.toLowerCase());
+	const id = model.id.toLowerCase();
+	return /gemini-3(?:\.\d+)?-flash/.test(id) || id === "gemini-flash-latest" || id === "gemini-flash-lite-latest";
 }
 
 function getDisabledThinkingConfig(model: Model<"google-generative-ai">): ThinkingConfig {

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

@@ -14,6 +14,7 @@ import type {
 	Context,
 	Model,
 	ThinkingLevel as PiThinkingLevel,
+	ProviderEnv,
 	SimpleStreamOptions,
 	StreamFunction,
 	StreamOptions,
@@ -23,6 +24,7 @@ import type {
 	ToolCall,
 } from "../types.ts";
 import { AssistantMessageEventStream } from "../utils/event-stream.ts";
+import { getProviderEnvValue } from "../utils/provider-env.ts";
 import { sanitizeSurrogates } from "../utils/sanitize-unicode.ts";
 import type { GoogleThinkingLevel } from "./google-shared.ts";
 import {
@@ -91,7 +93,7 @@ export const stream: StreamFunction<"google-vertex", GoogleVertexOptions> = (
 			// Create the client using either a Vertex API key, if provided, or ADC with project and location
 			const client = apiKey
 				? createClientWithApiKey(model, apiKey, options?.headers)
-				: createClient(model, resolveProject(options), resolveLocation(options), options?.headers);
+				: createClient(model, resolveProject(options), resolveLocation(options), options?.headers, options?.env);
 			let params = buildParams(model, context, options);
 			const nextParams = await options?.onPayload?.(params, model);
 			if (nextParams !== undefined) {
@@ -333,12 +335,15 @@ function createClient(
 	project: string,
 	location: string,
 	optionsHeaders?: Record<string, string>,
+	env?: ProviderEnv,
 ): GoogleGenAI {
+	const googleAuthOptions = buildGoogleAuthOptions(env);
 	return new GoogleGenAI({
 		vertexai: true,
 		project,
 		location,
 		apiVersion: API_VERSION,
+		...(googleAuthOptions ? { googleAuthOptions } : {}),
 		httpOptions: buildHttpOptions(model, optionsHeaders),
 	});
 }
@@ -394,6 +399,11 @@ function baseUrlIncludesApiVersion(baseUrl: string): boolean {
 	}
 }
 
+function buildGoogleAuthOptions(env?: ProviderEnv): { keyFilename: string } | undefined {
+	const keyFilename = getProviderEnvValue("GOOGLE_APPLICATION_CREDENTIALS", env);
+	return keyFilename ? { keyFilename } : undefined;
+}
+
 function resolveApiKey(options?: GoogleVertexOptions): string | undefined {
 	const apiKey = options?.apiKey?.trim();
 	if (!apiKey || apiKey === GCP_VERTEX_CREDENTIALS_MARKER || isPlaceholderApiKey(apiKey)) {
@@ -407,7 +417,10 @@ function isPlaceholderApiKey(apiKey: string): boolean {
 }
 
 function resolveProject(options?: GoogleVertexOptions): string {
-	const project = options?.project || process.env.GOOGLE_CLOUD_PROJECT || process.env.GCLOUD_PROJECT;
+	const project =
+		options?.project ||
+		getProviderEnvValue("GOOGLE_CLOUD_PROJECT", options?.env) ||
+		getProviderEnvValue("GCLOUD_PROJECT", options?.env);
 	if (!project) {
 		throw new Error(
 			"Vertex AI requires a project ID. Set GOOGLE_CLOUD_PROJECT/GCLOUD_PROJECT or pass project in options.",
@@ -417,7 +430,7 @@ function resolveProject(options?: GoogleVertexOptions): string {
 }
 
 function resolveLocation(options?: GoogleVertexOptions): string {
-	const location = options?.location || process.env.GOOGLE_CLOUD_LOCATION;
+	const location = options?.location || getProviderEnvValue("GOOGLE_CLOUD_LOCATION", options?.env);
 	if (!location) {
 		throw new Error("Vertex AI requires a location. Set GOOGLE_CLOUD_LOCATION or pass location in options.");
 	}
@@ -490,7 +503,8 @@ function isGemini3ProModel(model: Model<"google-generative-ai">): boolean {
 }
 
 function isGemini3FlashModel(model: Model<"google-generative-ai">): boolean {
-	return /gemini-3(?:\.\d+)?-flash/.test(model.id.toLowerCase());
+	const id = model.id.toLowerCase();
+	return /gemini-3(?:\.\d+)?-flash/.test(id) || id === "gemini-flash-latest" || id === "gemini-flash-lite-latest";
 }
 
 function getDisabledThinkingConfig(model: Model<"google-vertex">): ThinkingConfig {

packages/ai/src/api/mistral-conversations.ts

@@ -226,7 +226,7 @@ function buildRequestOptions(model: Model<"mistral-conversations">, options?: Mi
 
 	// Mistral infrastructure uses `x-affinity` for KV-cache reuse (prefix caching).
 	// Respect explicit caller-provided header values.
-	if (options?.sessionId && !headers["x-affinity"]) {
+	if (shouldUsePromptCaching(options) && !headers["x-affinity"]) {
 		headers["x-affinity"] = options.sessionId;
 	}
 
@@ -255,6 +255,7 @@ function buildChatPayload(
 	if (options?.toolChoice) payload.toolChoice = mapToolChoice(options.toolChoice);
 	if (options?.promptMode) payload.promptMode = options.promptMode;
 	if (options?.reasoningEffort) payload.reasoningEffort = options.reasoningEffort;
+	if (shouldUsePromptCaching(options)) payload.promptCacheKey = options.sessionId;
 
 	if (context.systemPrompt) {
 		payload.messages.unshift({
@@ -266,6 +267,31 @@ function buildChatPayload(
 	return payload;
 }
 
+function shouldUsePromptCaching(options?: MistralOptions): options is MistralOptions & { sessionId: string } {
+	return options?.cacheRetention !== "none" && !!options?.sessionId;
+}
+
+function getMistralCachedPromptTokens(usage: unknown, promptTokens: number): number {
+	const rawUsage = usage as {
+		promptTokensDetails?: { cachedTokens?: unknown } | null;
+		prompt_tokens_details?: { cached_tokens?: unknown } | null;
+		promptTokenDetails?: { cachedTokens?: unknown } | null;
+		prompt_token_details?: { cached_tokens?: unknown } | null;
+		numCachedTokens?: unknown;
+		num_cached_tokens?: unknown;
+	};
+	const rawCachedTokens =
+		rawUsage.promptTokensDetails?.cachedTokens ??
+		rawUsage.prompt_tokens_details?.cached_tokens ??
+		rawUsage.promptTokenDetails?.cachedTokens ??
+		rawUsage.prompt_token_details?.cached_tokens ??
+		rawUsage.numCachedTokens ??
+		rawUsage.num_cached_tokens ??
+		0;
+	const cachedTokens = typeof rawCachedTokens === "number" && Number.isFinite(rawCachedTokens) ? rawCachedTokens : 0;
+	return Math.min(promptTokens, Math.max(0, cachedTokens));
+}
+
 async function consumeChatStream(
 	model: Model<"mistral-conversations">,
 	output: AssistantMessage,
@@ -305,11 +331,16 @@ async function consumeChatStream(
 		output.responseId ||= chunk.id;
 
 		if (chunk.usage) {
-			output.usage.input = chunk.usage.promptTokens || 0;
+			const promptTokens = chunk.usage.promptTokens || 0;
+			const cachedPromptTokens = getMistralCachedPromptTokens(chunk.usage, promptTokens);
+
+			output.usage.input = Math.max(0, promptTokens - cachedPromptTokens);
 			output.usage.output = chunk.usage.completionTokens || 0;
-			output.usage.cacheRead = 0;
+			output.usage.cacheRead = cachedPromptTokens;
 			output.usage.cacheWrite = 0;
-			output.usage.totalTokens = chunk.usage.totalTokens || output.usage.input + output.usage.output;
+			output.usage.totalTokens =
+				chunk.usage.totalTokens ||
+				output.usage.input + output.usage.output + output.usage.cacheRead + output.usage.cacheWrite;
 			calculateCost(model, output.usage);
 		}
 

packages/ai/src/api/openai-codex-responses.ts

@@ -27,6 +27,7 @@ import type {
 	AssistantMessage,
 	Context,
 	Model,
+	ProviderEnv,
 	SimpleStreamOptions,
 	StreamFunction,
 	StreamOptions,
@@ -40,6 +41,7 @@ import {
 } from "../utils/diagnostics.ts";
 import { AssistantMessageEventStream } from "../utils/event-stream.ts";
 import { headersToRecord } from "../utils/headers.ts";
+import { resolveHttpProxyUrlForTarget } from "../utils/node-http-proxy.ts";
 import { clampOpenAIPromptCacheKey } from "./openai-prompt-cache.ts";
 import { convertResponsesMessages, convertResponsesTools, processResponsesStream } from "./openai-responses-shared.ts";
 import { buildBaseOptions } from "./simple-options.ts";
@@ -53,7 +55,9 @@ const JWT_CLAIM_PATH = "https://api.openai.com/auth" as const;
 const DEFAULT_MAX_RETRIES = 0;
 const BASE_DELAY_MS = 1000;
 const DEFAULT_MAX_RETRY_DELAY_MS = 60_000;
-const DEFAULT_SSE_HEADER_TIMEOUT_MS = 10_000;
+// Keep a bounded pre-header timeout so zero-event Codex SSE stalls fail instead of
+// leaving callers stuck on "Working..." indefinitely. See #4945.
+const DEFAULT_SSE_HEADER_TIMEOUT_MS = 20_000;
 const DEFAULT_WEBSOCKET_CONNECT_TIMEOUT_MS = 15_000;
 const CODEX_TOOL_CALL_PROVIDERS = new Set(["openai", "openai-codex", "opencode"]);
 const WEBSOCKET_MESSAGE_TOO_BIG_CLOSE_CODE = 1009;
@@ -812,19 +816,13 @@ type WebSocketConstructor = new (
 ) => WebSocketLike;
 
 let _cachedWebsocket: WebSocketConstructor | null = null;
-async function getWebSocketConstructor(): Promise<WebSocketConstructor | null> {
-	if (_cachedWebsocket) return _cachedWebsocket;
+async function getWebSocketConstructor(env?: ProviderEnv): Promise<WebSocketConstructor | null> {
+	if (!env && _cachedWebsocket) return _cachedWebsocket;
 
 	// bun doesn't respect http proxy envs, ref: https://github.com/oven-sh/bun/issues/15489
 	// TODO: remove this when bun supports proxy envs in websocket.
-	if (
-		process?.versions?.bun &&
-		(process.env.HTTP_PROXY || process.env.HTTPS_PROXY || process.env.http_proxy || process.env.https_proxy)
-	) {
-		const m = await dynamicImport("proxy-from-env");
-		const getProxyForUrl = (m as { getProxyForUrl: (url: string | object | URL) => string }).getProxyForUrl;
-
-		_cachedWebsocket = class extends WebSocket {
+	if (typeof process !== "undefined" && process.versions?.bun) {
+		const WebSocketWithProxy = class extends WebSocket {
 			constructor(url: string | URL, options?: string | string[] | Record<string, unknown>) {
 				let _opts: Record<string, unknown> = {};
 				if (Array.isArray(options) || typeof options === "string") {
@@ -833,11 +831,17 @@ async function getWebSocketConstructor(): Promise<WebSocketConstructor | null> {
 					_opts = { ...options };
 				}
 
-				const proxy = getProxyForUrl(url.toString().replace(/^wss:/, "https:").replace(/^ws:/, "http:"));
-				super(url, { ..._opts, ...(proxy ? { proxy } : {}) } as any);
+				const proxyUrl = resolveHttpProxyUrlForTarget(
+					url.toString().replace(/^wss:/, "https:").replace(/^ws:/, "http:"),
+					env,
+				);
+				super(url, { ..._opts, ...(proxyUrl ? { proxy: proxyUrl.toString() } : {}) } as any);
 			}
 		};
-		return _cachedWebsocket;
+		if (!env) {
+			_cachedWebsocket = WebSocketWithProxy;
+		}
+		return WebSocketWithProxy;
 	}
 
 	const ctor = (globalThis as { WebSocket?: unknown }).WebSocket;
@@ -892,8 +896,9 @@ async function connectWebSocket(
 	headers: Headers,
 	signal?: AbortSignal,
 	connectTimeoutMs = DEFAULT_WEBSOCKET_CONNECT_TIMEOUT_MS,
+	env?: ProviderEnv,
 ): Promise<WebSocketLike> {
-	const WebSocketCtor = await getWebSocketConstructor();
+	const WebSocketCtor = await getWebSocketConstructor(env);
 	if (!WebSocketCtor) {
 		throw new Error("WebSocket transport is not available in this runtime");
 	}
@@ -970,6 +975,7 @@ async function acquireWebSocket(
 	sessionId: string | undefined,
 	signal?: AbortSignal,
 	connectTimeoutMs?: number,
+	env?: ProviderEnv,
 ): Promise<{
 	socket: WebSocketLike;
 	entry?: CachedWebSocketConnection;
@@ -977,7 +983,7 @@ async function acquireWebSocket(
 	release: (options?: { keep?: boolean }) => void;
 }> {
 	if (!sessionId) {
-		const socket = await connectWebSocket(url, headers, signal, connectTimeoutMs);
+		const socket = await connectWebSocket(url, headers, signal, connectTimeoutMs, env);
 		return {
 			socket,
 			reused: false,
@@ -1009,7 +1015,7 @@ async function acquireWebSocket(
 			};
 		}
 		if (cached.busy) {
-			const socket = await connectWebSocket(url, headers, signal, connectTimeoutMs);
+			const socket = await connectWebSocket(url, headers, signal, connectTimeoutMs, env);
 			return {
 				socket,
 				reused: false,
@@ -1024,7 +1030,7 @@ async function acquireWebSocket(
 		}
 	}
 
-	const socket = await connectWebSocket(url, headers, signal, connectTimeoutMs);
+	const socket = await connectWebSocket(url, headers, signal, connectTimeoutMs, env);
 	const entry: CachedWebSocketConnection = { socket, busy: true };
 	websocketSessionCache.set(sessionId, entry);
 	return {
@@ -1310,6 +1316,7 @@ async function processWebSocketStream(
 		options?.sessionId,
 		options?.signal,
 		websocketConnectTimeoutMs,
+		options?.env,
 	);
 	let keepConnection = true;
 	const useCachedContext = options?.transport === "websocket-cached" || options?.transport === "auto";

packages/ai/src/api/openai-completions.ts

@@ -14,11 +14,13 @@ import { calculateCost, clampThinkingLevel } from "../models.ts";
 import type {
 	AssistantMessage,
 	CacheRetention,
+	ChatTemplateKwargValue,
 	Context,
 	ImageContent,
 	Message,
 	Model,
 	OpenAICompletionsCompat,
+	ProviderEnv,
 	SimpleStreamOptions,
 	StopReason,
 	StreamFunction,
@@ -32,6 +34,7 @@ import type {
 import { AssistantMessageEventStream } from "../utils/event-stream.ts";
 import { headersToRecord } from "../utils/headers.ts";
 import { parseStreamingJson } from "../utils/json-parse.ts";
+import { getProviderEnvValue } from "../utils/provider-env.ts";
 import { sanitizeSurrogates } from "../utils/sanitize-unicode.ts";
 import { isCloudflareProvider, resolveCloudflareBaseUrl } from "./cloudflare.ts";
 import { buildCopilotDynamicHeaders, hasCopilotVisionInput } from "./github-copilot-headers.ts";
@@ -74,6 +77,20 @@ function isImageContentBlock(block: { type: string }): block is ImageContent {
 	return block.type === "image";
 }
 
+function isEncryptedReasoningDetail(detail: unknown): detail is OpenAIEncryptedReasoningDetail {
+	if (typeof detail !== "object" || detail === null) {
+		return false;
+	}
+	const candidate = detail as Record<string, unknown>;
+	return (
+		candidate.type === "reasoning.encrypted" &&
+		typeof candidate.id === "string" &&
+		candidate.id.length > 0 &&
+		typeof candidate.data === "string" &&
+		candidate.data.length > 0
+	);
+}
+
 export interface OpenAICompletionsOptions extends StreamOptions {
 	toolChoice?: "auto" | "none" | "required" | { type: "function"; function: { name: string } };
 	reasoningEffort?: "minimal" | "low" | "medium" | "high" | "xhigh";
@@ -88,8 +105,16 @@ type ResolvedOpenAICompletionsCompat = Omit<Required<OpenAICompletionsCompat>, "
 	cacheControlFormat?: OpenAICompletionsCompat["cacheControlFormat"];
 };
 
+type ResolvedChatTemplateKwargValue = string | number | boolean | null;
+
 type ChatCompletionInstructionMessageParam = ChatCompletionDeveloperMessageParam | ChatCompletionSystemMessageParam;
 
+type OpenAIEncryptedReasoningDetail = {
+	type: "reasoning.encrypted";
+	id: string;
+	data: string;
+};
+
 type ChatCompletionTextPartWithCacheControl = ChatCompletionContentPartText & {
 	cache_control?: OpenAICompatCacheControl;
 };
@@ -98,11 +123,11 @@ type ChatCompletionToolWithCacheControl = OpenAI.Chat.Completions.ChatCompletion
 	cache_control?: OpenAICompatCacheControl;
 };
 
-function resolveCacheRetention(cacheRetention?: CacheRetention): CacheRetention {
+function resolveCacheRetention(cacheRetention?: CacheRetention, env?: ProviderEnv): CacheRetention {
 	if (cacheRetention) {
 		return cacheRetention;
 	}
-	if (typeof process !== "undefined" && process.env.PI_CACHE_RETENTION === "long") {
+	if (getProviderEnvValue("PI_CACHE_RETENTION", env) === "long") {
 		return "long";
 	}
 	return "short";
@@ -140,9 +165,9 @@ export const stream: StreamFunction<"openai-completions", OpenAICompletionsOptio
 				throw new Error(`No API key for provider: ${model.provider}`);
 			}
 			const compat = getCompat(model);
-			const cacheRetention = resolveCacheRetention(options?.cacheRetention);
+			const cacheRetention = resolveCacheRetention(options?.cacheRetention, options?.env);
 			const cacheSessionId = cacheRetention === "none" ? undefined : options?.sessionId;
-			const client = createClient(model, context, apiKey, options?.headers, cacheSessionId, compat);
+			const client = createClient(model, context, apiKey, options?.headers, cacheSessionId, compat, options?.env);
 			let params = buildParams(model, context, options, compat, cacheRetention);
 			const nextParams = await options?.onPayload?.(params, model);
 			if (nextParams !== undefined) {
@@ -171,6 +196,7 @@ export const stream: StreamFunction<"openai-completions", OpenAICompletionsOptio
 			let hasFinishReason = false;
 			const toolCallBlocksByIndex = new Map<number, StreamingToolCallBlock>();
 			const toolCallBlocksById = new Map<string, StreamingToolCallBlock>();
+			const pendingReasoningDetailsByToolCallId = new Map<string, string>();
 			const blocks = output.content as StreamingBlock[];
 			const getContentIndex = (block: StreamingBlock) => blocks.indexOf(block);
 			const finishBlock = (block: StreamingBlock) => {
@@ -226,6 +252,16 @@ export const stream: StreamFunction<"openai-completions", OpenAICompletionsOptio
 				}
 				return thinkingBlock;
 			};
+			const applyPendingReasoningDetail = (block: StreamingToolCallBlock) => {
+				if (!block.id) {
+					return;
+				}
+				const pendingReasoningDetail = pendingReasoningDetailsByToolCallId.get(block.id);
+				if (pendingReasoningDetail) {
+					block.thoughtSignature = pendingReasoningDetail;
+					pendingReasoningDetailsByToolCallId.delete(block.id);
+				}
+			};
 			const ensureToolCallBlock = (toolCall: StreamingToolCallDelta) => {
 				const streamIndex = typeof toolCall.index === "number" ? toolCall.index : undefined;
 				let block = streamIndex !== undefined ? toolCallBlocksByIndex.get(streamIndex) : undefined;
@@ -261,6 +297,7 @@ export const stream: StreamFunction<"openai-completions", OpenAICompletionsOptio
 				if (toolCall.id) {
 					toolCallBlocksById.set(toolCall.id, block);
 				}
+				applyPendingReasoningDetail(block);
 				return block;
 			};
 
@@ -370,15 +407,16 @@ export const stream: StreamFunction<"openai-completions", OpenAICompletionsOptio
 						}
 					}
 
-					const reasoningDetails = (choice.delta as any).reasoning_details;
-					if (reasoningDetails && Array.isArray(reasoningDetails)) {
+					const reasoningDetails = (choice.delta as { reasoning_details?: unknown }).reasoning_details;
+					if (Array.isArray(reasoningDetails)) {
 						for (const detail of reasoningDetails) {
-							if (detail.type === "reasoning.encrypted" && detail.id && detail.data) {
-								const matchingToolCall = output.content.find(
-									(b) => b.type === "toolCall" && b.id === detail.id,
-								) as ToolCall | undefined;
+							if (isEncryptedReasoningDetail(detail)) {
+								const serializedDetail = JSON.stringify(detail);
+								const matchingToolCall = toolCallBlocksById.get(detail.id);
 								if (matchingToolCall) {
-									matchingToolCall.thoughtSignature = JSON.stringify(detail);
+									matchingToolCall.thoughtSignature = serializedDetail;
+								} else {
+									pendingReasoningDetailsByToolCallId.set(detail.id, serializedDetail);
 								}
 							}
 						}
@@ -454,6 +492,7 @@ function createClient(
 	optionsHeaders?: Record<string, string>,
 	sessionId?: string,
 	compat: ResolvedOpenAICompletionsCompat = getCompat(model),
+	env?: ProviderEnv,
 ) {
 	const headers = { ...model.headers };
 	if (model.provider === "github-copilot") {
@@ -487,7 +526,7 @@ function createClient(
 
 	return new OpenAI({
 		apiKey,
-		baseURL: isCloudflareProvider(model.provider) ? resolveCloudflareBaseUrl(model) : model.baseUrl,
+		baseURL: isCloudflareProvider(model.provider) ? resolveCloudflareBaseUrl(model, env) : model.baseUrl,
 		dangerouslyAllowBrowser: true,
 		defaultHeaders,
 	});
@@ -498,7 +537,7 @@ function buildParams(
 	context: Context,
 	options?: OpenAICompletionsOptions,
 	compat: ResolvedOpenAICompletionsCompat = getCompat(model),
-	cacheRetention: CacheRetention = resolveCacheRetention(options?.cacheRetention),
+	cacheRetention: CacheRetention = resolveCacheRetention(options?.cacheRetention, options?.env),
 ) {
 	const messages = convertMessages(model, context, compat);
 	const cacheControl = getCompatCacheControl(compat, cacheRetention);
@@ -554,8 +593,18 @@ function buildParams(
 	}
 
 	if (compat.thinkingFormat === "zai" && model.reasoning) {
-		const zaiParams = params as typeof params & { thinking?: { type: "enabled" | "disabled" } };
+		const zaiParams = params as Omit<typeof params, "reasoning_effort"> & {
+			thinking?: { type: "enabled" | "disabled" };
+			reasoning_effort?: string;
+		};
 		zaiParams.thinking = { type: options?.reasoningEffort ? "enabled" : "disabled" };
+		if (options?.reasoningEffort && compat.supportsReasoningEffort) {
+			const mappedEffort = model.thinkingLevelMap?.[options.reasoningEffort];
+			const effort = mappedEffort === undefined ? options.reasoningEffort : mappedEffort;
+			if (typeof effort === "string") {
+				zaiParams.reasoning_effort = effort;
+			}
+		}
 	} else if (compat.thinkingFormat === "qwen" && model.reasoning) {
 		(params as any).enable_thinking = !!options?.reasoningEffort;
 	} else if (compat.thinkingFormat === "qwen-chat-template" && model.reasoning) {
@@ -563,8 +612,17 @@ function buildParams(
 			enable_thinking: !!options?.reasoningEffort,
 			preserve_thinking: true,
 		};
+	} else if (compat.thinkingFormat === "chat-template" && model.reasoning) {
+		const chatTemplateKwargs = buildChatTemplateKwargs(model, options, compat);
+		if (chatTemplateKwargs) {
+			(params as any).chat_template_kwargs = chatTemplateKwargs;
+		}
 	} else if (compat.thinkingFormat === "deepseek" && model.reasoning) {
-		(params as any).thinking = { type: options?.reasoningEffort ? "enabled" : "disabled" };
+		if (options?.reasoningEffort) {
+			(params as any).thinking = { type: "enabled" };
+		} else if (model.thinkingLevelMap?.off !== null) {
+			(params as any).thinking = { type: "disabled" };
+		}
 		if (options?.reasoningEffort && compat.supportsReasoningEffort) {
 			(params as any).reasoning_effort =
 				model.thinkingLevelMap?.[options.reasoningEffort] ?? options.reasoningEffort;
@@ -629,6 +687,44 @@ function buildParams(
 	return params;
 }
 
+function buildChatTemplateKwargs(
+	model: Model<"openai-completions">,
+	options: OpenAICompletionsOptions | undefined,
+	compat: ResolvedOpenAICompletionsCompat,
+): Record<string, ResolvedChatTemplateKwargValue> | undefined {
+	const kwargs: Record<string, ResolvedChatTemplateKwargValue> = {};
+
+	for (const [key, value] of Object.entries(compat.chatTemplateKwargs)) {
+		const resolved = resolveChatTemplateKwargValue(model, options, value);
+		if (resolved !== undefined) {
+			kwargs[key] = resolved;
+		}
+	}
+
+	return Object.keys(kwargs).length > 0 ? kwargs : undefined;
+}
+
+function resolveChatTemplateKwargValue(
+	model: Model<"openai-completions">,
+	options: OpenAICompletionsOptions | undefined,
+	value: ChatTemplateKwargValue,
+): ResolvedChatTemplateKwargValue | undefined {
+	if (typeof value !== "object" || value === null) {
+		return value;
+	}
+
+	const reasoningEffort = options?.reasoningEffort;
+	if (!reasoningEffort && value.omitWhenOff) {
+		return undefined;
+	}
+	if (value.$var === "thinking.enabled") {
+		return !!reasoningEffort;
+	}
+
+	const mappedValue = reasoningEffort ? model.thinkingLevelMap?.[reasoningEffort] : model.thinkingLevelMap?.off;
+	return mappedValue === undefined ? reasoningEffort : typeof mappedValue === "string" ? mappedValue : undefined;
+}
+
 function getCompatCacheControl(
 	compat: ResolvedOpenAICompletionsCompat,
 	cacheRetention: CacheRetention,
@@ -1141,6 +1237,7 @@ function detectCompat(model: Model<"openai-completions">): ResolvedOpenAIComplet
 							: "openai",
 		openRouterRouting: {},
 		vercelGatewayRouting: {},
+		chatTemplateKwargs: {},
 		zaiToolStream: false,
 		supportsStrictMode: !isMoonshot && !isTogether && !isCloudflareAiGateway && !isNvidia,
 		cacheControlFormat,
@@ -1179,6 +1276,7 @@ function getCompat(model: Model<"openai-completions">): ResolvedOpenAICompletion
 		thinkingFormat: model.compat.thinkingFormat ?? detected.thinkingFormat,
 		openRouterRouting: model.compat.openRouterRouting ?? {},
 		vercelGatewayRouting: model.compat.vercelGatewayRouting ?? detected.vercelGatewayRouting,
+		chatTemplateKwargs: model.compat.chatTemplateKwargs ?? detected.chatTemplateKwargs,
 		zaiToolStream: model.compat.zaiToolStream ?? detected.zaiToolStream,
 		supportsStrictMode: model.compat.supportsStrictMode ?? detected.supportsStrictMode,
 		cacheControlFormat: model.compat.cacheControlFormat ?? detected.cacheControlFormat,

packages/ai/src/api/openai-responses-shared.ts

@@ -456,7 +456,8 @@ export async function processResponsesStream<TApi extends Api>(
 				});
 				currentBlock = null;
 			} else if (item.type === "message" && currentBlock?.type === "text") {
-				currentBlock.text = item.content.map((c) => (c.type === "output_text" ? c.text : c.refusal)).join("");
+				currentBlock.text =
+					item.content?.map((c) => (c.type === "output_text" ? c.text : c.refusal)).join("") || "";
 				currentBlock.textSignature = encodeTextSignatureV1(item.id, item.phase ?? undefined);
 				stream.push({
 					type: "text_end",

packages/ai/src/api/openai-responses.ts

@@ -8,6 +8,7 @@ import type {
 	Context,
 	Model,
 	OpenAIResponsesCompat,
+	ProviderEnv,
 	SimpleStreamOptions,
 	StreamFunction,
 	StreamOptions,
@@ -15,6 +16,7 @@ import type {
 } from "../types.ts";
 import { AssistantMessageEventStream } from "../utils/event-stream.ts";
 import { headersToRecord } from "../utils/headers.ts";
+import { getProviderEnvValue } from "../utils/provider-env.ts";
 import { isCloudflareProvider, resolveCloudflareBaseUrl } from "./cloudflare.ts";
 import { buildCopilotDynamicHeaders, hasCopilotVisionInput } from "./github-copilot-headers.ts";
 import { clampOpenAIPromptCacheKey } from "./openai-prompt-cache.ts";
@@ -27,11 +29,11 @@ const OPENAI_TOOL_CALL_PROVIDERS = new Set(["openai", "openai-codex", "opencode"
  * Resolve cache retention preference.
  * Defaults to "short" and uses PI_CACHE_RETENTION for backward compatibility.
  */
-function resolveCacheRetention(cacheRetention?: CacheRetention): CacheRetention {
+function resolveCacheRetention(cacheRetention?: CacheRetention, env?: ProviderEnv): CacheRetention {
 	if (cacheRetention) {
 		return cacheRetention;
 	}
-	if (typeof process !== "undefined" && process.env.PI_CACHE_RETENTION === "long") {
+	if (getProviderEnvValue("PI_CACHE_RETENTION", env) === "long") {
 		return "long";
 	}
 	return "short";
@@ -111,9 +113,9 @@ export const stream: StreamFunction<"openai-responses", OpenAIResponsesOptions>
 			if (!apiKey) {
 				throw new Error(`No API key for provider: ${model.provider}`);
 			}
-			const cacheRetention = resolveCacheRetention(options?.cacheRetention);
+			const cacheRetention = resolveCacheRetention(options?.cacheRetention, options?.env);
 			const cacheSessionId = cacheRetention === "none" ? undefined : options?.sessionId;
-			const client = createClient(model, context, apiKey, options?.headers, cacheSessionId);
+			const client = createClient(model, context, apiKey, options?.headers, cacheSessionId, options?.env);
 			let params = buildParams(model, context, options);
 			const nextParams = await options?.onPayload?.(params, model);
 			if (nextParams !== undefined) {
@@ -185,6 +187,7 @@ function createClient(
 	apiKey: string,
 	optionsHeaders?: Record<string, string>,
 	sessionId?: string,
+	env?: ProviderEnv,
 ) {
 	const compat = getCompat(model);
 	const headers = { ...model.headers };
@@ -220,7 +223,7 @@ function createClient(
 
 	return new OpenAI({
 		apiKey,
-		baseURL: isCloudflareProvider(model.provider) ? resolveCloudflareBaseUrl(model) : model.baseUrl,
+		baseURL: isCloudflareProvider(model.provider) ? resolveCloudflareBaseUrl(model, env) : model.baseUrl,
 		dangerouslyAllowBrowser: true,
 		defaultHeaders,
 	});
@@ -229,7 +232,7 @@ function createClient(
 function buildParams(model: Model<"openai-responses">, context: Context, options?: OpenAIResponsesOptions) {
 	const messages = convertResponsesMessages(model, context, OPENAI_TOOL_CALL_PROVIDERS);
 
-	const cacheRetention = resolveCacheRetention(options?.cacheRetention);
+	const cacheRetention = resolveCacheRetention(options?.cacheRetention, options?.env);
 	const compat = getCompat(model);
 	const params: ResponseCreateParamsStreaming = {
 		model: model.id,

packages/ai/src/api/simple-options.ts

@@ -17,6 +17,7 @@ export function buildBaseOptions(_model: Model<Api>, options?: SimpleStreamOptio
 		maxRetries: options?.maxRetries,
 		maxRetryDelayMs: options?.maxRetryDelayMs,
 		metadata: options?.metadata,
+		env: options?.env,
 	};
 }
 

packages/ai/src/env-api-keys.ts

@@ -23,44 +23,17 @@ if (typeof process !== "undefined" && (process.versions?.node || process.version
 	});
 }
 
-import type { KnownProvider } from "./types.ts";
-
-let _procEnvCache: Map<string, string> | null = null;
-
-/**
- * Fallback for https://github.com/oven-sh/bun/issues/27802
- * Bun compiled binaries have an empty `process.env` inside sandbox
- * environments on Linux. We can recover the env from `/proc/self/environ`.
- */
-function getProcEnv(key: string): string | undefined {
-	if (!process.versions?.bun) return undefined;
-	if (typeof process === "undefined") return undefined;
-
-	// If process.env already has entries, the bug is not triggered.
-	if (Object.keys(process.env).length > 0) return undefined;
-
-	if (_procEnvCache === null) {
-		_procEnvCache = new Map();
-		try {
-			const { readFileSync } = require("node:fs") as typeof import("node:fs");
-			const data = readFileSync("/proc/self/environ", "utf-8");
-			for (const entry of data.split("\0")) {
-				const idx = entry.indexOf("=");
-				if (idx > 0) {
-					_procEnvCache.set(entry.slice(0, idx), entry.slice(idx + 1));
-				}
-			}
-		} catch {
-			// /proc/self/environ may not be readable.
-		}
-	}
-
-	return _procEnvCache.get(key);
-}
+import type { KnownProvider, ProviderEnv } from "./types.ts";
+import { getProviderEnvValue } from "./utils/provider-env.ts";
 
 let cachedVertexAdcCredentialsExists: boolean | null = null;
 
-function hasVertexAdcCredentials(): boolean {
+function hasVertexAdcCredentials(env?: ProviderEnv): boolean {
+	const explicitCredentialsPath = env?.GOOGLE_APPLICATION_CREDENTIALS;
+	if (explicitCredentialsPath) {
+		return _existsSync ? _existsSync(explicitCredentialsPath) : false;
+	}
+
 	if (cachedVertexAdcCredentialsExists === null) {
 		// If node modules haven't loaded yet (async import race at startup),
 		// return false WITHOUT caching so the next call retries once they're ready.
@@ -75,7 +48,7 @@ function hasVertexAdcCredentials(): boolean {
 		}
 
 		// Check GOOGLE_APPLICATION_CREDENTIALS env var first (standard way)
-		const gacPath = process.env.GOOGLE_APPLICATION_CREDENTIALS || getProcEnv("GOOGLE_APPLICATION_CREDENTIALS");
+		const gacPath = getProviderEnvValue("GOOGLE_APPLICATION_CREDENTIALS", env);
 		if (gacPath) {
 			cachedVertexAdcCredentialsExists = _existsSync(gacPath);
 		} else {
@@ -143,13 +116,13 @@ function getApiKeyEnvVars(provider: string): readonly string[] | undefined {
  * credential sources such as AWS profiles, AWS IAM credentials, and Google
  * Application Default Credentials.
  */
-export function findEnvKeys(provider: KnownProvider): string[] | undefined;
-export function findEnvKeys(provider: string): string[] | undefined;
-export function findEnvKeys(provider: string): string[] | undefined {
+export function findEnvKeys(provider: KnownProvider, env?: ProviderEnv): string[] | undefined;
+export function findEnvKeys(provider: string, env?: ProviderEnv): string[] | undefined;
+export function findEnvKeys(provider: string, env?: ProviderEnv): string[] | undefined {
 	const envVars = getApiKeyEnvVars(provider);
 	if (!envVars) return undefined;
 
-	const found = envVars.filter((envVar) => !!process.env[envVar] || !!getProcEnv(envVar));
+	const found = envVars.filter((envVar) => !!getProviderEnvValue(envVar, env));
 	return found.length > 0 ? found : undefined;
 }
 
@@ -158,25 +131,22 @@ export function findEnvKeys(provider: string): string[] | undefined {
  *
  * Will not return API keys for providers that require OAuth tokens.
  */
-export function getEnvApiKey(provider: KnownProvider): string | undefined;
-export function getEnvApiKey(provider: string): string | undefined;
-export function getEnvApiKey(provider: string): string | undefined {
-	const envKeys = findEnvKeys(provider);
+export function getEnvApiKey(provider: KnownProvider, env?: ProviderEnv): string | undefined;
+export function getEnvApiKey(provider: string, env?: ProviderEnv): string | undefined;
+export function getEnvApiKey(provider: string, env?: ProviderEnv): string | undefined {
+	const envKeys = findEnvKeys(provider, env);
 	if (envKeys?.[0]) {
-		return process.env[envKeys[0]] || getProcEnv(envKeys[0]);
+		return getProviderEnvValue(envKeys[0], env);
 	}
 
 	// Vertex AI supports either an explicit API key or Application Default Credentials.
 	// Auth is configured via `gcloud auth application-default login`.
 	if (provider === "google-vertex") {
-		const hasCredentials = hasVertexAdcCredentials();
+		const hasCredentials = hasVertexAdcCredentials(env);
 		const hasProject = !!(
-			process.env.GOOGLE_CLOUD_PROJECT ||
-			process.env.GCLOUD_PROJECT ||
-			getProcEnv("GOOGLE_CLOUD_PROJECT") ||
-			getProcEnv("GCLOUD_PROJECT")
+			getProviderEnvValue("GOOGLE_CLOUD_PROJECT", env) || getProviderEnvValue("GCLOUD_PROJECT", env)
 		);
-		const hasLocation = !!(process.env.GOOGLE_CLOUD_LOCATION || getProcEnv("GOOGLE_CLOUD_LOCATION"));
+		const hasLocation = !!getProviderEnvValue("GOOGLE_CLOUD_LOCATION", env);
 
 		if (hasCredentials && hasProject && hasLocation) {
 			return "<authenticated>";
@@ -192,18 +162,12 @@ export function getEnvApiKey(provider: string): string | undefined {
 		// 5. AWS_CONTAINER_CREDENTIALS_FULL_URI - ECS task roles (full URI)
 		// 6. AWS_WEB_IDENTITY_TOKEN_FILE - IRSA (IAM Roles for Service Accounts)
 		if (
-			process.env.AWS_PROFILE ||
-			(process.env.AWS_ACCESS_KEY_ID && process.env.AWS_SECRET_ACCESS_KEY) ||
-			process.env.AWS_BEARER_TOKEN_BEDROCK ||
-			process.env.AWS_CONTAINER_CREDENTIALS_RELATIVE_URI ||
-			process.env.AWS_CONTAINER_CREDENTIALS_FULL_URI ||
-			process.env.AWS_WEB_IDENTITY_TOKEN_FILE ||
-			getProcEnv("AWS_PROFILE") ||
-			(getProcEnv("AWS_ACCESS_KEY_ID") && getProcEnv("AWS_SECRET_ACCESS_KEY")) ||
-			getProcEnv("AWS_BEARER_TOKEN_BEDROCK") ||
-			getProcEnv("AWS_CONTAINER_CREDENTIALS_RELATIVE_URI") ||
-			getProcEnv("AWS_CONTAINER_CREDENTIALS_FULL_URI") ||
-			getProcEnv("AWS_WEB_IDENTITY_TOKEN_FILE")
+			getProviderEnvValue("AWS_PROFILE", env) ||
+			(getProviderEnvValue("AWS_ACCESS_KEY_ID", env) && getProviderEnvValue("AWS_SECRET_ACCESS_KEY", env)) ||
+			getProviderEnvValue("AWS_BEARER_TOKEN_BEDROCK", env) ||
+			getProviderEnvValue("AWS_CONTAINER_CREDENTIALS_RELATIVE_URI", env) ||
+			getProviderEnvValue("AWS_CONTAINER_CREDENTIALS_FULL_URI", env) ||
+			getProviderEnvValue("AWS_WEB_IDENTITY_TOKEN_FILE", env)
 		) {
 			return "<authenticated>";
 		}

packages/ai/src/image-models.generated.ts

@@ -95,6 +95,21 @@ export const IMAGE_MODELS = {
 				cacheWrite: 0.08333333333333334,
 			},
 		} satisfies ImagesModel<"openrouter-images">,
+		"google/gemini-3-pro-image": {
+			id: "google/gemini-3-pro-image",
+			name: "Google: Nano Banana Pro (Gemini 3 Pro Image)",
+			api: "openrouter-images",
+			provider: "openrouter",
+			baseUrl: "https://openrouter.ai/api/v1",
+			input: ["image", "text"],
+			output: ["image", "text"],
+			cost: {
+				input: 2,
+				output: 12,
+				cacheRead: 0.19999999999999998,
+				cacheWrite: 0.375,
+			},
+		} satisfies ImagesModel<"openrouter-images">,
 		"google/gemini-3-pro-image-preview": {
 			id: "google/gemini-3-pro-image-preview",
 			name: "Google: Nano Banana Pro (Gemini 3 Pro Image Preview)",
@@ -110,6 +125,21 @@ export const IMAGE_MODELS = {
 				cacheWrite: 0.375,
 			},
 		} satisfies ImagesModel<"openrouter-images">,
+		"google/gemini-3.1-flash-image": {
+			id: "google/gemini-3.1-flash-image",
+			name: "Google: Nano Banana 2 (Gemini 3.1 Flash Image)",
+			api: "openrouter-images",
+			provider: "openrouter",
+			baseUrl: "https://openrouter.ai/api/v1",
+			input: ["image", "text"],
+			output: ["image", "text"],
+			cost: {
+				input: 0.5,
+				output: 3,
+				cacheRead: 0,
+				cacheWrite: 0,
+			},
+		} satisfies ImagesModel<"openrouter-images">,
 		"google/gemini-3.1-flash-image-preview": {
 			id: "google/gemini-3.1-flash-image-preview",
 			name: "Google: Nano Banana 2 (Gemini 3.1 Flash Image Preview)",

packages/ai/src/models.ts

@@ -372,10 +372,13 @@ export function hasApi<TApi extends Api>(model: Model<Api>, api: TApi): model is
 }
 
 export function calculateCost<TApi extends Api>(model: Model<TApi>, usage: Usage): Usage["cost"] {
+	// Anthropic charges 2x base input for 1h cache writes.
+	const longWrite = usage.cacheWrite1h ?? 0;
+	const shortWrite = usage.cacheWrite - longWrite;
 	usage.cost.input = (model.cost.input / 1000000) * usage.input;
 	usage.cost.output = (model.cost.output / 1000000) * usage.output;
 	usage.cost.cacheRead = (model.cost.cacheRead / 1000000) * usage.cacheRead;
-	usage.cost.cacheWrite = (model.cost.cacheWrite / 1000000) * usage.cacheWrite;
+	usage.cost.cacheWrite = (model.cost.cacheWrite * shortWrite + model.cost.input * 2 * longWrite) / 1000000;
 	usage.cost.total = usage.cost.input + usage.cost.output + usage.cost.cacheRead + usage.cost.cacheWrite;
 	return usage.cost;
 }

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

@@ -13,30 +13,13 @@ export const CEREBRAS_MODELS = {
 		reasoning: true,
 		input: ["text"],
 		cost: {
-			input: 0.25,
-			output: 0.69,
+			input: 0.35,
+			output: 0.75,
 			cacheRead: 0,
 			cacheWrite: 0,
 		},
 		contextWindow: 131072,
-		maxTokens: 32768,
-	} satisfies Model<"openai-completions">,
-	"llama3.1-8b": {
-		id: "llama3.1-8b",
-		name: "Llama 3.1 8B",
-		api: "openai-completions",
-		provider: "cerebras",
-		baseUrl: "https://api.cerebras.ai/v1",
-		reasoning: false,
-		input: ["text"],
-		cost: {
-			input: 0.1,
-			output: 0.1,
-			cacheRead: 0,
-			cacheWrite: 0,
-		},
-		contextWindow: 32000,
-		maxTokens: 8000,
+		maxTokens: 40960,
 	} satisfies Model<"openai-completions">,
 	"zai-glm-4.7": {
 		id: "zai-glm-4.7",
@@ -44,7 +27,7 @@ export const CEREBRAS_MODELS = {
 		api: "openai-completions",
 		provider: "cerebras",
 		baseUrl: "https://api.cerebras.ai/v1",
-		reasoning: false,
+		reasoning: true,
 		input: ["text"],
 		cost: {
 			input: 2.25,
@@ -53,6 +36,6 @@ export const CEREBRAS_MODELS = {
 			cacheWrite: 0,
 		},
 		contextWindow: 131072,
-		maxTokens: 40000,
+		maxTokens: 40960,
 	} satisfies Model<"openai-completions">,
 } as const;

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

@@ -112,6 +112,24 @@ export const CLOUDFLARE_WORKERS_AI_MODELS = {
 		contextWindow: 262144,
 		maxTokens: 256000,
 	} satisfies Model<"openai-completions">,
+	"@cf/moonshotai/kimi-k2.7-code": {
+		id: "@cf/moonshotai/kimi-k2.7-code",
+		name: "Kimi K2.7 Code",
+		api: "openai-completions",
+		provider: "cloudflare-workers-ai",
+		baseUrl: "https://api.cloudflare.com/client/v4/accounts/{CLOUDFLARE_ACCOUNT_ID}/ai/v1",
+		compat: {"sendSessionAffinityHeaders":true},
+		reasoning: true,
+		input: ["text", "image"],
+		cost: {
+			input: 0.95,
+			output: 4,
+			cacheRead: 0.19,
+			cacheWrite: 0,
+		},
+		contextWindow: 262144,
+		maxTokens: 262144,
+	} satisfies Model<"openai-completions">,
 	"@cf/nvidia/nemotron-3-120b-a12b": {
 		id: "@cf/nvidia/nemotron-3-120b-a12b",
 		name: "Nemotron 3 Super 120B",
@@ -202,4 +220,22 @@ export const CLOUDFLARE_WORKERS_AI_MODELS = {
 		contextWindow: 131072,
 		maxTokens: 131072,
 	} satisfies Model<"openai-completions">,
+	"@cf/zai-org/glm-5.2": {
+		id: "@cf/zai-org/glm-5.2",
+		name: "Glm 5.2",
+		api: "openai-completions",
+		provider: "cloudflare-workers-ai",
+		baseUrl: "https://api.cloudflare.com/client/v4/accounts/{CLOUDFLARE_ACCOUNT_ID}/ai/v1",
+		compat: {"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/fireworks.models.ts

@@ -16,7 +16,7 @@ export const FIREWORKS_MODELS = {
 		cost: {
 			input: 0.14,
 			output: 0.28,
-			cacheRead: 0.03,
+			cacheRead: 0.028,
 			cacheWrite: 0,
 		},
 		contextWindow: 1000000,
@@ -58,6 +58,25 @@ export const FIREWORKS_MODELS = {
 		contextWindow: 202800,
 		maxTokens: 131072,
 	} satisfies Model<"anthropic-messages">,
+	"accounts/fireworks/models/glm-5p2": {
+		id: "accounts/fireworks/models/glm-5p2",
+		name: "GLM 5.2",
+		api: "openai-completions",
+		provider: "fireworks",
+		baseUrl: "https://api.fireworks.ai/inference/v1",
+		compat: {"supportsStore":false,"supportsDeveloperRole":false},
+		reasoning: true,
+		thinkingLevelMap: {"off":"none","minimal":null,"low":"high","medium":"high","xhigh":"max"},
+		input: ["text"],
+		cost: {
+			input: 1.4,
+			output: 4.4,
+			cacheRead: 0.26,
+			cacheWrite: 0,
+		},
+		contextWindow: 1048576,
+		maxTokens: 131072,
+	} satisfies Model<"openai-completions">,
 	"accounts/fireworks/models/gpt-oss-120b": {
 		id: "accounts/fireworks/models/gpt-oss-120b",
 		name: "GPT OSS 120B",
@@ -94,24 +113,6 @@ export const FIREWORKS_MODELS = {
 		contextWindow: 131072,
 		maxTokens: 32768,
 	} satisfies Model<"anthropic-messages">,
-	"accounts/fireworks/models/kimi-k2p5": {
-		id: "accounts/fireworks/models/kimi-k2p5",
-		name: "Kimi K2.5",
-		api: "anthropic-messages",
-		provider: "fireworks",
-		baseUrl: "https://api.fireworks.ai/inference",
-		compat: {"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false},
-		reasoning: true,
-		input: ["text", "image"],
-		cost: {
-			input: 0.6,
-			output: 3,
-			cacheRead: 0.1,
-			cacheWrite: 0,
-		},
-		contextWindow: 256000,
-		maxTokens: 256000,
-	} satisfies Model<"anthropic-messages">,
 	"accounts/fireworks/models/kimi-k2p6": {
 		id: "accounts/fireworks/models/kimi-k2p6",
 		name: "Kimi K2.6",
@@ -130,23 +131,23 @@ export const FIREWORKS_MODELS = {
 		contextWindow: 262000,
 		maxTokens: 262000,
 	} satisfies Model<"anthropic-messages">,
-	"accounts/fireworks/models/minimax-m2p5": {
-		id: "accounts/fireworks/models/minimax-m2p5",
-		name: "MiniMax-M2.5",
+	"accounts/fireworks/models/kimi-k2p7-code": {
+		id: "accounts/fireworks/models/kimi-k2p7-code",
+		name: "Kimi K2.7 Code",
 		api: "anthropic-messages",
 		provider: "fireworks",
 		baseUrl: "https://api.fireworks.ai/inference",
 		compat: {"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false},
 		reasoning: true,
-		input: ["text"],
+		input: ["text", "image"],
 		cost: {
-			input: 0.3,
-			output: 1.2,
-			cacheRead: 0.03,
+			input: 0.95,
+			output: 4,
+			cacheRead: 0.19,
 			cacheWrite: 0,
 		},
-		contextWindow: 196608,
-		maxTokens: 196608,
+		contextWindow: 262000,
+		maxTokens: 262000,
 	} satisfies Model<"anthropic-messages">,
 	"accounts/fireworks/models/minimax-m2p7": {
 		id: "accounts/fireworks/models/minimax-m2p7",
@@ -166,9 +167,27 @@ export const FIREWORKS_MODELS = {
 		contextWindow: 196608,
 		maxTokens: 196608,
 	} satisfies Model<"anthropic-messages">,
-	"accounts/fireworks/models/qwen3p6-plus": {
-		id: "accounts/fireworks/models/qwen3p6-plus",
-		name: "Qwen 3.6 Plus",
+	"accounts/fireworks/models/minimax-m3": {
+		id: "accounts/fireworks/models/minimax-m3",
+		name: "MiniMax-M3",
+		api: "anthropic-messages",
+		provider: "fireworks",
+		baseUrl: "https://api.fireworks.ai/inference",
+		compat: {"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false},
+		reasoning: true,
+		input: ["text"],
+		cost: {
+			input: 0.3,
+			output: 1.2,
+			cacheRead: 0.06,
+			cacheWrite: 0,
+		},
+		contextWindow: 512000,
+		maxTokens: 512000,
+	} satisfies Model<"anthropic-messages">,
+	"accounts/fireworks/models/qwen3p7-plus": {
+		id: "accounts/fireworks/models/qwen3p7-plus",
+		name: "Qwen 3.7 Plus",
 		api: "anthropic-messages",
 		provider: "fireworks",
 		baseUrl: "https://api.fireworks.ai/inference",
@@ -176,9 +195,9 @@ export const FIREWORKS_MODELS = {
 		reasoning: true,
 		input: ["text", "image"],
 		cost: {
-			input: 0.5,
-			output: 3,
-			cacheRead: 0.1,
+			input: 0.4,
+			output: 1.6,
+			cacheRead: 0.08,
 			cacheWrite: 0,
 		},
 		contextWindow: 262144,
@@ -238,4 +257,22 @@ export const FIREWORKS_MODELS = {
 		contextWindow: 262000,
 		maxTokens: 262000,
 	} satisfies Model<"anthropic-messages">,
+	"accounts/fireworks/routers/kimi-k2p7-code-fast": {
+		id: "accounts/fireworks/routers/kimi-k2p7-code-fast",
+		name: "Kimi K2.7 Code Fast",
+		api: "anthropic-messages",
+		provider: "fireworks",
+		baseUrl: "https://api.fireworks.ai/inference",
+		compat: {"sendSessionAffinityHeaders":true,"supportsEagerToolInputStreaming":false,"supportsCacheControlOnTools":false,"supportsLongCacheRetention":false},
+		reasoning: true,
+		input: ["text", "image"],
+		cost: {
+			input: 1.9,
+			output: 8,
+			cacheRead: 0.38,
+			cacheWrite: 0,
+		},
+		contextWindow: 262000,
+		maxTokens: 262000,
+	} satisfies Model<"anthropic-messages">,
 } as const;

packages/ai/src/providers/fireworks.ts

@@ -1,15 +1,19 @@
 import { anthropicMessagesApi } from "../api/anthropic-messages.lazy.ts";
+import { openAICompletionsApi } from "../api/openai-completions.lazy.ts";
 import { envApiKeyAuth } from "../auth/helpers.ts";
 import { createProvider, type Provider } from "../models.ts";
 import { FIREWORKS_MODELS } from "./fireworks.models.ts";
 
-export function fireworksProvider(): Provider<"anthropic-messages"> {
+export function fireworksProvider(): Provider<"anthropic-messages" | "openai-completions"> {
 	return createProvider({
 		id: "fireworks",
 		name: "Fireworks",
 		baseUrl: "https://api.fireworks.ai/inference",
 		auth: { apiKey: envApiKeyAuth("Fireworks API key", ["FIREWORKS_API_KEY"]) },
 		models: Object.values(FIREWORKS_MODELS),
-		api: anthropicMessagesApi(),
+		api: {
+			"anthropic-messages": anthropicMessagesApi(),
+			"openai-completions": openAICompletionsApi(),
+		},
 	});
 }

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

@@ -4,6 +4,25 @@
 import type { Model } from "../types.ts";
 
 export const GITHUB_COPILOT_MODELS = {
+	"claude-fable-5": {
+		id: "claude-fable-5",
+		name: "Claude Fable 5",
+		api: "openai-completions",
+		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"},
+		compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false},
+		reasoning: true,
+		input: ["text", "image"],
+		cost: {
+			input: 10,
+			output: 50,
+			cacheRead: 1,
+			cacheWrite: 12.5,
+		},
+		contextWindow: 1000000,
+		maxTokens: 128000,
+	} satisfies Model<"openai-completions">,
 	"claude-haiku-4.5": {
 		id: "claude-haiku-4.5",
 		name: "Claude Haiku 4.5 (latest)",
@@ -70,7 +89,7 @@ export const GITHUB_COPILOT_MODELS = {
 		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"},
 		compat: {"forceAdaptiveThinking":true,"supportsTemperature":false},
 		reasoning: true,
-		thinkingLevelMap: {"xhigh":"xhigh"},
+		thinkingLevelMap: {"xhigh":"xhigh","minimal":"low"},
 		input: ["text", "image"],
 		cost: {
 			input: 5,
@@ -90,7 +109,7 @@ export const GITHUB_COPILOT_MODELS = {
 		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"},
 		compat: {"forceAdaptiveThinking":true,"supportsTemperature":false},
 		reasoning: true,
-		thinkingLevelMap: {"xhigh":"xhigh"},
+		thinkingLevelMap: {"xhigh":"xhigh","minimal":"low"},
 		input: ["text", "image"],
 		cost: {
 			input: 5,
@@ -148,6 +167,7 @@ export const GITHUB_COPILOT_MODELS = {
 		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"},
 		compat: {"forceAdaptiveThinking":true},
 		reasoning: true,
+		thinkingLevelMap: {"minimal":"low","xhigh":"max"},
 		input: ["text", "image"],
 		cost: {
 			input: 3,
@@ -405,23 +425,4 @@ export const GITHUB_COPILOT_MODELS = {
 		contextWindow: 400000,
 		maxTokens: 128000,
 	} satisfies Model<"openai-responses">,
-	"raptor-mini": {
-		id: "raptor-mini",
-		name: "Raptor mini",
-		api: "openai-completions",
-		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"},
-		compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false},
-		reasoning: true,
-		input: ["text", "image"],
-		cost: {
-			input: 0.25,
-			output: 2,
-			cacheRead: 0.025,
-			cacheWrite: 0,
-		},
-		contextWindow: 400000,
-		maxTokens: 128000,
-	} satisfies Model<"openai-completions">,
 } as const;

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

@@ -4,94 +4,9 @@
 import type { Model } from "../types.ts";
 
 export const GOOGLE_VERTEX_MODELS = {
-	"gemini-1.5-flash": {
-		id: "gemini-1.5-flash",
-		name: "Gemini 1.5 Flash (Vertex)",
-		api: "google-vertex",
-		provider: "google-vertex",
-		baseUrl: "https://{location}-aiplatform.googleapis.com",
-		reasoning: false,
-		input: ["text", "image"],
-		cost: {
-			input: 0.075,
-			output: 0.3,
-			cacheRead: 0.01875,
-			cacheWrite: 0,
-		},
-		contextWindow: 1000000,
-		maxTokens: 8192,
-	} satisfies Model<"google-vertex">,
-	"gemini-1.5-flash-8b": {
-		id: "gemini-1.5-flash-8b",
-		name: "Gemini 1.5 Flash-8B (Vertex)",
-		api: "google-vertex",
-		provider: "google-vertex",
-		baseUrl: "https://{location}-aiplatform.googleapis.com",
-		reasoning: false,
-		input: ["text", "image"],
-		cost: {
-			input: 0.0375,
-			output: 0.15,
-			cacheRead: 0.01,
-			cacheWrite: 0,
-		},
-		contextWindow: 1000000,
-		maxTokens: 8192,
-	} satisfies Model<"google-vertex">,
-	"gemini-1.5-pro": {
-		id: "gemini-1.5-pro",
-		name: "Gemini 1.5 Pro (Vertex)",
-		api: "google-vertex",
-		provider: "google-vertex",
-		baseUrl: "https://{location}-aiplatform.googleapis.com",
-		reasoning: false,
-		input: ["text", "image"],
-		cost: {
-			input: 1.25,
-			output: 5,
-			cacheRead: 0.3125,
-			cacheWrite: 0,
-		},
-		contextWindow: 1000000,
-		maxTokens: 8192,
-	} satisfies Model<"google-vertex">,
-	"gemini-2.0-flash": {
-		id: "gemini-2.0-flash",
-		name: "Gemini 2.0 Flash (Vertex)",
-		api: "google-vertex",
-		provider: "google-vertex",
-		baseUrl: "https://{location}-aiplatform.googleapis.com",
-		reasoning: false,
-		input: ["text", "image"],
-		cost: {
-			input: 0.15,
-			output: 0.6,
-			cacheRead: 0.0375,
-			cacheWrite: 0,
-		},
-		contextWindow: 1048576,
-		maxTokens: 8192,
-	} satisfies Model<"google-vertex">,
-	"gemini-2.0-flash-lite": {
-		id: "gemini-2.0-flash-lite",
-		name: "Gemini 2.0 Flash Lite (Vertex)",
-		api: "google-vertex",
-		provider: "google-vertex",
-		baseUrl: "https://{location}-aiplatform.googleapis.com",
-		reasoning: true,
-		input: ["text", "image"],
-		cost: {
-			input: 0.075,
-			output: 0.3,
-			cacheRead: 0.01875,
-			cacheWrite: 0,
-		},
-		contextWindow: 1048576,
-		maxTokens: 65536,
-	} satisfies Model<"google-vertex">,
 	"gemini-2.5-flash": {
 		id: "gemini-2.5-flash",
-		name: "Gemini 2.5 Flash (Vertex)",
+		name: "Gemini 2.5 Flash",
 		api: "google-vertex",
 		provider: "google-vertex",
 		baseUrl: "https://{location}-aiplatform.googleapis.com",
@@ -108,24 +23,7 @@ export const GOOGLE_VERTEX_MODELS = {
 	} satisfies Model<"google-vertex">,
 	"gemini-2.5-flash-lite": {
 		id: "gemini-2.5-flash-lite",
-		name: "Gemini 2.5 Flash Lite (Vertex)",
-		api: "google-vertex",
-		provider: "google-vertex",
-		baseUrl: "https://{location}-aiplatform.googleapis.com",
-		reasoning: true,
-		input: ["text", "image"],
-		cost: {
-			input: 0.1,
-			output: 0.4,
-			cacheRead: 0.01,
-			cacheWrite: 0,
-		},
-		contextWindow: 1048576,
-		maxTokens: 65536,
-	} satisfies Model<"google-vertex">,
-	"gemini-2.5-flash-lite-preview-09-2025": {
-		id: "gemini-2.5-flash-lite-preview-09-2025",
-		name: "Gemini 2.5 Flash Lite Preview 09-25 (Vertex)",
+		name: "Gemini 2.5 Flash-Lite",
 		api: "google-vertex",
 		provider: "google-vertex",
 		baseUrl: "https://{location}-aiplatform.googleapis.com",
@@ -142,7 +40,7 @@ export const GOOGLE_VERTEX_MODELS = {
 	} satisfies Model<"google-vertex">,
 	"gemini-2.5-pro": {
 		id: "gemini-2.5-pro",
-		name: "Gemini 2.5 Pro (Vertex)",
+		name: "Gemini 2.5 Pro",
 		api: "google-vertex",
 		provider: "google-vertex",
 		baseUrl: "https://{location}-aiplatform.googleapis.com",
@@ -159,7 +57,7 @@ export const GOOGLE_VERTEX_MODELS = {
 	} satisfies Model<"google-vertex">,
 	"gemini-3-flash-preview": {
 		id: "gemini-3-flash-preview",
-		name: "Gemini 3 Flash Preview (Vertex)",
+		name: "Gemini 3 Flash Preview",
 		api: "google-vertex",
 		provider: "google-vertex",
 		baseUrl: "https://{location}-aiplatform.googleapis.com",
@@ -175,27 +73,27 @@ export const GOOGLE_VERTEX_MODELS = {
 		contextWindow: 1048576,
 		maxTokens: 65536,
 	} satisfies Model<"google-vertex">,
-	"gemini-3-pro-preview": {
-		id: "gemini-3-pro-preview",
-		name: "Gemini 3 Pro Preview (Vertex)",
+	"gemini-3.1-flash-lite": {
+		id: "gemini-3.1-flash-lite",
+		name: "Gemini 3.1 Flash Lite",
 		api: "google-vertex",
 		provider: "google-vertex",
 		baseUrl: "https://{location}-aiplatform.googleapis.com",
 		reasoning: true,
-		thinkingLevelMap: {"off":null,"minimal":null,"low":"LOW","medium":null,"high":"HIGH"},
+		thinkingLevelMap: {"off":null},
 		input: ["text", "image"],
 		cost: {
-			input: 2,
-			output: 12,
-			cacheRead: 0.2,
+			input: 0.25,
+			output: 1.5,
+			cacheRead: 0.025,
 			cacheWrite: 0,
 		},
-		contextWindow: 1000000,
-		maxTokens: 64000,
+		contextWindow: 1048576,
+		maxTokens: 65536,
 	} satisfies Model<"google-vertex">,
 	"gemini-3.1-pro-preview": {
 		id: "gemini-3.1-pro-preview",
-		name: "Gemini 3.1 Pro Preview (Vertex)",
+		name: "Gemini 3.1 Pro Preview",
 		api: "google-vertex",
 		provider: "google-vertex",
 		baseUrl: "https://{location}-aiplatform.googleapis.com",
@@ -213,7 +111,7 @@ export const GOOGLE_VERTEX_MODELS = {
 	} satisfies Model<"google-vertex">,
 	"gemini-3.1-pro-preview-customtools": {
 		id: "gemini-3.1-pro-preview-customtools",
-		name: "Gemini 3.1 Pro Preview Custom Tools (Vertex)",
+		name: "Gemini 3.1 Pro Preview Custom Tools",
 		api: "google-vertex",
 		provider: "google-vertex",
 		baseUrl: "https://{location}-aiplatform.googleapis.com",
@@ -229,4 +127,58 @@ export const GOOGLE_VERTEX_MODELS = {
 		contextWindow: 1048576,
 		maxTokens: 65536,
 	} satisfies Model<"google-vertex">,
+	"gemini-3.5-flash": {
+		id: "gemini-3.5-flash",
+		name: "Gemini 3.5 Flash",
+		api: "google-vertex",
+		provider: "google-vertex",
+		baseUrl: "https://{location}-aiplatform.googleapis.com",
+		reasoning: true,
+		thinkingLevelMap: {"off":null},
+		input: ["text", "image"],
+		cost: {
+			input: 1.5,
+			output: 9,
+			cacheRead: 0.15,
+			cacheWrite: 0,
+		},
+		contextWindow: 1048576,
+		maxTokens: 65536,
+	} satisfies Model<"google-vertex">,
+	"gemini-flash-latest": {
+		id: "gemini-flash-latest",
+		name: "Gemini Flash Latest",
+		api: "google-vertex",
+		provider: "google-vertex",
+		baseUrl: "https://{location}-aiplatform.googleapis.com",
+		reasoning: true,
+		thinkingLevelMap: {"off":null},
+		input: ["text", "image"],
+		cost: {
+			input: 1.5,
+			output: 9,
+			cacheRead: 0.15,
+			cacheWrite: 0,
+		},
+		contextWindow: 1048576,
+		maxTokens: 65536,
+	} satisfies Model<"google-vertex">,
+	"gemini-flash-lite-latest": {
+		id: "gemini-flash-lite-latest",
+		name: "Gemini Flash-Lite Latest",
+		api: "google-vertex",
+		provider: "google-vertex",
+		baseUrl: "https://{location}-aiplatform.googleapis.com",
+		reasoning: true,
+		thinkingLevelMap: {"off":null},
+		input: ["text", "image"],
+		cost: {
+			input: 0.25,
+			output: 1.5,
+			cacheRead: 0.025,
+			cacheWrite: 0,
+		},
+		contextWindow: 1048576,
+		maxTokens: 65536,
+	} satisfies Model<"google-vertex">,
 } as const;

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

@@ -222,11 +222,12 @@ export const GOOGLE_MODELS = {
 		provider: "google",
 		baseUrl: "https://generativelanguage.googleapis.com/v1beta",
 		reasoning: true,
+		thinkingLevelMap: {"off":null},
 		input: ["text", "image"],
 		cost: {
-			input: 0.3,
-			output: 2.5,
-			cacheRead: 0.075,
+			input: 1.5,
+			output: 9,
+			cacheRead: 0.15,
 			cacheWrite: 0,
 		},
 		contextWindow: 1048576,
@@ -239,10 +240,11 @@ export const GOOGLE_MODELS = {
 		provider: "google",
 		baseUrl: "https://generativelanguage.googleapis.com/v1beta",
 		reasoning: true,
+		thinkingLevelMap: {"off":null},
 		input: ["text", "image"],
 		cost: {
-			input: 0.1,
-			output: 0.4,
+			input: 0.25,
+			output: 1.5,
 			cacheRead: 0.025,
 			cacheWrite: 0,
 		},

packages/ai/src/providers/kimi-coding.models.ts

@@ -4,6 +4,24 @@
 import type { Model } from "../types.ts";
 
 export const KIMI_CODING_MODELS = {
+	"k2p7": {
+		id: "k2p7",
+		name: "Kimi K2.7 Code",
+		api: "anthropic-messages",
+		provider: "kimi-coding",
+		baseUrl: "https://api.kimi.com/coding",
+		headers: {"User-Agent":"KimiCLI/1.5"},
+		reasoning: true,
+		input: ["text", "image"],
+		cost: {
+			input: 0,
+			output: 0,
+			cacheRead: 0,
+			cacheWrite: 0,
+		},
+		contextWindow: 262144,
+		maxTokens: 32768,
+	} satisfies Model<"anthropic-messages">,
 	"kimi-for-coding": {
 		id: "kimi-for-coding",
 		name: "Kimi For Coding",

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

@@ -15,7 +15,7 @@ export const MISTRAL_MODELS = {
 		cost: {
 			input: 0.3,
 			output: 0.9,
-			cacheRead: 0,
+			cacheRead: 0.03,
 			cacheWrite: 0,
 		},
 		contextWindow: 256000,
@@ -32,7 +32,7 @@ export const MISTRAL_MODELS = {
 		cost: {
 			input: 0.4,
 			output: 2,
-			cacheRead: 0,
+			cacheRead: 0.04,
 			cacheWrite: 0,
 		},
 		contextWindow: 262144,
@@ -49,7 +49,7 @@ export const MISTRAL_MODELS = {
 		cost: {
 			input: 0.4,
 			output: 2,
-			cacheRead: 0,
+			cacheRead: 0.04,
 			cacheWrite: 0,
 		},
 		contextWindow: 262144,
@@ -66,7 +66,7 @@ export const MISTRAL_MODELS = {
 		cost: {
 			input: 0.4,
 			output: 2,
-			cacheRead: 0,
+			cacheRead: 0.04,
 			cacheWrite: 0,
 		},
 		contextWindow: 128000,
@@ -83,7 +83,7 @@ export const MISTRAL_MODELS = {
 		cost: {
 			input: 0.4,
 			output: 2,
-			cacheRead: 0,
+			cacheRead: 0.04,
 			cacheWrite: 0,
 		},
 		contextWindow: 262144,
@@ -100,7 +100,7 @@ export const MISTRAL_MODELS = {
 		cost: {
 			input: 0.1,
 			output: 0.3,
-			cacheRead: 0,
+			cacheRead: 0.01,
 			cacheWrite: 0,
 		},
 		contextWindow: 128000,
@@ -117,7 +117,7 @@ export const MISTRAL_MODELS = {
 		cost: {
 			input: 0.1,
 			output: 0.3,
-			cacheRead: 0,
+			cacheRead: 0.01,
 			cacheWrite: 0,
 		},
 		contextWindow: 128000,
@@ -151,7 +151,7 @@ export const MISTRAL_MODELS = {
 		cost: {
 			input: 2,
 			output: 5,
-			cacheRead: 0,
+			cacheRead: 0.2,
 			cacheWrite: 0,
 		},
 		contextWindow: 128000,
@@ -168,7 +168,7 @@ export const MISTRAL_MODELS = {
 		cost: {
 			input: 0.5,
 			output: 1.5,
-			cacheRead: 0,
+			cacheRead: 0.05,
 			cacheWrite: 0,
 		},
 		contextWindow: 128000,
@@ -185,7 +185,7 @@ export const MISTRAL_MODELS = {
 		cost: {
 			input: 0.04,
 			output: 0.04,
-			cacheRead: 0,
+			cacheRead: 0.004,
 			cacheWrite: 0,
 		},
 		contextWindow: 128000,
@@ -202,7 +202,7 @@ export const MISTRAL_MODELS = {
 		cost: {
 			input: 0.1,
 			output: 0.1,
-			cacheRead: 0,
+			cacheRead: 0.01,
 			cacheWrite: 0,
 		},
 		contextWindow: 128000,
@@ -219,7 +219,7 @@ export const MISTRAL_MODELS = {
 		cost: {
 			input: 2,
 			output: 6,
-			cacheRead: 0,
+			cacheRead: 0.2,
 			cacheWrite: 0,
 		},
 		contextWindow: 131072,
@@ -236,7 +236,7 @@ export const MISTRAL_MODELS = {
 		cost: {
 			input: 0.5,
 			output: 1.5,
-			cacheRead: 0,
+			cacheRead: 0.05,
 			cacheWrite: 0,
 		},
 		contextWindow: 262144,
@@ -253,7 +253,7 @@ export const MISTRAL_MODELS = {
 		cost: {
 			input: 0.5,
 			output: 1.5,
-			cacheRead: 0,
+			cacheRead: 0.05,
 			cacheWrite: 0,
 		},
 		contextWindow: 262144,
@@ -270,7 +270,7 @@ export const MISTRAL_MODELS = {
 		cost: {
 			input: 0.4,
 			output: 2,
-			cacheRead: 0,
+			cacheRead: 0.04,
 			cacheWrite: 0,
 		},
 		contextWindow: 131072,
@@ -287,7 +287,7 @@ export const MISTRAL_MODELS = {
 		cost: {
 			input: 0.4,
 			output: 2,
-			cacheRead: 0,
+			cacheRead: 0.04,
 			cacheWrite: 0,
 		},
 		contextWindow: 262144,
@@ -304,7 +304,7 @@ export const MISTRAL_MODELS = {
 		cost: {
 			input: 1.5,
 			output: 7.5,
-			cacheRead: 0,
+			cacheRead: 0.15,
 			cacheWrite: 0,
 		},
 		contextWindow: 262144,
@@ -338,7 +338,7 @@ export const MISTRAL_MODELS = {
 		cost: {
 			input: 0.4,
 			output: 2,
-			cacheRead: 0,
+			cacheRead: 0.04,
 			cacheWrite: 0,
 		},
 		contextWindow: 262144,
@@ -355,7 +355,7 @@ export const MISTRAL_MODELS = {
 		cost: {
 			input: 0.15,
 			output: 0.15,
-			cacheRead: 0,
+			cacheRead: 0.015,
 			cacheWrite: 0,
 		},
 		contextWindow: 128000,
@@ -372,7 +372,7 @@ export const MISTRAL_MODELS = {
 		cost: {
 			input: 0.1,
 			output: 0.3,
-			cacheRead: 0,
+			cacheRead: 0.01,
 			cacheWrite: 0,
 		},
 		contextWindow: 128000,
@@ -389,7 +389,7 @@ export const MISTRAL_MODELS = {
 		cost: {
 			input: 0.15,
 			output: 0.6,
-			cacheRead: 0,
+			cacheRead: 0.015,
 			cacheWrite: 0,
 		},
 		contextWindow: 256000,
@@ -406,7 +406,7 @@ export const MISTRAL_MODELS = {
 		cost: {
 			input: 0.15,
 			output: 0.6,
-			cacheRead: 0,
+			cacheRead: 0.015,
 			cacheWrite: 0,
 		},
 		contextWindow: 256000,
@@ -423,7 +423,7 @@ export const MISTRAL_MODELS = {
 		cost: {
 			input: 0.25,
 			output: 0.25,
-			cacheRead: 0,
+			cacheRead: 0.025,
 			cacheWrite: 0,
 		},
 		contextWindow: 8000,
@@ -440,7 +440,7 @@ export const MISTRAL_MODELS = {
 		cost: {
 			input: 0.15,
 			output: 0.15,
-			cacheRead: 0,
+			cacheRead: 0.015,
 			cacheWrite: 0,
 		},
 		contextWindow: 128000,
@@ -457,7 +457,7 @@ export const MISTRAL_MODELS = {
 		cost: {
 			input: 2,
 			output: 6,
-			cacheRead: 0,
+			cacheRead: 0.2,
 			cacheWrite: 0,
 		},
 		contextWindow: 64000,
@@ -474,7 +474,7 @@ export const MISTRAL_MODELS = {
 		cost: {
 			input: 0.7,
 			output: 0.7,
-			cacheRead: 0,
+			cacheRead: 0.07,
 			cacheWrite: 0,
 		},
 		contextWindow: 32000,
@@ -491,7 +491,7 @@ export const MISTRAL_MODELS = {
 		cost: {
 			input: 0.15,
 			output: 0.15,
-			cacheRead: 0,
+			cacheRead: 0.015,
 			cacheWrite: 0,
 		},
 		contextWindow: 128000,
@@ -508,7 +508,7 @@ export const MISTRAL_MODELS = {
 		cost: {
 			input: 2,
 			output: 6,
-			cacheRead: 0,
+			cacheRead: 0.2,
 			cacheWrite: 0,
 		},
 		contextWindow: 128000,

packages/ai/src/providers/moonshotai-cn.models.ts

@@ -130,4 +130,42 @@ export const MOONSHOTAI_CN_MODELS = {
 		contextWindow: 262144,
 		maxTokens: 262144,
 	} satisfies Model<"openai-completions">,
+	"kimi-k2.7-code": {
+		id: "kimi-k2.7-code",
+		name: "Kimi K2.7 Code",
+		api: "openai-completions",
+		provider: "moonshotai-cn",
+		baseUrl: "https://api.moonshot.cn/v1",
+		compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"thinkingFormat":"deepseek"},
+		reasoning: true,
+		thinkingLevelMap: {"off":null},
+		input: ["text", "image"],
+		cost: {
+			input: 0.95,
+			output: 4,
+			cacheRead: 0.19,
+			cacheWrite: 0,
+		},
+		contextWindow: 262144,
+		maxTokens: 262144,
+	} satisfies Model<"openai-completions">,
+	"kimi-k2.7-code-highspeed": {
+		id: "kimi-k2.7-code-highspeed",
+		name: "Kimi K2.7 Code HighSpeed",
+		api: "openai-completions",
+		provider: "moonshotai-cn",
+		baseUrl: "https://api.moonshot.cn/v1",
+		compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"thinkingFormat":"deepseek"},
+		reasoning: true,
+		thinkingLevelMap: {"off":null},
+		input: ["text", "image"],
+		cost: {
+			input: 1.9,
+			output: 8,
+			cacheRead: 0.38,
+			cacheWrite: 0,
+		},
+		contextWindow: 262144,
+		maxTokens: 262144,
+	} satisfies Model<"openai-completions">,
 } as const;

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

@@ -130,4 +130,42 @@ export const MOONSHOTAI_MODELS = {
 		contextWindow: 262144,
 		maxTokens: 262144,
 	} satisfies Model<"openai-completions">,
+	"kimi-k2.7-code": {
+		id: "kimi-k2.7-code",
+		name: "Kimi K2.7 Code",
+		api: "openai-completions",
+		provider: "moonshotai",
+		baseUrl: "https://api.moonshot.ai/v1",
+		compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"thinkingFormat":"deepseek"},
+		reasoning: true,
+		thinkingLevelMap: {"off":null},
+		input: ["text", "image"],
+		cost: {
+			input: 0.95,
+			output: 4,
+			cacheRead: 0.19,
+			cacheWrite: 0,
+		},
+		contextWindow: 262144,
+		maxTokens: 262144,
+	} satisfies Model<"openai-completions">,
+	"kimi-k2.7-code-highspeed": {
+		id: "kimi-k2.7-code-highspeed",
+		name: "Kimi K2.7 Code HighSpeed",
+		api: "openai-completions",
+		provider: "moonshotai",
+		baseUrl: "https://api.moonshot.ai/v1",
+		compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"thinkingFormat":"deepseek"},
+		reasoning: true,
+		thinkingLevelMap: {"off":null},
+		input: ["text", "image"],
+		cost: {
+			input: 1.9,
+			output: 8,
+			cacheRead: 0.38,
+			cacheWrite: 0,
+		},
+		contextWindow: 262144,
+		maxTokens: 262144,
+	} satisfies Model<"openai-completions">,
 } as const;

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

@@ -289,25 +289,6 @@ export const NVIDIA_MODELS = {
 		contextWindow: 131072,
 		maxTokens: 32768,
 	} satisfies Model<"openai-completions">,
-	"qwen/qwen3-coder-480b-a35b-instruct": {
-		id: "qwen/qwen3-coder-480b-a35b-instruct",
-		name: "Qwen3 Coder 480B A35B Instruct",
-		api: "openai-completions",
-		provider: "nvidia",
-		baseUrl: "https://integrate.api.nvidia.com/v1",
-		headers: {"NVCF-POLL-SECONDS":"3600"},
-		compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false},
-		reasoning: false,
-		input: ["text"],
-		cost: {
-			input: 0,
-			output: 0,
-			cacheRead: 0,
-			cacheWrite: 0,
-		},
-		contextWindow: 262144,
-		maxTokens: 66536,
-	} satisfies Model<"openai-completions">,
 	"qwen/qwen3.5-122b-a10b": {
 		id: "qwen/qwen3.5-122b-a10b",
 		name: "Qwen3.5 122B-A10B",

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

@@ -42,24 +42,6 @@ export const OPENCODE_GO_MODELS = {
 		contextWindow: 1000000,
 		maxTokens: 384000,
 	} satisfies Model<"openai-completions">,
-	"glm-5": {
-		id: "glm-5",
-		name: "GLM-5",
-		api: "openai-completions",
-		provider: "opencode-go",
-		baseUrl: "https://opencode.ai/zen/go/v1",
-		compat: {"maxTokensField":"max_tokens"},
-		reasoning: true,
-		input: ["text"],
-		cost: {
-			input: 1,
-			output: 3.2,
-			cacheRead: 0.2,
-			cacheWrite: 0,
-		},
-		contextWindow: 202752,
-		maxTokens: 32768,
-	} satisfies Model<"openai-completions">,
 	"glm-5.1": {
 		id: "glm-5.1",
 		name: "GLM-5.1",
@@ -78,23 +60,23 @@ export const OPENCODE_GO_MODELS = {
 		contextWindow: 202752,
 		maxTokens: 32768,
 	} satisfies Model<"openai-completions">,
-	"kimi-k2.5": {
-		id: "kimi-k2.5",
-		name: "Kimi K2.5",
+	"glm-5.2": {
+		id: "glm-5.2",
+		name: "GLM-5.2",
 		api: "openai-completions",
 		provider: "opencode-go",
 		baseUrl: "https://opencode.ai/zen/go/v1",
 		compat: {"maxTokensField":"max_tokens"},
 		reasoning: true,
-		input: ["text", "image"],
+		input: ["text"],
 		cost: {
-			input: 0.6,
-			output: 3,
-			cacheRead: 0.1,
+			input: 1.4,
+			output: 4.4,
+			cacheRead: 0.26,
 			cacheWrite: 0,
 		},
-		contextWindow: 262144,
-		maxTokens: 65536,
+		contextWindow: 1000000,
+		maxTokens: 131072,
 	} satisfies Model<"openai-completions">,
 	"kimi-k2.6": {
 		id: "kimi-k2.6",
@@ -102,7 +84,7 @@ export const OPENCODE_GO_MODELS = {
 		api: "openai-completions",
 		provider: "opencode-go",
 		baseUrl: "https://opencode.ai/zen/go/v1",
-		compat: {"thinkingFormat":"deepseek","supportsReasoningEffort":false,"maxTokensField":"max_tokens"},
+		compat: {"thinkingFormat":"deepseek","supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsLongCacheRetention":false},
 		reasoning: true,
 		thinkingLevelMap: {"minimal":null,"low":null,"medium":null},
 		input: ["text", "image"],
@@ -115,6 +97,24 @@ export const OPENCODE_GO_MODELS = {
 		contextWindow: 262144,
 		maxTokens: 65536,
 	} satisfies Model<"openai-completions">,
+	"kimi-k2.7-code": {
+		id: "kimi-k2.7-code",
+		name: "Kimi K2.7 Code",
+		api: "openai-completions",
+		provider: "opencode-go",
+		baseUrl: "https://opencode.ai/zen/go/v1",
+		compat: {"maxTokensField":"max_tokens"},
+		reasoning: true,
+		input: ["text", "image"],
+		cost: {
+			input: 0.95,
+			output: 4,
+			cacheRead: 0.19,
+			cacheWrite: 0,
+		},
+		contextWindow: 262144,
+		maxTokens: 262144,
+	} satisfies Model<"openai-completions">,
 	"mimo-v2.5": {
 		id: "mimo-v2.5",
 		name: "MiMo V2.5",
@@ -151,23 +151,6 @@ export const OPENCODE_GO_MODELS = {
 		contextWindow: 1048576,
 		maxTokens: 128000,
 	} satisfies Model<"openai-completions">,
-	"minimax-m2.5": {
-		id: "minimax-m2.5",
-		name: "MiniMax M2.5",
-		api: "anthropic-messages",
-		provider: "opencode-go",
-		baseUrl: "https://opencode.ai/zen/go",
-		reasoning: true,
-		input: ["text"],
-		cost: {
-			input: 0.3,
-			output: 1.2,
-			cacheRead: 0.03,
-			cacheWrite: 0,
-		},
-		contextWindow: 204800,
-		maxTokens: 65536,
-	} satisfies Model<"anthropic-messages">,
 	"minimax-m2.7": {
 		id: "minimax-m2.7",
 		name: "MiniMax M2.7",
@@ -188,16 +171,16 @@ export const OPENCODE_GO_MODELS = {
 	} satisfies Model<"openai-completions">,
 	"minimax-m3": {
 		id: "minimax-m3",
-		name: "MiniMax M3",
+		name: "MiniMax M3 (3x usage)",
 		api: "anthropic-messages",
 		provider: "opencode-go",
 		baseUrl: "https://opencode.ai/zen/go",
 		reasoning: true,
 		input: ["text", "image"],
 		cost: {
-			input: 0.3,
-			output: 1.2,
-			cacheRead: 0.06,
+			input: 0.1,
+			output: 0.4,
+			cacheRead: 0.02,
 			cacheWrite: 0,
 		},
 		contextWindow: 512000,

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

@@ -22,25 +22,6 @@ export const OPENCODE_MODELS = {
 		contextWindow: 200000,
 		maxTokens: 32000,
 	} satisfies Model<"openai-completions">,
-	"claude-fable-5": {
-		id: "claude-fable-5",
-		name: "Claude Fable 5",
-		api: "anthropic-messages",
-		provider: "opencode",
-		baseUrl: "https://opencode.ai/zen",
-		compat: {"forceAdaptiveThinking":true},
-		reasoning: true,
-		thinkingLevelMap: {"off":null,"xhigh":"xhigh"},
-		input: ["text", "image"],
-		cost: {
-			input: 10,
-			output: 50,
-			cacheRead: 1,
-			cacheWrite: 12.5,
-		},
-		contextWindow: 1000000,
-		maxTokens: 128000,
-	} satisfies Model<"anthropic-messages">,
 	"claude-haiku-4-5": {
 		id: "claude-haiku-4-5",
 		name: "Claude Haiku 4.5",
@@ -207,7 +188,7 @@ export const OPENCODE_MODELS = {
 		api: "openai-completions",
 		provider: "opencode",
 		baseUrl: "https://opencode.ai/zen/v1",
-		compat: {"maxTokensField":"max_tokens","requiresReasoningContentOnAssistantMessages":true,"thinkingFormat":"deepseek"},
+		compat: {"maxTokensField":"max_tokens","supportsLongCacheRetention":false,"requiresReasoningContentOnAssistantMessages":true},
 		reasoning: true,
 		thinkingLevelMap: {"minimal":null,"low":null,"medium":null,"high":"high","xhigh":"max"},
 		input: ["text"],
@@ -226,7 +207,7 @@ export const OPENCODE_MODELS = {
 		api: "openai-completions",
 		provider: "opencode",
 		baseUrl: "https://opencode.ai/zen/v1",
-		compat: {"maxTokensField":"max_tokens","requiresReasoningContentOnAssistantMessages":true,"thinkingFormat":"deepseek"},
+		compat: {"maxTokensField":"max_tokens","requiresReasoningContentOnAssistantMessages":true},
 		reasoning: true,
 		thinkingLevelMap: {"minimal":null,"low":null,"medium":null,"high":"high","xhigh":"max"},
 		input: ["text"],
@@ -245,7 +226,7 @@ export const OPENCODE_MODELS = {
 		api: "openai-completions",
 		provider: "opencode",
 		baseUrl: "https://opencode.ai/zen/v1",
-		compat: {"maxTokensField":"max_tokens","requiresReasoningContentOnAssistantMessages":true,"thinkingFormat":"deepseek"},
+		compat: {"maxTokensField":"max_tokens","supportsLongCacheRetention":false,"requiresReasoningContentOnAssistantMessages":true},
 		reasoning: true,
 		thinkingLevelMap: {"minimal":null,"low":null,"medium":null,"high":"high","xhigh":"max"},
 		input: ["text"],
@@ -661,7 +642,7 @@ export const OPENCODE_MODELS = {
 		api: "openai-completions",
 		provider: "opencode",
 		baseUrl: "https://opencode.ai/zen/v1",
-		compat: {"maxTokensField":"max_tokens"},
+		compat: {"maxTokensField":"max_tokens","supportsLongCacheRetention":false},
 		reasoning: true,
 		input: ["text", "image"],
 		cost: {
@@ -679,7 +660,7 @@ export const OPENCODE_MODELS = {
 		api: "openai-completions",
 		provider: "opencode",
 		baseUrl: "https://opencode.ai/zen/v1",
-		compat: {"thinkingFormat":"deepseek","supportsReasoningEffort":false,"maxTokensField":"max_tokens"},
+		compat: {"thinkingFormat":"deepseek","supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsLongCacheRetention":false},
 		reasoning: true,
 		input: ["text", "image"],
 		cost: {
@@ -733,7 +714,7 @@ export const OPENCODE_MODELS = {
 		api: "openai-completions",
 		provider: "opencode",
 		baseUrl: "https://opencode.ai/zen/v1",
-		compat: {"maxTokensField":"max_tokens"},
+		compat: {"maxTokensField":"max_tokens","supportsLongCacheRetention":false},
 		reasoning: true,
 		input: ["text"],
 		cost: {

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

@@ -4,25 +4,6 @@
 import type { Model } from "../types.ts";
 
 export const TOGETHER_MODELS = {
-	"MiniMaxAI/MiniMax-M2.5": {
-		id: "MiniMaxAI/MiniMax-M2.5",
-		name: "MiniMax-M2.5",
-		api: "openai-completions",
-		provider: "together",
-		baseUrl: "https://api.together.ai/v1",
-		compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false,"thinkingFormat":"together"},
-		reasoning: true,
-		thinkingLevelMap: {"minimal":null,"low":null,"medium":null},
-		input: ["text"],
-		cost: {
-			input: 0.3,
-			output: 1.2,
-			cacheRead: 0.06,
-			cacheWrite: 0,
-		},
-		contextWindow: 204800,
-		maxTokens: 131072,
-	} satisfies Model<"openai-completions">,
 	"MiniMaxAI/MiniMax-M2.7": {
 		id: "MiniMaxAI/MiniMax-M2.7",
 		name: "MiniMax-M2.7",
@@ -42,28 +23,28 @@ export const TOGETHER_MODELS = {
 		contextWindow: 202752,
 		maxTokens: 131072,
 	} satisfies Model<"openai-completions">,
-	"Qwen/Qwen3-235B-A22B-Instruct-2507-tput": {
-		id: "Qwen/Qwen3-235B-A22B-Instruct-2507-tput",
-		name: "Qwen3 235B A22B Instruct 2507 FP8",
+	"MiniMaxAI/MiniMax-M3": {
+		id: "MiniMaxAI/MiniMax-M3",
+		name: "MiniMax-M3",
 		api: "openai-completions",
 		provider: "together",
 		baseUrl: "https://api.together.ai/v1",
 		compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false,"thinkingFormat":"together"},
 		reasoning: true,
 		thinkingLevelMap: {"minimal":null,"low":null,"medium":null},
-		input: ["text"],
+		input: ["text", "image"],
 		cost: {
-			input: 0.2,
-			output: 0.6,
-			cacheRead: 0,
+			input: 0.3,
+			output: 1.2,
+			cacheRead: 0.06,
 			cacheWrite: 0,
 		},
-		contextWindow: 262144,
-		maxTokens: 262144,
+		contextWindow: 524288,
+		maxTokens: 250000,
 	} satisfies Model<"openai-completions">,
-	"Qwen/Qwen3-Coder-480B-A35B-Instruct-FP8": {
-		id: "Qwen/Qwen3-Coder-480B-A35B-Instruct-FP8",
-		name: "Qwen3 Coder 480B A35B Instruct",
+	"Qwen/Qwen2.5-7B-Instruct-Turbo": {
+		id: "Qwen/Qwen2.5-7B-Instruct-Turbo",
+		name: "Qwen 2.5 7B Instruct Turbo",
 		api: "openai-completions",
 		provider: "together",
 		baseUrl: "https://api.together.ai/v1",
@@ -71,27 +52,26 @@ export const TOGETHER_MODELS = {
 		reasoning: false,
 		input: ["text"],
 		cost: {
-			input: 2,
-			output: 2,
+			input: 0.3,
+			output: 0.3,
 			cacheRead: 0,
 			cacheWrite: 0,
 		},
-		contextWindow: 262144,
-		maxTokens: 262144,
+		contextWindow: 32768,
+		maxTokens: 32768,
 	} satisfies Model<"openai-completions">,
-	"Qwen/Qwen3-Coder-Next-FP8": {
-		id: "Qwen/Qwen3-Coder-Next-FP8",
-		name: "Qwen3 Coder Next FP8",
+	"Qwen/Qwen3-235B-A22B-Instruct-2507-tput": {
+		id: "Qwen/Qwen3-235B-A22B-Instruct-2507-tput",
+		name: "Qwen3 235B A22B Instruct 2507 FP8",
 		api: "openai-completions",
 		provider: "together",
 		baseUrl: "https://api.together.ai/v1",
-		compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false,"thinkingFormat":"together"},
-		reasoning: true,
-		thinkingLevelMap: {"minimal":null,"low":null,"medium":null},
+		compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false},
+		reasoning: false,
 		input: ["text"],
 		cost: {
-			input: 0.5,
-			output: 1.2,
+			input: 0.2,
+			output: 0.6,
 			cacheRead: 0,
 			cacheWrite: 0,
 		},
@@ -117,6 +97,25 @@ export const TOGETHER_MODELS = {
 		contextWindow: 262144,
 		maxTokens: 130000,
 	} satisfies Model<"openai-completions">,
+	"Qwen/Qwen3.5-9B": {
+		id: "Qwen/Qwen3.5-9B",
+		name: "Qwen3.5 9B",
+		api: "openai-completions",
+		provider: "together",
+		baseUrl: "https://api.together.ai/v1",
+		compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false,"thinkingFormat":"together"},
+		reasoning: true,
+		thinkingLevelMap: {"minimal":null,"low":null,"medium":null},
+		input: ["text", "image"],
+		cost: {
+			input: 0.17,
+			output: 0.25,
+			cacheRead: 0,
+			cacheWrite: 0,
+		},
+		contextWindow: 262144,
+		maxTokens: 65536,
+	} satisfies Model<"openai-completions">,
 	"Qwen/Qwen3.6-Plus": {
 		id: "Qwen/Qwen3.6-Plus",
 		name: "Qwen3.6 Plus",
@@ -142,57 +141,18 @@ export const TOGETHER_MODELS = {
 		api: "openai-completions",
 		provider: "together",
 		baseUrl: "https://api.together.ai/v1",
-		compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false,"thinkingFormat":"together"},
-		reasoning: true,
-		thinkingLevelMap: {"minimal":null,"low":null,"medium":null},
+		compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false},
+		reasoning: false,
 		input: ["text"],
 		cost: {
-			input: 2.5,
-			output: 7.5,
+			input: 1.25,
+			output: 3.75,
 			cacheRead: 0,
 			cacheWrite: 0,
 		},
 		contextWindow: 1000000,
 		maxTokens: 500000,
 	} satisfies Model<"openai-completions">,
-	"deepseek-ai/DeepSeek-V3": {
-		id: "deepseek-ai/DeepSeek-V3",
-		name: "DeepSeek-V3",
-		api: "openai-completions",
-		provider: "together",
-		baseUrl: "https://api.together.ai/v1",
-		compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false,"thinkingFormat":"together"},
-		reasoning: true,
-		thinkingLevelMap: {"minimal":null,"low":null,"medium":null},
-		input: ["text"],
-		cost: {
-			input: 1.25,
-			output: 1.25,
-			cacheRead: 0,
-			cacheWrite: 0,
-		},
-		contextWindow: 131072,
-		maxTokens: 131072,
-	} satisfies Model<"openai-completions">,
-	"deepseek-ai/DeepSeek-V3-1": {
-		id: "deepseek-ai/DeepSeek-V3-1",
-		name: "DeepSeek V3.1",
-		api: "openai-completions",
-		provider: "together",
-		baseUrl: "https://api.together.ai/v1",
-		compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false,"thinkingFormat":"together"},
-		reasoning: true,
-		thinkingLevelMap: {"minimal":null,"low":null,"medium":null},
-		input: ["text"],
-		cost: {
-			input: 0.6,
-			output: 1.7,
-			cacheRead: 0,
-			cacheWrite: 0,
-		},
-		contextWindow: 131072,
-		maxTokens: 131072,
-	} satisfies Model<"openai-completions">,
 	"deepseek-ai/DeepSeek-V4-Pro": {
 		id: "deepseek-ai/DeepSeek-V4-Pro",
 		name: "DeepSeek V4 Pro",
@@ -204,8 +164,8 @@ export const TOGETHER_MODELS = {
 		thinkingLevelMap: {"minimal":null,"low":null,"medium":null,"high":"high","xhigh":null},
 		input: ["text"],
 		cost: {
-			input: 2.1,
-			output: 4.4,
+			input: 1.74,
+			output: 3.48,
 			cacheRead: 0.2,
 			cacheWrite: 0,
 		},
@@ -241,8 +201,8 @@ export const TOGETHER_MODELS = {
 		thinkingLevelMap: {"minimal":null,"low":null,"medium":null},
 		input: ["text", "image"],
 		cost: {
-			input: 0.2,
-			output: 0.5,
+			input: 0.39,
+			output: 0.97,
 			cacheRead: 0,
 			cacheWrite: 0,
 		},
@@ -267,25 +227,6 @@ export const TOGETHER_MODELS = {
 		contextWindow: 131072,
 		maxTokens: 131072,
 	} satisfies Model<"openai-completions">,
-	"moonshotai/Kimi-K2.5": {
-		id: "moonshotai/Kimi-K2.5",
-		name: "Kimi K2.5",
-		api: "openai-completions",
-		provider: "together",
-		baseUrl: "https://api.together.ai/v1",
-		compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false,"thinkingFormat":"together"},
-		reasoning: true,
-		thinkingLevelMap: {"minimal":null,"low":null,"medium":null},
-		input: ["text", "image"],
-		cost: {
-			input: 0.5,
-			output: 2.8,
-			cacheRead: 0,
-			cacheWrite: 0,
-		},
-		contextWindow: 262144,
-		maxTokens: 262144,
-	} satisfies Model<"openai-completions">,
 	"moonshotai/Kimi-K2.6": {
 		id: "moonshotai/Kimi-K2.6",
 		name: "Kimi K2.6",
@@ -305,6 +246,25 @@ export const TOGETHER_MODELS = {
 		contextWindow: 262144,
 		maxTokens: 131000,
 	} satisfies Model<"openai-completions">,
+	"moonshotai/Kimi-K2.7-Code": {
+		id: "moonshotai/Kimi-K2.7-Code",
+		name: "Kimi K2.7 Code",
+		api: "openai-completions",
+		provider: "together",
+		baseUrl: "https://api.together.ai/v1",
+		compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false,"thinkingFormat":"together"},
+		reasoning: true,
+		thinkingLevelMap: {"minimal":null,"low":null,"medium":null},
+		input: ["text"],
+		cost: {
+			input: 0.95,
+			output: 4,
+			cacheRead: 0.19,
+			cacheWrite: 0,
+		},
+		contextWindow: 262144,
+		maxTokens: 131072,
+	} satisfies Model<"openai-completions">,
 	"nvidia/nemotron-3-ultra-550b-a55b": {
 		id: "nvidia/nemotron-3-ultra-550b-a55b",
 		name: "Nemotron 3 Ultra 550B A55B",
@@ -343,6 +303,44 @@ export const TOGETHER_MODELS = {
 		contextWindow: 131072,
 		maxTokens: 131072,
 	} satisfies Model<"openai-completions">,
+	"openai/gpt-oss-20b": {
+		id: "openai/gpt-oss-20b",
+		name: "GPT OSS 20B",
+		api: "openai-completions",
+		provider: "together",
+		baseUrl: "https://api.together.ai/v1",
+		compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":true,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false,"thinkingFormat":"openai"},
+		reasoning: true,
+		thinkingLevelMap: {"off":null,"minimal":null},
+		input: ["text"],
+		cost: {
+			input: 0.05,
+			output: 0.2,
+			cacheRead: 0,
+			cacheWrite: 0,
+		},
+		contextWindow: 131072,
+		maxTokens: 131072,
+	} satisfies Model<"openai-completions">,
+	"zai-org/GLM-5": {
+		id: "zai-org/GLM-5",
+		name: "GLM-5",
+		api: "openai-completions",
+		provider: "together",
+		baseUrl: "https://api.together.ai/v1",
+		compat: {"supportsStore":false,"supportsDeveloperRole":false,"supportsReasoningEffort":false,"maxTokensField":"max_tokens","supportsStrictMode":false,"supportsLongCacheRetention":false,"thinkingFormat":"together"},
+		reasoning: true,
+		thinkingLevelMap: {"minimal":null,"low":null,"medium":null},
+		input: ["text"],
+		cost: {
+			input: 1,
+			output: 3.2,
+			cacheRead: 0,
+			cacheWrite: 0,
+		},
+		contextWindow: 202752,
+		maxTokens: 131072,
+	} satisfies Model<"openai-completions">,
 	"zai-org/GLM-5.1": {
 		id: "zai-org/GLM-5.1",
 		name: "GLM-5.1",

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

@@ -98,7 +98,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		reasoning: true,
 		input: ["text", "image"],
 		cost: {
-			input: 0.39999999999999997,
+			input: 0.4,
 			output: 4,
 			cacheRead: 0,
 			cacheWrite: 0,
@@ -168,7 +168,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		cost: {
 			input: 1,
 			output: 5,
-			cacheRead: 0.19999999999999998,
+			cacheRead: 0.2,
 			cacheWrite: 0,
 		},
 		contextWindow: 1000000,
@@ -268,7 +268,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		reasoning: true,
 		input: ["text", "image"],
 		cost: {
-			input: 0.39999999999999997,
+			input: 0.4,
 			output: 4,
 			cacheRead: 0,
 			cacheWrite: 0,
@@ -285,8 +285,8 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		reasoning: true,
 		input: ["text", "image"],
 		cost: {
-			input: 0.09999999999999999,
-			output: 0.39999999999999997,
+			input: 0.1,
+			output: 0.4,
 			cacheRead: 0.001,
 			cacheWrite: 0.125,
 		},
@@ -302,7 +302,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		reasoning: true,
 		input: ["text", "image"],
 		cost: {
-			input: 0.39999999999999997,
+			input: 0.4,
 			output: 2.4,
 			cacheRead: 0.04,
 			cacheWrite: 0.5,
@@ -320,7 +320,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		input: ["text", "image"],
 		cost: {
 			input: 0.6,
-			output: 3.5999999999999996,
+			output: 3.6,
 			cacheRead: 0,
 			cacheWrite: 0,
 		},
@@ -338,7 +338,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		cost: {
 			input: 0.5,
 			output: 3,
-			cacheRead: 0.09999999999999999,
+			cacheRead: 0.1,
 			cacheWrite: 0.625,
 		},
 		contextWindow: 1000000,
@@ -370,8 +370,8 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		reasoning: true,
 		input: ["text", "image"],
 		cost: {
-			input: 0.39999999999999997,
-			output: 1.5999999999999999,
+			input: 0.4,
+			output: 1.6,
 			cacheRead: 0.08,
 			cacheWrite: 0.5,
 		},
@@ -404,7 +404,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		reasoning: false,
 		input: ["text", "image"],
 		cost: {
-			input: 0.7999999999999999,
+			input: 0.8,
 			output: 4,
 			cacheRead: 0.08,
 			cacheWrite: 1,
@@ -412,25 +412,6 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		contextWindow: 200000,
 		maxTokens: 8192,
 	} satisfies Model<"anthropic-messages">,
-	"anthropic/claude-fable-5": {
-		id: "anthropic/claude-fable-5",
-		name: "Claude Fable 5",
-		api: "anthropic-messages",
-		provider: "vercel-ai-gateway",
-		baseUrl: "https://ai-gateway.vercel.sh",
-		compat: {"forceAdaptiveThinking":true},
-		reasoning: true,
-		thinkingLevelMap: {"off":null,"xhigh":"xhigh"},
-		input: ["text", "image"],
-		cost: {
-			input: 10,
-			output: 50,
-			cacheRead: 1,
-			cacheWrite: 12.5,
-		},
-		contextWindow: 1000000,
-		maxTokens: 128000,
-	} satisfies Model<"anthropic-messages">,
 	"anthropic/claude-haiku-4.5": {
 		id: "anthropic/claude-haiku-4.5",
 		name: "Claude Haiku 4.5",
@@ -442,7 +423,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		cost: {
 			input: 1,
 			output: 5,
-			cacheRead: 0.09999999999999999,
+			cacheRead: 0.1,
 			cacheWrite: 1.25,
 		},
 		contextWindow: 200000,
@@ -635,7 +616,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		input: ["text"],
 		cost: {
 			input: 0.25,
-			output: 0.8999999999999999,
+			output: 0.9,
 			cacheRead: 0,
 			cacheWrite: 0,
 		},
@@ -653,7 +634,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		cost: {
 			input: 0.25,
 			output: 2,
-			cacheRead: 0.049999999999999996,
+			cacheRead: 0.05,
 			cacheWrite: 0,
 		},
 		contextWindow: 256000,
@@ -838,8 +819,8 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		reasoning: true,
 		input: ["text", "image"],
 		cost: {
-			input: 0.09999999999999999,
-			output: 0.39999999999999997,
+			input: 0.1,
+			output: 0.4,
 			cacheRead: 0.01,
 			cacheWrite: 0,
 		},
@@ -874,7 +855,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		cost: {
 			input: 0.5,
 			output: 3,
-			cacheRead: 0.049999999999999996,
+			cacheRead: 0.05,
 			cacheWrite: 0,
 		},
 		contextWindow: 1000000,
@@ -891,7 +872,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		cost: {
 			input: 2,
 			output: 12,
-			cacheRead: 0.19999999999999998,
+			cacheRead: 0.2,
 			cacheWrite: 0,
 		},
 		contextWindow: 1000000,
@@ -942,7 +923,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		cost: {
 			input: 2,
 			output: 12,
-			cacheRead: 0.19999999999999998,
+			cacheRead: 0.2,
 			cacheWrite: 0,
 		},
 		contextWindow: 1000000,
@@ -992,7 +973,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		input: ["text", "image"],
 		cost: {
 			input: 0.14,
-			output: 0.39999999999999997,
+			output: 0.4,
 			cacheRead: 0,
 			cacheWrite: 0,
 		},
@@ -1010,7 +991,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		cost: {
 			input: 0.25,
 			output: 0.75,
-			cacheRead: 0.024999999999999998,
+			cacheRead: 0.025,
 			cacheWrite: 0,
 		},
 		contextWindow: 128000,
@@ -1162,7 +1143,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		input: ["text", "image"],
 		cost: {
 			input: 0.24,
-			output: 0.9700000000000001,
+			output: 0.97,
 			cacheRead: 0,
 			cacheWrite: 0,
 		},
@@ -1178,7 +1159,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		reasoning: false,
 		input: ["text", "image"],
 		cost: {
-			input: 0.16999999999999998,
+			input: 0.17,
 			output: 0.66,
 			cacheRead: 0,
 			cacheWrite: 0,
@@ -1332,7 +1313,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		input: ["text"],
 		cost: {
 			input: 0.3,
-			output: 0.8999999999999999,
+			output: 0.9,
 			cacheRead: 0,
 			cacheWrite: 0,
 		},
@@ -1348,7 +1329,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		reasoning: false,
 		input: ["text"],
 		cost: {
-			input: 0.39999999999999997,
+			input: 0.4,
 			output: 2,
 			cacheRead: 0,
 			cacheWrite: 0,
@@ -1365,7 +1346,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		reasoning: false,
 		input: ["text"],
 		cost: {
-			input: 0.09999999999999999,
+			input: 0.1,
 			output: 0.3,
 			cacheRead: 0,
 			cacheWrite: 0,
@@ -1382,7 +1363,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		reasoning: false,
 		input: ["text"],
 		cost: {
-			input: 0.09999999999999999,
+			input: 0.1,
 			output: 0.3,
 			cacheRead: 0,
 			cacheWrite: 0,
@@ -1399,8 +1380,8 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		reasoning: false,
 		input: ["text"],
 		cost: {
-			input: 0.09999999999999999,
-			output: 0.09999999999999999,
+			input: 0.1,
+			output: 0.1,
 			cacheRead: 0,
 			cacheWrite: 0,
 		},
@@ -1433,7 +1414,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		reasoning: false,
 		input: ["text", "image"],
 		cost: {
-			input: 0.39999999999999997,
+			input: 0.4,
 			output: 2,
 			cacheRead: 0,
 			cacheWrite: 0,
@@ -1467,13 +1448,13 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		reasoning: false,
 		input: ["text"],
 		cost: {
-			input: 0.02,
-			output: 0.04,
+			input: 0.15,
+			output: 0.15,
 			cacheRead: 0,
 			cacheWrite: 0,
 		},
-		contextWindow: 131072,
-		maxTokens: 131072,
+		contextWindow: 128000,
+		maxTokens: 128000,
 	} satisfies Model<"anthropic-messages">,
 	"mistral/mistral-small": {
 		id: "mistral/mistral-small",
@@ -1484,7 +1465,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		reasoning: false,
 		input: ["text", "image"],
 		cost: {
-			input: 0.09999999999999999,
+			input: 0.1,
 			output: 0.3,
 			cacheRead: 0,
 			cacheWrite: 0,
@@ -1535,7 +1516,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		reasoning: false,
 		input: ["text"],
 		cost: {
-			input: 0.5700000000000001,
+			input: 0.57,
 			output: 2.3,
 			cacheRead: 0,
 			cacheWrite: 0,
@@ -1560,40 +1541,6 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		contextWindow: 262114,
 		maxTokens: 262114,
 	} satisfies Model<"anthropic-messages">,
-	"moonshotai/kimi-k2-thinking-turbo": {
-		id: "moonshotai/kimi-k2-thinking-turbo",
-		name: "Kimi K2 Thinking Turbo",
-		api: "anthropic-messages",
-		provider: "vercel-ai-gateway",
-		baseUrl: "https://ai-gateway.vercel.sh",
-		reasoning: true,
-		input: ["text"],
-		cost: {
-			input: 1.15,
-			output: 8,
-			cacheRead: 0.15,
-			cacheWrite: 0,
-		},
-		contextWindow: 262114,
-		maxTokens: 262114,
-	} satisfies Model<"anthropic-messages">,
-	"moonshotai/kimi-k2-turbo": {
-		id: "moonshotai/kimi-k2-turbo",
-		name: "Kimi K2 Turbo",
-		api: "anthropic-messages",
-		provider: "vercel-ai-gateway",
-		baseUrl: "https://ai-gateway.vercel.sh",
-		reasoning: false,
-		input: ["text"],
-		cost: {
-			input: 1.15,
-			output: 8,
-			cacheRead: 0.15,
-			cacheWrite: 0,
-		},
-		contextWindow: 256000,
-		maxTokens: 16384,
-	} satisfies Model<"anthropic-messages">,
 	"moonshotai/kimi-k2.5": {
 		id: "moonshotai/kimi-k2.5",
 		name: "Kimi K2.5",
@@ -1605,7 +1552,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		cost: {
 			input: 0.6,
 			output: 3,
-			cacheRead: 0.09999999999999999,
+			cacheRead: 0.1,
 			cacheWrite: 0,
 		},
 		contextWindow: 262114,
@@ -1628,6 +1575,40 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		contextWindow: 262000,
 		maxTokens: 262000,
 	} satisfies Model<"anthropic-messages">,
+	"moonshotai/kimi-k2.7-code": {
+		id: "moonshotai/kimi-k2.7-code",
+		name: "Kimi K2.7 Code",
+		api: "anthropic-messages",
+		provider: "vercel-ai-gateway",
+		baseUrl: "https://ai-gateway.vercel.sh",
+		reasoning: true,
+		input: ["text", "image"],
+		cost: {
+			input: 0.95,
+			output: 4,
+			cacheRead: 0.19,
+			cacheWrite: 0,
+		},
+		contextWindow: 256000,
+		maxTokens: 32768,
+	} satisfies Model<"anthropic-messages">,
+	"moonshotai/kimi-k2.7-code-highspeed": {
+		id: "moonshotai/kimi-k2.7-code-highspeed",
+		name: "Kimi K2.7 Code High Speed",
+		api: "anthropic-messages",
+		provider: "vercel-ai-gateway",
+		baseUrl: "https://ai-gateway.vercel.sh",
+		reasoning: true,
+		input: ["text", "image"],
+		cost: {
+			input: 1.9,
+			output: 8,
+			cacheRead: 0.38,
+			cacheWrite: 0,
+		},
+		contextWindow: 262144,
+		maxTokens: 32768,
+	} satisfies Model<"anthropic-messages">,
 	"nvidia/nemotron-3-super-120b-a12b": {
 		id: "nvidia/nemotron-3-super-120b-a12b",
 		name: "NVIDIA Nemotron 3 Super 120B A12B",
@@ -1671,7 +1652,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		reasoning: true,
 		input: ["text", "image"],
 		cost: {
-			input: 0.19999999999999998,
+			input: 0.2,
 			output: 0.6,
 			cacheRead: 0,
 			cacheWrite: 0,
@@ -1689,7 +1670,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		input: ["text"],
 		cost: {
 			input: 0.06,
-			output: 0.22999999999999998,
+			output: 0.23,
 			cacheRead: 0,
 			cacheWrite: 0,
 		},
@@ -1739,9 +1720,9 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		reasoning: false,
 		input: ["text", "image"],
 		cost: {
-			input: 0.39999999999999997,
-			output: 1.5999999999999999,
-			cacheRead: 0.09999999999999999,
+			input: 0.4,
+			output: 1.6,
+			cacheRead: 0.1,
 			cacheWrite: 0,
 		},
 		contextWindow: 1047576,
@@ -1756,9 +1737,9 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		reasoning: false,
 		input: ["text", "image"],
 		cost: {
-			input: 0.09999999999999999,
-			output: 0.39999999999999997,
-			cacheRead: 0.024999999999999998,
+			input: 0.1,
+			output: 0.4,
+			cacheRead: 0.025,
 			cacheWrite: 0,
 		},
 		contextWindow: 1047576,
@@ -1860,7 +1841,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		cost: {
 			input: 0.25,
 			output: 2,
-			cacheRead: 0.024999999999999998,
+			cacheRead: 0.025,
 			cacheWrite: 0,
 		},
 		contextWindow: 400000,
@@ -1875,8 +1856,8 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		reasoning: true,
 		input: ["text", "image"],
 		cost: {
-			input: 0.049999999999999996,
-			output: 0.39999999999999997,
+			input: 0.05,
+			output: 0.4,
 			cacheRead: 0.005,
 			cacheWrite: 0,
 		},
@@ -1945,7 +1926,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		cost: {
 			input: 0.25,
 			output: 2,
-			cacheRead: 0.024999999999999998,
+			cacheRead: 0.025,
 			cacheWrite: 0,
 		},
 		contextWindow: 400000,
@@ -2139,7 +2120,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		thinkingLevelMap: {"xhigh":"xhigh"},
 		input: ["text", "image"],
 		cost: {
-			input: 0.19999999999999998,
+			input: 0.2,
 			output: 1.25,
 			cacheRead: 0.02,
 			cacheWrite: 0,
@@ -2227,8 +2208,8 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		reasoning: true,
 		input: ["text"],
 		cost: {
-			input: 0.049999999999999996,
-			output: 0.19999999999999998,
+			input: 0.05,
+			output: 0.2,
 			cacheRead: 0,
 			cacheWrite: 0,
 		},
@@ -2388,6 +2369,23 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		contextWindow: 200000,
 		maxTokens: 8000,
 	} satisfies Model<"anthropic-messages">,
+	"sakana/fugu-ultra": {
+		id: "sakana/fugu-ultra",
+		name: "Fugu Ultra",
+		api: "anthropic-messages",
+		provider: "vercel-ai-gateway",
+		baseUrl: "https://ai-gateway.vercel.sh",
+		reasoning: true,
+		input: ["text", "image"],
+		cost: {
+			input: 5,
+			output: 30,
+			cacheRead: 0.5,
+			cacheWrite: 0,
+		},
+		contextWindow: 1000000,
+		maxTokens: 1000000,
+	} satisfies Model<"anthropic-messages">,
 	"stepfun/step-3.5-flash": {
 		id: "stepfun/step-3.5-flash",
 		name: "StepFun 3.5 Flash",
@@ -2399,8 +2397,8 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		cost: {
 			input: 0.09,
 			output: 0.3,
-			cacheRead: 0,
-			cacheWrite: 0.02,
+			cacheRead: 0.02,
+			cacheWrite: 0,
 		},
 		contextWindow: 262114,
 		maxTokens: 262114,
@@ -2414,7 +2412,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		reasoning: true,
 		input: ["text", "image"],
 		cost: {
-			input: 0.19999999999999998,
+			input: 0.2,
 			output: 1.15,
 			cacheRead: 0.04,
 			cacheWrite: 0,
@@ -2431,9 +2429,9 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		reasoning: false,
 		input: ["text", "image"],
 		cost: {
-			input: 0.19999999999999998,
+			input: 0.2,
 			output: 0.5,
-			cacheRead: 0.049999999999999996,
+			cacheRead: 0.05,
 			cacheWrite: 0,
 		},
 		contextWindow: 1000000,
@@ -2448,9 +2446,9 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		reasoning: true,
 		input: ["text", "image"],
 		cost: {
-			input: 0.19999999999999998,
+			input: 0.2,
 			output: 0.5,
-			cacheRead: 0.049999999999999996,
+			cacheRead: 0.05,
 			cacheWrite: 0,
 		},
 		contextWindow: 1000000,
@@ -2467,7 +2465,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		cost: {
 			input: 1.25,
 			output: 2.5,
-			cacheRead: 0.19999999999999998,
+			cacheRead: 0.2,
 			cacheWrite: 0,
 		},
 		contextWindow: 2000000,
@@ -2484,7 +2482,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		cost: {
 			input: 1.25,
 			output: 2.5,
-			cacheRead: 0.19999999999999998,
+			cacheRead: 0.2,
 			cacheWrite: 0,
 		},
 		contextWindow: 2000000,
@@ -2501,7 +2499,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		cost: {
 			input: 1.25,
 			output: 2.5,
-			cacheRead: 0.19999999999999998,
+			cacheRead: 0.2,
 			cacheWrite: 0,
 		},
 		contextWindow: 2000000,
@@ -2518,7 +2516,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		cost: {
 			input: 1.25,
 			output: 2.5,
-			cacheRead: 0.19999999999999998,
+			cacheRead: 0.2,
 			cacheWrite: 0,
 		},
 		contextWindow: 2000000,
@@ -2535,7 +2533,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		cost: {
 			input: 1.25,
 			output: 2.5,
-			cacheRead: 0.19999999999999998,
+			cacheRead: 0.2,
 			cacheWrite: 0,
 		},
 		contextWindow: 2000000,
@@ -2552,7 +2550,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		cost: {
 			input: 1.25,
 			output: 2.5,
-			cacheRead: 0.19999999999999998,
+			cacheRead: 0.2,
 			cacheWrite: 0,
 		},
 		contextWindow: 2000000,
@@ -2569,7 +2567,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		cost: {
 			input: 1.25,
 			output: 2.5,
-			cacheRead: 0.19999999999999998,
+			cacheRead: 0.2,
 			cacheWrite: 0,
 		},
 		contextWindow: 1000000,
@@ -2586,7 +2584,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		cost: {
 			input: 1,
 			output: 2,
-			cacheRead: 0.19999999999999998,
+			cacheRead: 0.2,
 			cacheWrite: 0,
 		},
 		contextWindow: 256000,
@@ -2601,7 +2599,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		reasoning: true,
 		input: ["text"],
 		cost: {
-			input: 0.09999999999999999,
+			input: 0.1,
 			output: 0.3,
 			cacheRead: 0.01,
 			cacheWrite: 0,
@@ -2620,7 +2618,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		cost: {
 			input: 1,
 			output: 3,
-			cacheRead: 0.19999999999999998,
+			cacheRead: 0.2,
 			cacheWrite: 0,
 		},
 		contextWindow: 1000000,
@@ -2686,7 +2684,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		reasoning: true,
 		input: ["text"],
 		cost: {
-			input: 0.19999999999999998,
+			input: 0.2,
 			output: 1.1,
 			cacheRead: 0.03,
 			cacheWrite: 0,
@@ -2704,7 +2702,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		input: ["text", "image"],
 		cost: {
 			input: 0.6,
-			output: 1.7999999999999998,
+			output: 1.8,
 			cacheRead: 0.11,
 			cacheWrite: 0,
 		},
@@ -2738,8 +2736,8 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		input: ["text", "image"],
 		cost: {
 			input: 0.3,
-			output: 0.8999999999999999,
-			cacheRead: 0.049999999999999996,
+			output: 0.9,
+			cacheRead: 0.05,
 			cacheWrite: 0,
 		},
 		contextWindow: 128000,
@@ -2789,7 +2787,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		input: ["text"],
 		cost: {
 			input: 0.07,
-			output: 0.39999999999999997,
+			output: 0.4,
 			cacheRead: 0,
 			cacheWrite: 0,
 		},
@@ -2806,7 +2804,7 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		input: ["text"],
 		cost: {
 			input: 0.06,
-			output: 0.39999999999999997,
+			output: 0.4,
 			cacheRead: 0.01,
 			cacheWrite: 0,
 		},
@@ -2823,8 +2821,8 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		input: ["text"],
 		cost: {
 			input: 1,
-			output: 3.1999999999999997,
-			cacheRead: 0.19999999999999998,
+			output: 3.2,
+			cacheRead: 0.2,
 			cacheWrite: 0,
 		},
 		contextWindow: 202800,
@@ -2864,6 +2862,23 @@ export const VERCEL_AI_GATEWAY_MODELS = {
 		contextWindow: 202800,
 		maxTokens: 64000,
 	} satisfies Model<"anthropic-messages">,
+	"zai/glm-5.2": {
+		id: "zai/glm-5.2",
+		name: "GLM 5.2",
+		api: "anthropic-messages",
+		provider: "vercel-ai-gateway",
+		baseUrl: "https://ai-gateway.vercel.sh",
+		reasoning: true,
+		input: ["text"],
+		cost: {
+			input: 1.5,
+			output: 4.5,
+			cacheRead: 0.3,
+			cacheWrite: 0,
+		},
+		contextWindow: 1000000,
+		maxTokens: 128000,
+	} satisfies Model<"anthropic-messages">,
 	"zai/glm-5v-turbo": {
 		id: "zai/glm-5v-turbo",
 		name: "GLM 5V Turbo",

packages/ai/src/providers/zai-coding-cn.models.ts

@@ -76,6 +76,25 @@ export const ZAI_CODING_CN_MODELS = {
 		contextWindow: 200000,
 		maxTokens: 131072,
 	} satisfies Model<"openai-completions">,
+	"glm-5.2": {
+		id: "glm-5.2",
+		name: "GLM-5.2",
+		api: "openai-completions",
+		provider: "zai-coding-cn",
+		baseUrl: "https://open.bigmodel.cn/api/coding/paas/v4",
+		compat: {"supportsDeveloperRole":false,"thinkingFormat":"zai","supportsReasoningEffort":true,"zaiToolStream":true},
+		reasoning: true,
+		thinkingLevelMap: {"minimal":null,"low":"high","medium":"high","high":"high","xhigh":"max"},
+		input: ["text"],
+		cost: {
+			input: 0,
+			output: 0,
+			cacheRead: 0,
+			cacheWrite: 0,
+		},
+		contextWindow: 1000000,
+		maxTokens: 131072,
+	} satisfies Model<"openai-completions">,
 	"glm-5v-turbo": {
 		id: "glm-5v-turbo",
 		name: "GLM-5V-Turbo",

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

@@ -76,6 +76,25 @@ export const ZAI_MODELS = {
 		contextWindow: 200000,
 		maxTokens: 131072,
 	} satisfies Model<"openai-completions">,
+	"glm-5.2": {
+		id: "glm-5.2",
+		name: "GLM-5.2",
+		api: "openai-completions",
+		provider: "zai",
+		baseUrl: "https://api.z.ai/api/coding/paas/v4",
+		compat: {"supportsDeveloperRole":false,"thinkingFormat":"zai","supportsReasoningEffort":true,"zaiToolStream":true},
+		reasoning: true,
+		thinkingLevelMap: {"minimal":null,"low":"high","medium":"high","high":"high","xhigh":"max"},
+		input: ["text"],
+		cost: {
+			input: 0,
+			output: 0,
+			cacheRead: 0,
+			cacheWrite: 0,
+		},
+		contextWindow: 1000000,
+		maxTokens: 131072,
+	} satisfies Model<"openai-completions">,
 	"glm-5v-turbo": {
 		id: "glm-5v-turbo",
 		name: "GLM-5V-Turbo",

packages/ai/src/types.ts

@@ -74,6 +74,15 @@ export type ImagesProviderId = KnownImagesProvider | string;
 export type ThinkingLevel = "minimal" | "low" | "medium" | "high" | "xhigh";
 export type ModelThinkingLevel = "off" | ThinkingLevel;
 export type ThinkingLevelMap = Partial<Record<ModelThinkingLevel, string | null>>;
+export type ChatTemplateKwargValue =
+	| string
+	| number
+	| boolean
+	| null
+	| {
+			$var: "thinking.enabled" | "thinking.effort";
+			omitWhenOff?: boolean;
+	  };
 
 /** Token budgets for each thinking level (token-based providers only) */
 export interface ThinkingBudgets {
@@ -88,6 +97,9 @@ export type CacheRetention = "none" | "short" | "long";
 
 export type Transport = "sse" | "websocket" | "websocket-cached" | "auto";
 
+/** Provider-scoped environment overrides. Values take precedence over process.env. */
+export type ProviderEnv = Record<string, string>;
+
 export interface ProviderResponse {
 	status: number;
 	headers: Record<string, string>;
@@ -162,6 +174,12 @@ export interface StreamOptions {
 	 * For example, Anthropic uses `user_id` for abuse tracking and rate limiting.
 	 */
 	metadata?: Record<string, unknown>;
+	/**
+	 * Provider-scoped environment values. These take precedence over process.env for
+	 * provider configuration such as regional settings, endpoint placeholders, and
+	 * proxy variables.
+	 */
+	env?: ProviderEnv;
 }
 
 export type ProviderStreamOptions = StreamOptions & Record<string, unknown>;
@@ -328,6 +346,8 @@ export interface Usage {
 	output: number;
 	cacheRead: number;
 	cacheWrite: number;
+	/** Subset of `cacheWrite` written with 1h retention. Only Anthropic reports this split. */
+	cacheWrite1h?: number;
 	totalTokens: number;
 	cost: {
 		input: number;
@@ -453,7 +473,7 @@ export interface OpenAICompletionsCompat {
 	requiresThinkingAsText?: boolean;
 	/** Whether all replayed assistant messages must include an empty reasoning_content field when reasoning is enabled. Default: auto-detected from URL. */
 	requiresReasoningContentOnAssistantMessages?: boolean;
-	/** Format for reasoning/thinking parameter. "openai" uses reasoning_effort, "openrouter" uses reasoning: { effort }, "deepseek" uses thinking: { type } plus reasoning_effort when supported, "together" uses reasoning: { enabled } plus reasoning_effort when supported, "zai" uses thinking: { type }, "qwen" uses top-level enable_thinking: boolean, "qwen-chat-template" uses chat_template_kwargs.enable_thinking, "string-thinking" uses top-level thinking: string, and "ant-ling" uses reasoning: { effort } only when the mapped effort is non-null. Default: "openai". */
+	/** Format for reasoning/thinking parameter. "openai" uses reasoning_effort, "openrouter" uses reasoning: { effort }, "deepseek" uses thinking: { type } plus reasoning_effort when supported, "together" uses reasoning: { enabled } plus reasoning_effort when supported, "zai" uses thinking: { type }, "qwen" uses top-level enable_thinking: boolean, "qwen-chat-template" uses chat_template_kwargs.enable_thinking and preserve_thinking, "chat-template" uses configurable chat_template_kwargs, "string-thinking" uses top-level thinking: string, and "ant-ling" uses reasoning: { effort } only when the mapped effort is non-null. Default: "openai". */
 	thinkingFormat?:
 		| "openai"
 		| "openrouter"
@@ -461,9 +481,12 @@ export interface OpenAICompletionsCompat {
 		| "together"
 		| "zai"
 		| "qwen"
+		| "chat-template"
 		| "qwen-chat-template"
 		| "string-thinking"
 		| "ant-ling";
+	/** Kwargs to send as `chat_template_kwargs` when `thinkingFormat` is `chat-template`. Use `{ "$var": "thinking.enabled" }` or `{ "$var": "thinking.effort" }` for pi-controlled thinking values. */
+	chatTemplateKwargs?: Record<string, ChatTemplateKwargValue>;
 	/** OpenRouter-compatible routing preferences sent as the `provider` request field. */
 	openRouterRouting?: OpenRouterRouting;
 	/** Vercel AI Gateway routing preferences. Only used when baseUrl points to Vercel AI Gateway. */

packages/ai/src/utils/node-http-proxy.ts

@@ -1,7 +1,5 @@
-import type { Agent as HttpAgent } from "node:http";
-import type { Agent as HttpsAgent } from "node:https";
-import { HttpProxyAgent } from "http-proxy-agent";
-import { HttpsProxyAgent } from "https-proxy-agent";
+import type { ProviderEnv } from "../types.ts";
+import { getProviderEnvValue } from "./provider-env.ts";
 
 const DEFAULT_PROXY_PORTS: Record<string, number> = {
 	ftp: 21,
@@ -12,16 +10,16 @@ const DEFAULT_PROXY_PORTS: Record<string, number> = {
 	wss: 443,
 };
 
-export interface NodeHttpProxyAgents {
-	httpAgent: HttpAgent;
-	httpsAgent: HttpsAgent;
-}
-
-export const UNSUPPORTED_PROXY_PROTOCOL_MESSAGE =
-	"Unsupported proxy protocol. SOCKS and PAC proxy URLs are not supported; use an HTTP or HTTPS proxy URL.";
-
-function getProxyEnv(key: string): string {
-	return process.env[key.toLowerCase()] || process.env[key.toUpperCase()] || "";
+function getProxyEnv(key: string, env?: ProviderEnv): string {
+	const lowercaseKey = key.toLowerCase();
+	const uppercaseKey = key.toUpperCase();
+	return (
+		env?.[lowercaseKey] ||
+		env?.[uppercaseKey] ||
+		getProviderEnvValue(lowercaseKey) ||
+		getProviderEnvValue(uppercaseKey) ||
+		""
+	);
 }
 
 function parseProxyTargetUrl(targetUrl: string | URL): URL | undefined {
@@ -36,8 +34,8 @@ function parseProxyTargetUrl(targetUrl: string | URL): URL | undefined {
 	}
 }
 
-function shouldProxyHostname(hostname: string, port: number): boolean {
-	const noProxy = getProxyEnv("no_proxy").toLowerCase();
+function shouldProxyHostname(hostname: string, port: number, env?: ProviderEnv): boolean {
+	const noProxy = getProxyEnv("no_proxy", env).toLowerCase();
 	if (!noProxy) {
 		return true;
 	}
@@ -68,7 +66,7 @@ function shouldProxyHostname(hostname: string, port: number): boolean {
 	});
 }
 
-function getProxyForUrl(targetUrl: string | URL): string {
+function getProxyForUrl(targetUrl: string | URL, env?: ProviderEnv): string {
 	const parsedUrl = parseProxyTargetUrl(targetUrl);
 	if (!parsedUrl?.protocol || !parsedUrl.host) {
 		return "";
@@ -77,19 +75,22 @@ function getProxyForUrl(targetUrl: string | URL): string {
 	const protocol = parsedUrl.protocol.split(":", 1)[0]!;
 	const hostname = parsedUrl.host.replace(/:\d*$/, "");
 	const port = Number.parseInt(parsedUrl.port, 10) || DEFAULT_PROXY_PORTS[protocol] || 0;
-	if (!shouldProxyHostname(hostname, port)) {
+	if (!shouldProxyHostname(hostname, port, env)) {
 		return "";
 	}
 
-	let proxy = getProxyEnv(`${protocol}_proxy`) || getProxyEnv("all_proxy");
+	let proxy = getProxyEnv(`${protocol}_proxy`, env) || getProxyEnv("all_proxy", env);
 	if (proxy && !proxy.includes("://")) {
 		proxy = `${protocol}://${proxy}`;
 	}
 	return proxy;
 }
 
-export function resolveHttpProxyUrlForTarget(targetUrl: string | URL): URL | undefined {
-	const proxy = getProxyForUrl(targetUrl);
+export const UNSUPPORTED_PROXY_PROTOCOL_MESSAGE =
+	"Unsupported proxy protocol. SOCKS and PAC proxy URLs are not supported; use an HTTP or HTTPS proxy URL.";
+
+export function resolveHttpProxyUrlForTarget(targetUrl: string | URL, env?: ProviderEnv): URL | undefined {
+	const proxy = getProxyForUrl(targetUrl, env);
 	if (!proxy) {
 		return undefined;
 	}
@@ -109,15 +110,3 @@ export function resolveHttpProxyUrlForTarget(targetUrl: string | URL): URL | und
 
 	return proxyUrl;
 }
-
-export function createHttpProxyAgentsForTarget(targetUrl: string | URL): NodeHttpProxyAgents | undefined {
-	const proxyUrl = resolveHttpProxyUrlForTarget(targetUrl);
-	if (!proxyUrl) {
-		return undefined;
-	}
-
-	return {
-		httpAgent: new HttpProxyAgent(proxyUrl),
-		httpsAgent: new HttpsProxyAgent(proxyUrl) as unknown as HttpsAgent,
-	};
-}

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

@@ -7,6 +7,7 @@
 
 import type { Server } from "node:http";
 import type { OAuthAuth } from "../../auth/types.ts";
+import { getProviderEnvValue } from "../provider-env.ts";
 import { oauthErrorHtml, oauthSuccessHtml } from "./oauth-page.ts";
 import { generatePKCE } from "./pkce.ts";
 import type { OAuthCredentials, OAuthLoginCallbacks, OAuthPrompt, OAuthProviderInterface } from "./types.ts";
@@ -29,7 +30,7 @@ const decode = (s: string) => atob(s);
 const CLIENT_ID = decode("OWQxYzI1MGEtZTYxYi00NGQ5LTg4ZWQtNTk0NGQxOTYyZjVl");
 const AUTHORIZE_URL = "https://claude.ai/oauth/authorize";
 const TOKEN_URL = "https://platform.claude.com/v1/oauth/token";
-const CALLBACK_HOST = process.env.PI_OAUTH_CALLBACK_HOST || "127.0.0.1";
+const CALLBACK_HOST = getProviderEnvValue("PI_OAUTH_CALLBACK_HOST") || "127.0.0.1";
 const CALLBACK_PORT = 53692;
 const CALLBACK_PATH = "/callback";
 const REDIRECT_URI = `http://localhost:${CALLBACK_PORT}${CALLBACK_PATH}`;

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

@@ -10,6 +10,7 @@ import type { OAuthCredentials, OAuthDeviceCodeInfo, OAuthLoginCallbacks, OAuthP
 
 type CopilotCredentials = OAuthCredentials & {
 	enterpriseUrl?: string;
+	availableModelIds: string[];
 };
 
 const decode = (s: string) => atob(s);
@@ -21,6 +22,7 @@ const COPILOT_HEADERS = {
 	"Editor-Plugin-Version": "copilot-chat/0.35.0",
 	"Copilot-Integration-Id": "vscode-chat",
 } as const;
+const COPILOT_API_VERSION = "2026-06-01";
 
 type DeviceCodeResponse = {
 	device_code: string;
@@ -89,6 +91,48 @@ export function getGitHubCopilotBaseUrl(token?: string, enterpriseDomain?: strin
 	return "https://api.individual.githubcopilot.com";
 }
 
+function asRecord(value: unknown): Record<string, unknown> | undefined {
+	return value && typeof value === "object" ? (value as Record<string, unknown>) : undefined;
+}
+
+function isSelectableCopilotModel(item: Record<string, unknown>): boolean {
+	const policy = asRecord(item.policy);
+	const capabilities = asRecord(item.capabilities);
+	const supports = asRecord(capabilities?.supports);
+	return item.model_picker_enabled === true && policy?.state !== "disabled" && supports?.tool_calls !== false;
+}
+
+function parseAvailableCopilotModelIds(raw: unknown): string[] {
+	const data = asRecord(raw)?.data;
+	if (!Array.isArray(data)) {
+		throw new Error("Invalid Copilot models response");
+	}
+
+	const ids: string[] = [];
+	for (const rawItem of data) {
+		const item = asRecord(rawItem);
+		const id = item?.id;
+		if (typeof id === "string" && item && isSelectableCopilotModel(item)) {
+			ids.push(id);
+		}
+	}
+	return ids;
+}
+
+async function fetchAvailableGitHubCopilotModelIds(copilotToken: string, enterpriseDomain?: string): Promise<string[]> {
+	const baseUrl = getGitHubCopilotBaseUrl(copilotToken, enterpriseDomain);
+	const raw = await fetchJson(`${baseUrl}/models`, {
+		headers: {
+			Accept: "application/json",
+			Authorization: `Bearer ${copilotToken}`,
+			...COPILOT_HEADERS,
+			"X-GitHub-Api-Version": COPILOT_API_VERSION,
+		},
+		signal: AbortSignal.timeout(5000),
+	});
+	return parseAvailableCopilotModelIds(raw);
+}
+
 async function fetchJson(url: string, init: RequestInit): Promise<unknown> {
 	const response = await fetch(url, init);
 	if (!response.ok) {
@@ -202,10 +246,7 @@ async function pollForGitHubAccessToken(
 	});
 }
 
-/**
- * Refresh GitHub Copilot token
- */
-export async function refreshGitHubCopilotToken(
+async function refreshGitHubCopilotAccessToken(
 	refreshToken: string,
 	enterpriseDomain?: string,
 ): Promise<OAuthCredentials> {
@@ -239,6 +280,20 @@ export async function refreshGitHubCopilotToken(
 	};
 }
 
+/**
+ * Refresh GitHub Copilot token
+ */
+export async function refreshGitHubCopilotToken(
+	refreshToken: string,
+	enterpriseDomain?: string,
+): Promise<OAuthCredentials> {
+	const credentials = await refreshGitHubCopilotAccessToken(refreshToken, enterpriseDomain);
+	return {
+		...credentials,
+		availableModelIds: await fetchAvailableGitHubCopilotModelIds(credentials.access, enterpriseDomain),
+	};
+}
+
 /**
  * Enable a model for the user's GitHub Copilot account.
  * This is required for some models (like Claude, Grok) before they can be used.
@@ -323,12 +378,18 @@ export async function loginGitHubCopilot(options: {
 	});
 
 	const githubAccessToken = await pollForGitHubAccessToken(domain, device, options.signal);
-	const credentials = await refreshGitHubCopilotToken(githubAccessToken, enterpriseDomain ?? undefined);
+	const credentials = await refreshGitHubCopilotAccessToken(githubAccessToken, enterpriseDomain ?? undefined);
 
 	// Enable all models after successful login
 	options.onProgress?.("Enabling models...");
 	await enableAllGitHubCopilotModels(credentials.access, enterpriseDomain ?? undefined);
-	return credentials;
+
+	// 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),
+	};
 }
 
 function copilotEnterpriseDomain(credential: OAuthCredential): string | undefined {
@@ -393,6 +454,14 @@ export const githubCopilotOAuthProvider: OAuthProviderInterface = {
 		const creds = credentials as CopilotCredentials;
 		const domain = creds.enterpriseUrl ? (normalizeDomain(creds.enterpriseUrl) ?? undefined) : undefined;
 		const baseUrl = getGitHubCopilotBaseUrl(creds.access, domain);
-		return models.map((m) => (m.provider === "github-copilot" ? { ...m, baseUrl } : m));
+		// 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/utils/oauth/openai-codex.ts

@@ -18,6 +18,7 @@ if (typeof process !== "undefined" && (process.versions?.node || process.version
 }
 
 import type { OAuthAuth } from "../../auth/types.ts";
+import { getProviderEnvValue } from "../provider-env.ts";
 import { pollOAuthDeviceCodeFlow } from "./device-code.ts";
 import { oauthErrorHtml, oauthSuccessHtml } from "./oauth-page.ts";
 import { generatePKCE } from "./pkce.ts";
@@ -48,7 +49,7 @@ type OAuthToken = { access: string; refresh: string; expires: number };
 type TokenOperation = "exchange" | "refresh";
 
 function getCallbackHost(): string {
-	return typeof process !== "undefined" ? process.env.PI_OAUTH_CALLBACK_HOST || "127.0.0.1" : "127.0.0.1";
+	return getProviderEnvValue("PI_OAUTH_CALLBACK_HOST") || "127.0.0.1";
 }
 
 type DeviceAuthInfo = {

packages/ai/src/utils/overflow.ts

@@ -12,6 +12,7 @@ import type { AssistantMessage } from "../types.ts";
  * - Anthropic: "413 {\"error\":{\"type\":\"request_too_large\",\"message\":\"Request exceeds the maximum size\"}}"
  * - OpenAI: "Your input exceeds the context window of this model"
  * - OpenAI/LiteLLM: "Requested token count exceeds the model's maximum context length of 131072 tokens"
+ * - OpenAI-compatible: "Input length (265330) exceeds model's maximum context length (262144)."
  * - Google: "The input token count (1196265) exceeds the maximum number of tokens allowed (1048575)"
  * - xAI: "This model's maximum prompt length is 131072 but the request contains 537812 tokens"
  * - Groq: "Please reduce the length of the messages or completion"
@@ -36,7 +37,7 @@ const OVERFLOW_PATTERNS = [
 	/request_too_large/i, // Anthropic request byte-size overflow (HTTP 413)
 	/input is too long for requested model/i, // Amazon Bedrock
 	/exceeds the context window/i, // OpenAI (Completions & Responses API)
-	/exceeds (?:the )?(?:model'?s )?maximum context length of [\d,]+ tokens?/i, // OpenAI-compatible proxies (LiteLLM)
+	/exceeds (?:the )?(?:model'?s )?maximum context length(?: of [\d,]+ tokens?|\s*\([\d,]+\))/i, // OpenAI-compatible proxies (LiteLLM)
 	/input token count.*exceeds the maximum/i, // Google (Gemini)
 	/maximum prompt length is \d+/i, // xAI (Grok)
 	/reduce the length of the messages/i, // Groq
@@ -85,7 +86,7 @@ const NON_OVERFLOW_PATTERNS = [
  *
  * **Reliable detection (returns error with detectable message):**
  * - Anthropic: "prompt is too long: X tokens > Y maximum" or "request_too_large"
- * - OpenAI (Completions & Responses): "exceeds the context window" or "exceeds the model's maximum context length of X tokens"
+ * - OpenAI (Completions & Responses): "exceeds the context window", "exceeds the model's maximum context length of X tokens", or "exceeds model's maximum context length (X)"
  * - Google Gemini: "input token count exceeds the maximum"
  * - xAI (Grok): "maximum prompt length is X but request contains Y"
  * - Groq: "reduce the length of the messages"

packages/ai/src/utils/provider-env.ts

@@ -0,0 +1,52 @@
+import type { ProviderEnv } from "../types.ts";
+
+let procEnvCache: Map<string, string> | null = null;
+
+/**
+ * Fallback for https://github.com/oven-sh/bun/issues/27802.
+ * Bun compiled binaries can expose an empty process.env inside Linux sandboxes
+ * even though /proc/self/environ contains the environment.
+ *
+ * This intentionally duplicates restoreSandboxEnv() in
+ * packages/coding-agent/src/bun/restore-sandbox-env.ts. The ai package can be
+ * used directly, without going through that entrypoint, so provider env lookup
+ * must not depend on process.env having been patched.
+ */
+function getBunSandboxEnvValue(name: string): string | undefined {
+	if (typeof process === "undefined" || !process.versions?.bun || Object.keys(process.env).length > 0) {
+		return undefined;
+	}
+
+	if (procEnvCache === null) {
+		procEnvCache = new Map();
+		try {
+			const { readFileSync } = require("node:fs") as {
+				readFileSync(path: string, encoding: BufferEncoding): string;
+			};
+			const data = readFileSync("/proc/self/environ", "utf-8");
+			for (const entry of data.split("\0")) {
+				const idx = entry.indexOf("=");
+				if (idx > 0) {
+					procEnvCache.set(entry.slice(0, idx), entry.slice(idx + 1));
+				}
+			}
+		} catch {
+			// /proc/self/environ may not exist or may not be readable.
+		}
+	}
+
+	return procEnvCache.get(name);
+}
+
+/**
+ * Resolve a provider env value from scoped overrides, normal process.env, then
+ * the duplicated Bun sandbox fallback for direct pi-ai consumers.
+ */
+export function getProviderEnvValue(name: string, env?: ProviderEnv): string | undefined {
+	return (
+		env?.[name] ||
+		(typeof process !== "undefined" ? process.env[name] : undefined) ||
+		getBunSandboxEnvValue(name) ||
+		undefined
+	);
+}

packages/ai/test/anthropic-adaptive-thinking-models.test.ts

@@ -5,9 +5,8 @@ import type { Api, Model } from "../src/types.ts";
 const EXPECTED_CURRENT_ADAPTIVE_THINKING_MODELS = [
 	"anthropic/claude-fable-5",
 	"anthropic/claude-opus-4-8",
-	"opencode/claude-fable-5",
+	"cloudflare-ai-gateway/claude-fable-5",
 	"opencode/claude-opus-4-8",
-	"vercel-ai-gateway/anthropic/claude-fable-5",
 	"vercel-ai-gateway/anthropic/claude-opus-4.8",
 ];
 

packages/ai/test/anthropic-cache-write-1h-cost.test.ts

@@ -0,0 +1,86 @@
+import type Anthropic from "@anthropic-ai/sdk";
+import { describe, expect, it } from "vitest";
+import { stream as streamAnthropic } from "../src/api/anthropic-messages.ts";
+import { getModel } from "../src/compat.ts";
+import type { Context } from "../src/types.ts";
+
+function createSseResponse(events: Array<{ event: string; data: string }>): Response {
+	const body = events.map(({ event, data }) => `event: ${event}\ndata: ${data}\n`).join("\n");
+	return new Response(body, { status: 200, headers: { "content-type": "text/event-stream" } });
+}
+
+function createFakeAnthropicClient(response: Response): Anthropic {
+	return {
+		messages: { create: () => ({ asResponse: async () => response }) },
+	} as unknown as Anthropic;
+}
+
+function eventsWithCacheCreation(
+	cacheCreation: Record<string, number> | undefined,
+): Array<{ event: string; data: string }> {
+	const startUsage: Record<string, unknown> = {
+		input_tokens: 100,
+		output_tokens: 0,
+		cache_read_input_tokens: 0,
+		cache_creation_input_tokens: 1_000_000,
+	};
+	if (cacheCreation) startUsage.cache_creation = cacheCreation;
+	return [
+		{
+			event: "message_start",
+			data: JSON.stringify({ type: "message_start", message: { id: "msg_test", usage: startUsage } }),
+		},
+		{
+			event: "content_block_start",
+			data: JSON.stringify({ type: "content_block_start", index: 0, content_block: { type: "text", text: "" } }),
+		},
+		{
+			event: "content_block_delta",
+			data: JSON.stringify({ type: "content_block_delta", index: 0, delta: { type: "text_delta", text: "Hi" } }),
+		},
+		{ event: "content_block_stop", data: JSON.stringify({ type: "content_block_stop", index: 0 }) },
+		{
+			event: "message_delta",
+			data: JSON.stringify({
+				type: "message_delta",
+				delta: { stop_reason: "end_turn" },
+				usage: {
+					input_tokens: 100,
+					output_tokens: 5,
+					cache_read_input_tokens: 0,
+					cache_creation_input_tokens: 1_000_000,
+				},
+			}),
+		},
+		{ event: "message_stop", data: JSON.stringify({ type: "message_stop" }) },
+	];
+}
+
+// claude-opus-4-8: input 5, cacheWrite (5m) 6.25 per Mtok. 1h write = 2x input = 10.
+const context: Context = { messages: [{ role: "user", content: "hi", timestamp: Date.now() }] };
+
+describe("Anthropic 1h cache write cost", () => {
+	it("prices the 1h portion at 2x input and the rest at the 5m rate", async () => {
+		const model = getModel("anthropic", "claude-opus-4-8");
+		const response = createSseResponse(
+			eventsWithCacheCreation({ ephemeral_5m_input_tokens: 600_000, ephemeral_1h_input_tokens: 400_000 }),
+		);
+		const result = await streamAnthropic(model, context, { client: createFakeAnthropicClient(response) }).result();
+
+		expect(result.usage.cacheWrite).toBe(1_000_000);
+		expect(result.usage.cacheWrite1h).toBe(400_000);
+		// 600k * 6.25/Mtok + 400k * 10/Mtok = 3.75 + 4.0 = 7.75
+		expect(result.usage.cost.cacheWrite).toBeCloseTo(7.75, 10);
+	});
+
+	it("falls back to the 5m rate when no breakdown is reported", async () => {
+		const model = getModel("anthropic", "claude-opus-4-8");
+		const response = createSseResponse(eventsWithCacheCreation(undefined));
+		const result = await streamAnthropic(model, context, { client: createFakeAnthropicClient(response) }).result();
+
+		expect(result.usage.cacheWrite).toBe(1_000_000);
+		expect(result.usage.cacheWrite1h ?? 0).toBe(0);
+		// 1M * 6.25/Mtok = 6.25
+		expect(result.usage.cost.cacheWrite).toBeCloseTo(6.25, 10);
+	});
+});

packages/ai/test/anthropic-sse-parsing.test.ts

@@ -166,6 +166,64 @@ describe("Anthropic raw SSE parsing", () => {
 		});
 	});
 
+	it("preserves refusal stop details from message_delta", async () => {
+		const model = getModel("anthropic", "claude-fable-5");
+		const context: Context = {
+			messages: [{ role: "user", content: "blocked request", timestamp: Date.now() }],
+		};
+		const explanation =
+			"This request triggered restrictions on violative cyber content and was blocked under Anthropic's Usage Policy. To learn more, provide feedback, or request an exemption based on how you use Claude, visit our help center: https://support.claude.com/en/articles/14604842-real-time-cyber-safeguards-on-claude.";
+		const response = createSseResponse([
+			{
+				event: "message_start",
+				data: JSON.stringify({
+					type: "message_start",
+					message: {
+						id: "msg_01XFUDYJgAACzvnptvVoYEL",
+						usage: {
+							input_tokens: 412,
+							output_tokens: 0,
+							cache_read_input_tokens: 0,
+							cache_creation_input_tokens: 0,
+						},
+					},
+				}),
+			},
+			{
+				event: "message_delta",
+				data: JSON.stringify({
+					type: "message_delta",
+					delta: {
+						stop_reason: "refusal",
+						stop_details: {
+							type: "refusal",
+							category: "cyber",
+							explanation,
+						},
+					},
+					usage: {
+						input_tokens: 412,
+						output_tokens: 0,
+						cache_read_input_tokens: 0,
+						cache_creation_input_tokens: 0,
+					},
+				}),
+			},
+			{
+				event: "message_stop",
+				data: JSON.stringify({ type: "message_stop" }),
+			},
+		]);
+
+		const stream = streamAnthropic(model, context, {
+			client: createFakeAnthropicClient(response),
+		});
+		const result = await stream.result();
+
+		expect(result.stopReason).toBe("error");
+		expect(result.errorMessage).toBe(explanation);
+	});
+
 	it("ignores unknown SSE events after message_stop", async () => {
 		const model = getModel("anthropic", "claude-haiku-4-5");
 		const context: Context = {

packages/ai/test/bedrock-endpoint-resolution.test.ts

@@ -44,7 +44,7 @@ vi.mock("@aws-sdk/client-bedrock-runtime", () => {
 	};
 });
 
-import { stream as streamBedrock } from "../src/api/bedrock-converse-stream.ts";
+import { type BedrockOptions, stream as streamBedrock } from "../src/api/bedrock-converse-stream.ts";
 import { getModel } from "../src/compat.ts";
 import type { Context, Model } from "../src/types.ts";
 
@@ -83,8 +83,12 @@ afterEach(() => {
 	}
 });
 
-async function captureClientConfig(model: Model<"bedrock-converse-stream">): Promise<Record<string, unknown>> {
-	await streamBedrock(model, context, { cacheRetention: "none" }).result();
+async function captureClientConfig(
+	model: Model<"bedrock-converse-stream">,
+	options: BedrockOptions = {},
+): Promise<Record<string, unknown>> {
+	bedrockMock.constructorCalls.length = 0;
+	await streamBedrock(model, context, { cacheRetention: "none", ...options }).result();
 	expect(bedrockMock.constructorCalls).toHaveLength(1);
 	return bedrockMock.constructorCalls[0];
 }
@@ -115,6 +119,29 @@ describe("bedrock endpoint resolution", () => {
 		expect(config.region).toBe("eu-central-1");
 	});
 
+	it("handles missing regions for explicit, scoped, and ambient profiles", async () => {
+		const model = getModel("amazon-bedrock", "eu.anthropic.claude-sonnet-4-5-20250929-v1:0");
+
+		let config = await captureClientConfig(model, { profile: "bedrock-profile" });
+
+		expect(config.profile).toBe("bedrock-profile");
+		expect(config.endpoint).toBe("https://bedrock-runtime.eu-central-1.amazonaws.com");
+		expect(config.region).toBe("eu-central-1");
+
+		config = await captureClientConfig(model, { env: { AWS_PROFILE: "scoped-bedrock-profile" } });
+
+		expect(config.profile).toBe("scoped-bedrock-profile");
+		expect(config.endpoint).toBe("https://bedrock-runtime.eu-central-1.amazonaws.com");
+		expect(config.region).toBe("eu-central-1");
+
+		process.env.AWS_PROFILE = "ambient-bedrock-profile";
+		config = await captureClientConfig(model);
+
+		expect(config.profile).toBe("ambient-bedrock-profile");
+		expect(config.endpoint).toBeUndefined();
+		expect(config.region).toBeUndefined();
+	});
+
 	it("still passes custom Bedrock endpoints through to the SDK client", async () => {
 		process.env.AWS_REGION = "us-west-2";
 		const baseModel = getModel("amazon-bedrock", "us.anthropic.claude-opus-4-8");

packages/ai/test/cache-retention.test.ts

@@ -3,6 +3,7 @@ import { stream as streamAnthropic } from "../src/api/anthropic-messages.ts";
 import { stream as streamOpenAICompletions } from "../src/api/openai-completions.ts";
 import { stream as streamOpenAIResponses } from "../src/api/openai-responses.ts";
 import { getModel, stream } from "../src/compat.ts";
+import { MODELS } from "../src/models.generated.ts";
 import type { Context, Model } from "../src/types.ts";
 
 class PayloadCaptured extends Error {
@@ -12,6 +13,11 @@ class PayloadCaptured extends Error {
 	}
 }
 
+interface OpenAICompletionsCachePayload {
+	prompt_cache_key?: string;
+	prompt_cache_retention?: string;
+}
+
 function stopAfterPayload<TPayload>(capture: (payload: TPayload) => void): (payload: unknown) => never {
 	return (payload: unknown): never => {
 		capture(payload as TPayload);
@@ -454,5 +460,39 @@ describe("Cache Retention (PI_CACHE_RETENTION)", () => {
 			expect(capturedPayload.prompt_cache_key).toBeUndefined();
 			expect(capturedPayload.prompt_cache_retention).toBeUndefined();
 		});
+
+		it.each([
+			MODELS.opencode["deepseek-v4-flash"],
+			MODELS.opencode["deepseek-v4-pro"],
+			MODELS.opencode["kimi-k2.5"],
+			MODELS.opencode["kimi-k2.6"],
+			MODELS.opencode["minimax-m2.7"],
+			MODELS["opencode-go"]["kimi-k2.6"],
+		] as const)("should omit long cache retention for $provider/$id", async (metadata) => {
+			const model = metadata as Model<"openai-completions">;
+			let capturedPayload: OpenAICompletionsCachePayload | undefined;
+
+			try {
+				const s = streamOpenAICompletions(model, context, {
+					apiKey: "fake-key",
+					cacheRetention: "long",
+					sessionId: "session-opencode-long-cache-unsupported",
+					onPayload: stopAfterPayload<OpenAICompletionsCachePayload>((payload) => {
+						capturedPayload = payload;
+					}),
+				});
+
+				for await (const event of s) {
+					if (event.type === "error") break;
+				}
+			} catch {
+				// Expected to fail
+			}
+
+			expect(model.compat?.supportsLongCacheRetention).toBe(false);
+			expect(capturedPayload).toBeDefined();
+			expect(capturedPayload?.prompt_cache_key).toBeUndefined();
+			expect(capturedPayload?.prompt_cache_retention).toBeUndefined();
+		});
 	});
 });

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

@@ -1,6 +1,7 @@
 import { describe, expect, it, vi } from "vitest";
 import { stream as streamAnthropic } from "../src/api/anthropic-messages.ts";
 import { getModel } from "../src/compat.ts";
+import { getSupportedThinkingLevels } from "../src/models.ts";
 import type { Context } from "../src/types.ts";
 
 const mockState = vi.hoisted(() => ({
@@ -54,6 +55,16 @@ describe("Copilot Claude via Anthropic Messages", () => {
 		messages: [{ role: "user", content: "Hello", timestamp: Date.now() }],
 	};
 
+	it("applies Copilot-specific adaptive thinking effort overrides", () => {
+		const opus47 = getModel("github-copilot", "claude-opus-4.7");
+		expect(opus47.thinkingLevelMap).toMatchObject({ minimal: "low", xhigh: "xhigh" });
+		expect(getSupportedThinkingLevels(opus47)).toContain("xhigh");
+
+		const sonnet46 = getModel("github-copilot", "claude-sonnet-4.6");
+		expect(sonnet46.thinkingLevelMap).toMatchObject({ minimal: "low", xhigh: "max" });
+		expect(getSupportedThinkingLevels(sonnet46)).toContain("xhigh");
+	});
+
 	it("uses Bearer auth, Copilot headers, and valid Anthropic Messages payload", async () => {
 		const model = getModel("github-copilot", "claude-sonnet-4.6");
 		expect(model.api).toBe("anthropic-messages");

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

@@ -1,5 +1,10 @@
 import { afterEach, describe, expect, it, vi } from "vitest";
-import { loginGitHubCopilot } from "../src/utils/oauth/github-copilot.ts";
+import { getModels } from "../src/compat.ts";
+import {
+	githubCopilotOAuthProvider,
+	loginGitHubCopilot,
+	refreshGitHubCopilotToken,
+} from "../src/utils/oauth/github-copilot.ts";
 
 function jsonResponse(body: unknown, status: number = 200): Response {
 	return new Response(JSON.stringify(body), {
@@ -29,6 +34,57 @@ describe("GitHub Copilot OAuth device flow", () => {
 		vi.useRealTimers();
 	});
 
+	it("filters models to the authenticated account picker catalog", async () => {
+		const fetchMock = vi.fn(async (input: unknown, init?: RequestInit): Promise<Response> => {
+			const url = getUrl(input);
+
+			if (url.includes("/copilot_internal/v2/token")) {
+				return jsonResponse({
+					token: "tid=test;exp=9999999999;proxy-ep=proxy.individual.githubcopilot.com;",
+					expires_at: 9999999999,
+				});
+			}
+
+			if (url === "https://api.individual.githubcopilot.com/models") {
+				expect(init?.headers).toMatchObject({
+					Authorization: "Bearer tid=test;exp=9999999999;proxy-ep=proxy.individual.githubcopilot.com;",
+				});
+				return jsonResponse({
+					data: [
+						{
+							id: "gpt-4.1",
+							model_picker_enabled: true,
+							capabilities: { supports: { tool_calls: true } },
+						},
+						{
+							id: "claude-opus-4.7",
+							model_picker_enabled: true,
+							policy: { state: "disabled" },
+							capabilities: { supports: { tool_calls: true } },
+						},
+						{
+							id: "gpt-5.4-nano",
+							model_picker_enabled: false,
+							capabilities: { supports: { tool_calls: true } },
+						},
+					],
+				});
+			}
+
+			throw new Error(`Unexpected fetch URL: ${url}`);
+		});
+
+		vi.stubGlobal("fetch", fetchMock);
+
+		const credentials = await refreshGitHubCopilotToken("ghu_refresh_token");
+		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",
+		]);
+	});
+
 	it("reports device-code details through onDeviceCode", async () => {
 		vi.useFakeTimers();
 		vi.setSystemTime(new Date("2026-03-09T00:00:00Z"));
@@ -57,6 +113,10 @@ describe("GitHub Copilot OAuth device flow", () => {
 				});
 			}
 
+			if (url.endsWith("/models")) {
+				return jsonResponse({ data: [] });
+			}
+
 			if (url.includes("/models/") && url.endsWith("/policy")) {
 				return new Response("", { status: 200 });
 			}
@@ -146,6 +206,10 @@ describe("GitHub Copilot OAuth device flow", () => {
 				});
 			}
 
+			if (url.endsWith("/models")) {
+				return jsonResponse({ data: [] });
+			}
+
 			if (url.includes("/models/") && url.endsWith("/policy")) {
 				return new Response("", { status: 200 });
 			}
@@ -231,6 +295,10 @@ describe("GitHub Copilot OAuth device flow", () => {
 				});
 			}
 
+			if (url.endsWith("/models")) {
+				return jsonResponse({ data: [] });
+			}
+
 			if (url.includes("/models/") && url.endsWith("/policy")) {
 				return new Response("", { status: 200 });
 			}

packages/ai/test/mistral-reasoning-mode.test.ts

@@ -5,6 +5,7 @@ import type { Context, Model, SimpleStreamOptions } from "../src/types.ts";
 interface MistralPayload {
 	promptMode?: "reasoning";
 	reasoningEffort?: "none" | "high";
+	promptCacheKey?: string;
 }
 
 function makeContext(): Context {
@@ -76,4 +77,21 @@ describe("Mistral reasoning mode selection", () => {
 		expect(payload.reasoningEffort).toBeUndefined();
 		expect(payload.promptMode).toBeUndefined();
 	});
+
+	it("uses the session id as prompt cache key", async () => {
+		const payload = await capturePayload(getModel("mistral", "mistral-large-latest"), {
+			sessionId: "session-123",
+		});
+
+		expect(payload.promptCacheKey).toBe("session-123");
+	});
+
+	it("omits prompt cache key when cache retention is disabled", async () => {
+		const payload = await capturePayload(getModel("mistral", "mistral-large-latest"), {
+			sessionId: "session-123",
+			cacheRetention: "none",
+		});
+
+		expect(payload.promptCacheKey).toBeUndefined();
+	});
 });

packages/ai/test/node-http-proxy.test.ts

@@ -54,6 +54,17 @@ describe("node HTTP proxy resolution", () => {
 		);
 	});
 
+	it("prefers scoped proxy env aliases before process env aliases", () => {
+		resetProxyEnv();
+		process.env.https_proxy = "http://process-proxy.example:8080";
+
+		expect(
+			resolveHttpProxyUrlForTarget("https://bedrock-runtime.us-east-1.amazonaws.com", {
+				HTTPS_PROXY: "http://scoped-proxy.example:8080",
+			})?.toString(),
+		).toBe("http://scoped-proxy.example:8080/");
+	});
+
 	it("rejects SOCKS and PAC proxy URLs explicitly", () => {
 		resetProxyEnv();
 		process.env.HTTPS_PROXY = "socks5://proxy.example:1080";

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

@@ -361,13 +361,21 @@ describe("openai-codex streaming", () => {
 			apiKey: token,
 			transport: "sse",
 		}).result();
+		let settled = false;
+		const observedResultPromise = resultPromise.then((result) => {
+			settled = true;
+			return result;
+		});
 		await vi.advanceTimersByTimeAsync(0);
 		expect(fetchMock).toHaveBeenCalledTimes(1);
 
 		await vi.advanceTimersByTimeAsync(10_000);
-		const result = await resultPromise;
+		expect(settled).toBe(false);
+
+		await vi.advanceTimersByTimeAsync(10_000);
+		const result = await observedResultPromise;
 		expect(result.stopReason).toBe("error");
-		expect(result.errorMessage).toBe("Codex SSE response headers timed out after 10000ms");
+		expect(result.errorMessage).toBe("Codex SSE response headers timed out after 20000ms");
 	});
 
 	it("aborts SSE body reads after response headers arrive", async () => {

packages/ai/test/openai-completions-empty-tools.test.ts

@@ -162,6 +162,31 @@ describe("openai-completions empty tools handling", () => {
 		expect(clientOptions.defaultHeaders?.["cf-aig-authorization"]).toBe("Bearer test");
 	});
 
+	it("uses provider env before process.env for Cloudflare AI Gateway base URL", async () => {
+		process.env.CLOUDFLARE_ACCOUNT_ID = "process-account";
+		process.env.CLOUDFLARE_GATEWAY_ID = "process-gateway";
+		const model = getModel("cloudflare-ai-gateway", "workers-ai/@cf/moonshotai/kimi-k2.6")!;
+
+		await streamSimple(
+			model,
+			{
+				messages: [{ role: "user", content: "hi", timestamp: Date.now() }],
+			},
+			{
+				apiKey: "test",
+				env: {
+					CLOUDFLARE_ACCOUNT_ID: "provider-account",
+					CLOUDFLARE_GATEWAY_ID: "provider-gateway",
+				},
+			},
+		).result();
+
+		const clientOptions = mockState.lastClientOptions as { baseURL?: string };
+		expect(clientOptions.baseURL).toBe(
+			"https://gateway.ai.cloudflare.com/v1/provider-account/provider-gateway/compat",
+		);
+	});
+
 	it("preserves inline upstream Authorization for Cloudflare AI Gateway BYOK requests", async () => {
 		process.env.CLOUDFLARE_ACCOUNT_ID = "account-id";
 		process.env.CLOUDFLARE_GATEWAY_ID = "gateway-id";

packages/ai/test/openai-completions-reasoning-details.test.ts

@@ -0,0 +1,118 @@
+import { Type } from "typebox";
+import { beforeEach, describe, expect, it, vi } from "vitest";
+import { stream as streamOpenAICompletions } from "../src/api/openai-completions.ts";
+import type { AssistantMessage, Model, Tool } from "../src/types.ts";
+
+const mockState = vi.hoisted(() => ({
+	chunkSets: [] as unknown[][],
+	payloads: [] as unknown[],
+}));
+
+vi.mock("openai", () => {
+	class FakeOpenAI {
+		chat = {
+			completions: {
+				create: (payload: unknown) => {
+					mockState.payloads.push(payload);
+					const chunks = mockState.chunkSets.shift() ?? [];
+					const stream = {
+						async *[Symbol.asyncIterator]() {
+							for (const chunk of chunks) {
+								yield chunk;
+							}
+						},
+					};
+					const result = Promise.resolve(stream) as Promise<typeof stream> & {
+						withResponse: () => Promise<{ data: typeof stream; response: { status: number; headers: Headers } }>;
+					};
+					result.withResponse = async () => ({
+						data: stream,
+						response: { status: 200, headers: new Headers() },
+					});
+					return result;
+				},
+			},
+		};
+	}
+	return { default: FakeOpenAI };
+});
+
+const reasoningDetail = { type: "reasoning.encrypted", id: "call_1", data: "encrypted-signature" };
+const readTool: Tool = {
+	name: "read",
+	description: "Read a file",
+	parameters: Type.Object({ path: Type.String() }),
+};
+
+function model(): Model<"openai-completions"> {
+	return {
+		id: "google/gemini-test",
+		name: "Gemini Test",
+		api: "openai-completions",
+		provider: "openrouter",
+		baseUrl: "https://openrouter.ai/api/v1",
+		reasoning: true,
+		input: ["text"],
+		cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+		contextWindow: 100_000,
+		maxTokens: 4096,
+	};
+}
+
+function chunk(delta: Record<string, unknown>, finishReason: string | null = null): unknown {
+	return {
+		id: "chatcmpl-test",
+		model: "google/gemini-test",
+		choices: [{ index: 0, delta, finish_reason: finishReason }],
+	};
+}
+
+function toolCallChunk(): unknown {
+	return chunk({
+		tool_calls: [
+			{
+				index: 0,
+				id: "call_1",
+				type: "function",
+				function: { name: "read", arguments: '{"path":"README.md"}' },
+			},
+		],
+	});
+}
+
+async function runOpenAICompletionsStream(messages: AssistantMessage[] = []): Promise<AssistantMessage> {
+	return await streamOpenAICompletions(model(), { messages, tools: [readTool] }, { apiKey: "test" }).result();
+}
+
+function getAssistantPayload(payload: unknown): { reasoning_details?: unknown } | undefined {
+	const messages = (payload as { messages?: Array<{ role?: string; reasoning_details?: unknown }> }).messages ?? [];
+	return messages.find((message) => message.role === "assistant");
+}
+
+describe("openai-completions reasoning_details streaming", () => {
+	beforeEach(() => {
+		mockState.chunkSets = [];
+		mockState.payloads = [];
+	});
+
+	it("preserves reasoning_details that arrive before their matching tool call", async () => {
+		mockState.chunkSets = [
+			[chunk({ reasoning_details: [reasoningDetail] }), toolCallChunk(), chunk({}, "tool_calls")],
+			[chunk({ content: "ok" }), chunk({}, "stop")],
+		];
+
+		const assistantMessage = await runOpenAICompletionsStream();
+		const toolCall = assistantMessage.content.find((block) => block.type === "toolCall");
+		expect(toolCall).toMatchObject({
+			type: "toolCall",
+			id: "call_1",
+			name: "read",
+			arguments: { path: "README.md" },
+			thoughtSignature: JSON.stringify(reasoningDetail),
+		});
+
+		await runOpenAICompletionsStream([assistantMessage]);
+
+		expect(getAssistantPayload(mockState.payloads[1])?.reasoning_details).toEqual([reasoningDetail]);
+	});
+});

packages/ai/test/openai-completions-thinking-as-text.test.ts

@@ -34,6 +34,7 @@ const compat = {
 	thinkingFormat: "openai",
 	openRouterRouting: {},
 	vercelGatewayRouting: {},
+	chatTemplateKwargs: {},
 	zaiToolStream: false,
 	supportsStrictMode: true,
 	cacheControlFormat: undefined,

packages/ai/test/openai-completions-tool-choice.test.ts

@@ -2,7 +2,7 @@ import { Type } from "typebox";
 import { beforeEach, describe, expect, it, vi } from "vitest";
 import { convertMessages } from "../src/api/openai-completions.ts";
 import { getModel, stream, streamSimple } from "../src/compat.ts";
-import type { AssistantMessage, Model, Tool, ToolResultMessage } from "../src/types.ts";
+import type { AssistantMessage, Model, SimpleStreamOptions, Tool, ToolResultMessage } from "../src/types.ts";
 
 const mockState = vi.hoisted(() => ({
 	lastParams: undefined as unknown,
@@ -63,6 +63,46 @@ vi.mock("openai", () => {
 	return { default: FakeOpenAI };
 });
 
+const localOpenAICompletionsModel = {
+	api: "openai-completions",
+	provider: "local-vllm",
+	baseUrl: "http://localhost:8000/v1",
+	reasoning: true,
+	input: ["text"],
+	cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+	contextWindow: 128000,
+	maxTokens: 8192,
+} satisfies Omit<Model<"openai-completions">, "id" | "name" | "compat">;
+
+type CapturedParams = {
+	chat_template_kwargs?: Record<string, unknown>;
+	thinking?: unknown;
+	reasoning_effort?: string;
+};
+
+async function captureSimpleParams(
+	model: Model<"openai-completions">,
+	reasoning?: SimpleStreamOptions["reasoning"],
+): Promise<CapturedParams> {
+	let payload: unknown;
+
+	await streamSimple(
+		model,
+		{
+			messages: [{ role: "user", content: "Hi", timestamp: Date.now() }],
+		},
+		{
+			apiKey: "test",
+			reasoning,
+			onPayload: (params: unknown) => {
+				payload = params;
+			},
+		},
+	).result();
+
+	return (payload ?? mockState.lastParams) as CapturedParams;
+}
+
 describe("openai-completions tool_choice", () => {
 	beforeEach(() => {
 		mockState.lastParams = undefined;
@@ -256,6 +296,86 @@ describe("openai-completions tool_choice", () => {
 		expect(getModel("zai", "glm-4.5-air")?.compat?.zaiToolStream).toBeUndefined();
 	});
 
+	it("stores z.ai GLM-5.2 effort metadata", () => {
+		for (const provider of ["zai", "zai-coding-cn"] as const) {
+			const model = getModel(provider, "glm-5.2")!;
+			expect(model.compat?.supportsReasoningEffort).toBe(true);
+			expect(model.thinkingLevelMap).toEqual({
+				minimal: null,
+				low: "high",
+				medium: "high",
+				high: "high",
+				xhigh: "max",
+			});
+		}
+	});
+
+	it("maps z.ai GLM-5.2 thinking levels to reasoning_effort", async () => {
+		const model = getModel("zai", "glm-5.2")!;
+		const cases = [
+			{ reasoning: "low", effort: "high" },
+			{ reasoning: "medium", effort: "high" },
+			{ reasoning: "high", effort: "high" },
+			{ reasoning: "xhigh", effort: "max" },
+		] as const;
+
+		for (const testCase of cases) {
+			let payload: unknown;
+
+			await streamSimple(
+				model,
+				{
+					messages: [
+						{
+							role: "user",
+							content: "Hi",
+							timestamp: Date.now(),
+						},
+					],
+				},
+				{
+					apiKey: "test",
+					reasoning: testCase.reasoning,
+					onPayload: (params: unknown) => {
+						payload = params;
+					},
+				},
+			).result();
+
+			const params = (payload ?? mockState.lastParams) as { thinking?: unknown; reasoning_effort?: string };
+			expect(params.thinking).toEqual({ type: "enabled" });
+			expect(params.reasoning_effort).toBe(testCase.effort);
+		}
+	});
+
+	it("omits z.ai GLM-5.2 reasoning_effort when thinking is off", async () => {
+		const model = getModel("zai", "glm-5.2")!;
+		let payload: unknown;
+
+		await streamSimple(
+			model,
+			{
+				messages: [
+					{
+						role: "user",
+						content: "Hi",
+						timestamp: Date.now(),
+					},
+				],
+			},
+			{
+				apiKey: "test",
+				onPayload: (params: unknown) => {
+					payload = params;
+				},
+			},
+		).result();
+
+		const params = (payload ?? mockState.lastParams) as { thinking?: unknown; reasoning_effort?: string };
+		expect(params.thinking).toEqual({ type: "disabled" });
+		expect(params.reasoning_effort).toBeUndefined();
+	});
+
 	it("omits tool_stream for unsupported z.ai models", async () => {
 		const model = getModel("zai", "glm-4.5-air")!;
 		const tools: Tool[] = [
@@ -1063,6 +1183,7 @@ describe("openai-completions tool_choice", () => {
 				thinkingFormat: "openai",
 				openRouterRouting: {},
 				vercelGatewayRouting: {},
+				chatTemplateKwargs: {},
 				zaiToolStream: false,
 				supportsStrictMode: true,
 				sendSessionAffinityHeaders: false,
@@ -1119,6 +1240,54 @@ describe("openai-completions tool_choice", () => {
 		expect(params.reasoning_effort).toBeUndefined();
 	});
 
+	it("omits disabled thinking for Moonshot Kimi K2.7 Code models", async () => {
+		const cases = [getModel("moonshotai", "kimi-k2.7-code"), getModel("moonshotai-cn", "kimi-k2.7-code")];
+
+		for (const model of cases) {
+			expect(model).toBeDefined();
+			let payload: unknown;
+
+			await streamSimple(
+				model!,
+				{
+					messages: [{ role: "user", content: "Hi", timestamp: Date.now() }],
+				},
+				{
+					apiKey: "test",
+					onPayload: (params: unknown) => {
+						payload = params;
+					},
+				},
+			).result();
+
+			const params = (payload ?? mockState.lastParams) as { thinking?: unknown; reasoning_effort?: string };
+			expect(params.thinking).toBeUndefined();
+			expect(params.reasoning_effort).toBeUndefined();
+		}
+	});
+
+	it("keeps disabled thinking for Moonshot Kimi K2.6 when thinking is off", async () => {
+		const model = getModel("moonshotai-cn", "kimi-k2.6")!;
+		let payload: unknown;
+
+		await streamSimple(
+			model,
+			{
+				messages: [{ role: "user", content: "Hi", timestamp: Date.now() }],
+			},
+			{
+				apiKey: "test",
+				onPayload: (params: unknown) => {
+					payload = params;
+				},
+			},
+		).result();
+
+		const params = (payload ?? mockState.lastParams) as { thinking?: unknown; reasoning_effort?: string };
+		expect(params.thinking).toEqual({ type: "disabled" });
+		expect(params.reasoning_effort).toBeUndefined();
+	});
+
 	it("sends max_tokens for OpenCode completions models", async () => {
 		const cases = [getModel("opencode-go", "kimi-k2.6")!, getModel("opencode", "grok-build-0.1")!] as const;
 
@@ -1322,6 +1491,77 @@ describe("openai-completions tool_choice", () => {
 		expect(params.reasoning_effort).toBeUndefined();
 	});
 
+	it("uses configurable chat template boolean thinking kwargs", async () => {
+		const model = {
+			...localOpenAICompletionsModel,
+			id: "deepseek-ai/DeepSeek-V3.1",
+			name: "DeepSeek V3.1 via vLLM",
+			compat: {
+				thinkingFormat: "chat-template",
+				supportsReasoningEffort: false,
+				chatTemplateKwargs: { thinking: { $var: "thinking.enabled" } },
+			},
+		} satisfies Model<"openai-completions">;
+
+		for (const testCase of [
+			{ reasoning: "high" as const, expected: true },
+			{ reasoning: undefined, expected: false },
+		]) {
+			const params = await captureSimpleParams(model, testCase.reasoning);
+
+			expect(params.chat_template_kwargs).toEqual({ thinking: testCase.expected });
+			expect(params.thinking).toBeUndefined();
+			expect(params.reasoning_effort).toBeUndefined();
+		}
+	});
+
+	it("uses qwen chat template thinking kwargs", async () => {
+		const model = {
+			...localOpenAICompletionsModel,
+			id: "Qwen/Qwen3-Coder",
+			name: "Qwen3 Coder via vLLM",
+			compat: {
+				thinkingFormat: "qwen-chat-template",
+				supportsReasoningEffort: false,
+			},
+		} satisfies Model<"openai-completions">;
+
+		for (const testCase of [
+			{ reasoning: "high" as const, expected: true },
+			{ reasoning: undefined, expected: false },
+		]) {
+			const params = await captureSimpleParams(model, testCase.reasoning);
+
+			expect(params.chat_template_kwargs).toEqual({
+				enable_thinking: testCase.expected,
+				preserve_thinking: true,
+			});
+			expect(params.reasoning_effort).toBeUndefined();
+		}
+	});
+
+	it("uses configurable chat template effort kwargs with static kwargs", async () => {
+		const model = {
+			...localOpenAICompletionsModel,
+			id: "unsloth/gpt-oss-120b-GGUF",
+			name: "GPT OSS via vLLM",
+			thinkingLevelMap: { xhigh: "max" },
+			compat: {
+				thinkingFormat: "chat-template",
+				supportsReasoningEffort: false,
+				chatTemplateKwargs: {
+					preserve_thinking: true,
+					reasoning_effort: { $var: "thinking.effort", omitWhenOff: true },
+				},
+			},
+		} satisfies Model<"openai-completions">;
+
+		const params = await captureSimpleParams(model, "xhigh");
+
+		expect(params.chat_template_kwargs).toEqual({ preserve_thinking: true, reasoning_effort: "max" });
+		expect(params.reasoning_effort).toBeUndefined();
+	});
+
 	it("uses Ant Ling compatibility metadata", async () => {
 		const model = getModel("ant-ling", "Ring-2.6-1T")!;
 		let payload: unknown;

packages/ai/test/openai-completions-tool-result-images.test.ts

@@ -32,6 +32,7 @@ const compat: Required<OpenAICompletionsCompat> = {
 	thinkingFormat: "openai",
 	openRouterRouting: {},
 	vercelGatewayRouting: {},
+	chatTemplateKwargs: {},
 	zaiToolStream: false,
 	supportsStrictMode: true,
 	cacheControlFormat: "anthropic",

packages/ai/test/overflow.test.ts

@@ -49,6 +49,13 @@ describe("isContextOverflow", () => {
 		expect(isContextOverflow(message, 131072)).toBe(true);
 	});
 
+	it("detects OpenAI-compatible parenthesized maximum context length errors", () => {
+		const message = createErrorMessage(
+			"Error: 400 Input length (265330) exceeds model's maximum context length (262144).",
+		);
+		expect(isContextOverflow(message, 262144)).toBe(true);
+	});
+
 	it("detects OpenRouter Poolside maximum allowed input length errors", () => {
 		const message = createErrorMessage(
 			"Provider returned error: Input length 131393 exceeds the maximum allowed input length of 131040 tokens.",

packages/ai/test/supports-xhigh.test.ts

@@ -69,6 +69,15 @@ describe("getSupportedThinkingLevels", () => {
 		expect(getSupportedThinkingLevels(model!)).toEqual(["off", "high"]);
 	});
 
+	it("excludes thinking off for Moonshot Kimi K2.7 Code models", () => {
+		const cases = [getModel("moonshotai", "kimi-k2.7-code"), getModel("moonshotai-cn", "kimi-k2.7-code")];
+
+		for (const model of cases) {
+			expect(model).toBeDefined();
+			expect(getSupportedThinkingLevels(model!)).toEqual(["minimal", "low", "medium", "high"]);
+		}
+	});
+
 	it("includes only high for OpenCode Grok Build", () => {
 		const model = getModel("opencode", "grok-build-0.1");
 		expect(model).toBeDefined();

packages/coding-agent/CHANGELOG.md

@@ -10,6 +10,205 @@
 
 - Added an experimental first-time setup flow behind `PI_EXPERIMENTAL=1` that asks for a dark/light theme choice (preselecting the detected appearance) and opt-in analytics data sharing on first launch with the default agent directory; opting in stores a `trackingId` in `settings.json`.
 
+## [0.79.10] - 2026-06-22
+
+### New Features
+
+- **Extension compaction event context** - Extension `session_before_compact` and `session_compact` events now include `reason` and `willRetry`, so extensions can distinguish manual `/compact`, threshold auto-compaction, and overflow retry flows. See [session_before_compact / session_compact](docs/extensions.md#session_before_compact--session_compact) and [Custom Summarization via Extensions](docs/compaction.md#custom-summarization-via-extensions).
+- **Safer update flow** - `pi update` installs the exact checked Pi version, and update notices show the changelog URL, making upgrades more predictable. See [Install and Manage](docs/packages.md#install-and-manage).
+
+### Added
+
+- Added `reason` and `willRetry` metadata to extension `session_before_compact` and `session_compact` events so extensions can distinguish manual, threshold, and overflow compaction flows ([#5962](https://github.com/earendil-works/pi/pull/5962) by [@PizzaMarinara](https://github.com/PizzaMarinara)).
+
+### Fixed
+
+- Fixed the `find` tool to respect nested git repository boundaries when parent `.gitignore` rules ignore the nested repo ([#5960](https://github.com/earendil-works/pi/issues/5960)).
+- Fixed the usage docs slash command table to include `/trust` and `/import` ([#5959](https://github.com/earendil-works/pi/issues/5959)).
+- Fixed inherited OpenAI-compatible streaming to preserve encrypted `reasoning_details` that arrive before matching tool call deltas ([#5114](https://github.com/earendil-works/pi/issues/5114)).
+- Fixed broken TUI documentation links to the plan-mode extension example ([#5957](https://github.com/earendil-works/pi/issues/5957)).
+- Fixed transient extension UI and session-start messages emitted during session replacement or reload so they remain visible, and kept reload input blocked until reload completes ([#5943](https://github.com/earendil-works/pi/issues/5943)).
+- Fixed the plan-mode example to preserve active custom tools, skip the action prompt when no plan is found, and queue refinement/execution follow-ups correctly from `agent_end` ([#5940](https://github.com/earendil-works/pi/issues/5940)).
+- Fixed `pi update` to install the exact version returned by the Pi update check, make `--force` reinstall that checked version, fail instead of falling back to an unversioned reinstall when no version is available, and report both the old and updated versions.
+- Fixed update notifications to display the actual changelog URL as the hyperlink text.
+
+## [0.79.9] - 2026-06-20
+
+### New Features
+
+- **Chat-template thinking compatibility** - OpenAI-compatible custom providers can map Pi thinking levels into `chat_template_kwargs`, enabling vLLM/Hugging Face chat-template models such as DeepSeek to use provider-native thinking controls. See [Custom Provider API Types](docs/custom-provider.md#api-types) and [OpenAI Compatibility](docs/models.md#openai-compatibility).
+- **GLM-5.2 provider improvements** - GLM-5.2 now has corrected Fireworks OpenAI-compatible routing and OpenRouter `xhigh` thinking support, improving `/model` behavior and high-effort reasoning for GLM-5.2 users. See [Model Options](docs/usage.md#model-options).
+
+### Added
+
+- Added inherited configurable `chat-template` thinking support for OpenAI-compatible providers that use `chat_template_kwargs`, such as DeepSeek models behind vLLM ([#5673](https://github.com/earendil-works/pi/issues/5673)).
+
+### Fixed
+
+- Fixed inherited Fireworks GLM-5.2 metadata to use the OpenAI-compatible Chat Completions endpoint with `reasoning_effort` support ([#5923](https://github.com/earendil-works/pi/issues/5923)).
+- Fixed same-directory session switches to reuse imported extension modules while preserving fresh extension instances and lifecycle events ([#5905](https://github.com/earendil-works/pi/issues/5905)).
+- Fixed deep session branches taking quadratic time to build context or branch paths ([#5909](https://github.com/earendil-works/pi/issues/5909)).
+- Fixed inherited OpenRouter GLM-5.2 metadata to expose `xhigh` reasoning and send OpenRouter's native `xhigh` effort ([#5770](https://github.com/earendil-works/pi/issues/5770)).
+- Fixed inherited Markdown streaming code fence rendering so partial closing fences no longer make code blocks shrink or flicker while content streams ([#5846](https://github.com/earendil-works/pi/pull/5846) by [@xl0](https://github.com/xl0)).
+- Fixed fuzzy `edit` matches to preserve untouched line blocks instead of rewriting the whole file through normalized content ([#5899](https://github.com/earendil-works/pi/issues/5899)).
+- Fixed bash commands through legacy WSL `bash.exe` to pass scripts over stdin so shell variables expand in the target bash ([#5893](https://github.com/earendil-works/pi/issues/5893)).
+- Fixed `/model` to hide GitHub Copilot models that are unavailable to the authenticated account ([#5897](https://github.com/earendil-works/pi/issues/5897)).
+- Fixed `/model` selector search to rank exact provider-prefixed matches before proxy-provider model ID matches ([#5892](https://github.com/earendil-works/pi/issues/5892)).
+
+## [0.79.8] - 2026-06-19
+
+### New Features
+
+- **Selective provider base entry points** - SDK users can pair `@earendil-works/pi-ai/base` and `@earendil-works/pi-agent-core/base` with explicit provider registration to keep bundled applications from including unused provider transports. See [`pi-ai` Base Entry Point](../ai/README.md#base-entry-point) and [`pi-agent-core` Base Entry Point](../agent/README.md#base-entry-point).
+- **Mistral prompt caching** - Mistral sessions now use provider-side prompt caching with session affinity and cached-token usage/cost accounting. See [API Keys](docs/providers.md#api-keys) and [Environment Variables](docs/usage.md#environment-variables).
+- **Post-compaction token estimates** - Compact results and compaction events now include estimated post-compaction token counts so clients can show the approximate context reduction. See [RPC compact](docs/rpc.md#compact) and [compaction events](docs/rpc.md#compaction_start--compaction_end).
+- **OpenRouter Fusion alias** - `openrouter/fusion` is available as a built-in OpenRouter model alias. See [API Keys](docs/providers.md#api-keys).
+
+### Added
+
+- Added inherited `@earendil-works/pi-ai/base` and `@earendil-works/pi-agent-core/base` entry points for selective provider registration in bundled applications ([#5348](https://github.com/earendil-works/pi/pull/5348) by [@FredKSchott](https://github.com/FredKSchott)).
+- Added inherited Mistral prompt caching using the pi session ID as `prompt_cache_key`, including cached-token usage and cost accounting ([#5854](https://github.com/earendil-works/pi/issues/5854)).
+- Added estimated post-compaction token counts to compact results and compaction events ([#5877](https://github.com/earendil-works/pi/issues/5877)).
+- Added the inherited OpenRouter Fusion alias as `openrouter/fusion` ([#5866](https://github.com/earendil-works/pi/pull/5866) by [@dannote](https://github.com/dannote)).
+
+### Fixed
+
+- Updated vulnerable runtime dependencies, including `undici` and the packaged `protobufjs` transitive dependency.
+- Fixed compaction to refuse sessions with no eligible messages instead of producing empty summaries ([#4811](https://github.com/earendil-works/pi/issues/4811)).
+- Fixed successful overflow-triggered auto-compaction to avoid retrying completed assistant responses ([#5720](https://github.com/earendil-works/pi/issues/5720)).
+
+## [0.79.7] - 2026-06-18
+
+### New Features
+
+- **Automatic theme mode** - `/settings` can choose separate light and dark themes and follow terminal color-scheme changes. See [Selecting a Theme](docs/themes.md#selecting-a-theme).
+- **Self-only updates by default** - `pi update` now updates pi only, with `pi update --all` for updating pi and packages together. See [Install and Manage](docs/packages.md#install-and-manage).
+- **Extension API helpers** - extensions can use `CONFIG_DIR_NAME` for project config paths and import edit diff helpers for edit-style diffs. See [`ctx.cwd`](docs/extensions.md#ctxcwd) and [SDK Exports](docs/sdk.md#exports).
+- **Warp inline images** - Warp terminals now get inline image rendering through Kitty graphics detection. See [Image](docs/tui.md#image).
+
+### Added
+
+- Added automatic theme mode so `/settings` can use separate light and dark themes and follow terminal color-scheme changes ([#5874](https://github.com/earendil-works/pi/pull/5874)).
+- Added inherited Warp terminal image capability detection so inline images render through Warp's Kitty graphics support ([#5841](https://github.com/earendil-works/pi/pull/5841) by [@dodiego](https://github.com/dodiego)).
+- Exported `CONFIG_DIR_NAME` from the coding-agent public API so extensions can resolve project config paths without hardcoding `.pi` ([#5869](https://github.com/earendil-works/pi/pull/5869) by [@xl0](https://github.com/xl0)).
+- Exported edit diff helpers (`generateDiffString`, `generateUnifiedPatch`, and `EditDiffResult`) from the public API for extensions that need edit-style diffs ([#5756](https://github.com/earendil-works/pi/pull/5756) by [@xl0](https://github.com/xl0)).
+
+### Changed
+
+- Changed bare `pi update` to update only pi, added `pi update --all` for updating pi and extensions together, and clarified extension update prompts.
+- Reserved `/` in theme names for automatic light/dark theme settings.
+- Updated extension docs, examples, runtime help, trust prompts, and config labels to use the configured project config directory instead of hardcoded `.pi` paths.
+
+### Fixed
+
+- Fixed RPC unknown-command errors to include the request id so clients do not hang waiting for a response ([#5868](https://github.com/earendil-works/pi/issues/5868)).
+- Fixed `/model` autocomplete and model selection searches to match provider/model queries regardless of whether the provider or model token is typed first.
+- Fixed the tree navigator to horizontally pan deep entries so the selected item remains readable ([#5830](https://github.com/earendil-works/pi/issues/5830)).
+
+## [0.79.6] - 2026-06-16
+
+### Fixed
+
+- Fixed HTTP dispatcher configuration to preserve a caller's deliberate `fetch` override instead of reinstalling the undici global fetch over it.
+- Fixed inherited OpenCode Go DeepSeek V4 thinking-off requests to send the provider's `thinking: { type: "disabled" }` compatibility parameter.
+
+## [0.79.5] - 2026-06-16
+
+### New Features
+
+- **Provider-scoped API key environments** - `auth.json` API key entries can now include `env` overrides for provider-specific Cloudflare, Azure OpenAI, Google Vertex, Amazon Bedrock, cache retention, and proxy settings without changing the project shell. See [Auth File](docs/providers.md#auth-file).
+- **Global HTTP proxy setting** - Configure `httpProxy` once in global settings to apply `HTTP_PROXY` and `HTTPS_PROXY` to Pi-managed HTTP clients. See [Network](docs/settings.md#network).
+- **Vercel AI Gateway attribution** - Vercel AI Gateway requests now include Pi attribution headers by default. See [API Keys](docs/providers.md#api-keys).
+
+### Added
+
+- Added Vercel AI Gateway request attribution headers (`http-referer` and `x-title`) for Vercel AI Gateway models ([#5798](https://github.com/earendil-works/pi/pull/5798) by [@rwachtler](https://github.com/rwachtler)).
+- Added an `xp` footer marker when experimental features are enabled.
+- Added a global `httpProxy` setting that applies as `HTTP_PROXY` and `HTTPS_PROXY` for Pi-managed HTTP clients ([#5790](https://github.com/earendil-works/pi/issues/5790)).
+- Added `auth.json` API key `env` values so provider-specific environment overrides can be scoped to Pi and propagated to inherited provider configuration ([#5728](https://github.com/earendil-works/pi/issues/5728)).
+
+### Changed
+
+- Updated the vendored Markdown parser used by HTML session exports to `marked` 18.0.5.
+
+### Fixed
+
+- Fixed inherited OpenAI Responses streaming to tolerate null message content from OpenAI-compatible servers before tool calls ([#5819](https://github.com/earendil-works/pi/issues/5819)).
+- Fixed inherited OpenCode DeepSeek V4 thinking requests to avoid sending both `thinking` and `reasoning_effort` ([#5818](https://github.com/earendil-works/pi/issues/5818)).
+- Fixed device-code login to stop opening the browser automatically.
+- Fixed inherited editor Cursor Up handling so non-empty drafts jump to the start of the line before browsing input history ([#5789](https://github.com/earendil-works/pi/pull/5789) by [@4h9fbZ](https://github.com/4h9fbZ)).
+- Fixed inherited Z.AI GLM-5.2 thinking requests to send `reasoning_effort` with the provider's `high`/`max` effort mapping ([#5770](https://github.com/earendil-works/pi/issues/5770)).
+- Fixed successful `pi update` on Windows to exit naturally instead of calling `process.exit(0)`, avoiding a Node.js/libuv assertion after version-check network requests ([#5805](https://github.com/earendil-works/pi/issues/5805)).
+- Fixed inherited Google and `google-vertex` Gemini model metadata to map `latest` aliases to the current models, add Gemini 3.5 Flash for Vertex, correct Gemini 2.5 Flash Vertex cache pricing, and remove shut-down Vertex preview models ([#5761](https://github.com/earendil-works/pi/issues/5761)).
+- Fixed the session selector to stay open and show the all-sessions empty state when both current-folder and all-scope session lists are empty ([#5747](https://github.com/earendil-works/pi/issues/5747)).
+- Fixed inherited Moonshot AI China model metadata to include Kimi K2.7 Code, and omitted unsupported thinking-off payloads for Kimi K2.7 Code models ([#5760](https://github.com/earendil-works/pi/issues/5760)).
+
+## [0.79.4] - 2026-06-15
+
+### New Features
+
+- **Automatic first-run theme selection** - pi detects the terminal background on first run and defaults to the `dark` or `light` theme. See [Selecting a Theme](docs/themes.md#selecting-a-theme).
+- **Standalone binary integrity checksums** - GitHub release assets now include `SHA256SUMS` files for verifying standalone binary downloads. See [Quickstart Install](docs/quickstart.md#install).
+
+### Added
+
+- Added `SHA256SUMS` integrity files to standalone binary GitHub release assets ([#5739](https://github.com/earendil-works/pi/issues/5739)).
+- Added first-run interactive theme detection from the terminal background ([#5385](https://github.com/earendil-works/pi/pull/5385) by [@vegarsti](https://github.com/vegarsti)).
+
+### Fixed
+
+- Fixed bash tool output collection to keep draining stdout/stderr after the child exits while descendants still write, avoiding truncated late output ([#5753](https://github.com/earendil-works/pi/pull/5753) by [@Mearman](https://github.com/Mearman)).
+- Fixed `/tree` help rendering to show compact wrapped controls instead of truncating them on narrow terminals ([#5055](https://github.com/earendil-works/pi/issues/5055)).
+- Fixed SIGTERM/SIGHUP interactive shutdown to keep signal handlers installed until terminal cleanup completes, preventing `signal-exit` from re-sending the signal and leaving the terminal in raw/Kitty keyboard mode ([#5724](https://github.com/earendil-works/pi/issues/5724)).
+- Fixed extensions documentation to clarify that `pi.getActiveTools()` returns active tool names while `pi.getAllTools()` returns tool metadata ([#5729](https://github.com/earendil-works/pi/issues/5729)).
+- Fixed question and questionnaire extension examples to wrap long prompt, option, and help text instead of truncating it ([#5708](https://github.com/earendil-works/pi/pull/5708) by [@xl0](https://github.com/xl0)).
+- Fixed package commands such as `pi list`, `pi install`, and `pi update` to terminate after completing even if an extension leaves background handles open ([#5687](https://github.com/earendil-works/pi/issues/5687)).
+- Fixed `pi update` for pnpm global installs whose configured `global-bin-dir` no longer matches the active pnpm home ([#5689](https://github.com/earendil-works/pi/issues/5689)).
+- Fixed npm package specs that use ranges or tags (for example `@^1.2.7`) so installed package resources still load instead of being treated as mismatched exact pins ([#5695](https://github.com/earendil-works/pi/issues/5695)).
+- Fixed inherited Anthropic 1-hour prompt-cache write cost accounting to price 1-hour cache writes at 2x input instead of the 5-minute cache-write rate ([#5738](https://github.com/earendil-works/pi/pull/5738) by [@theBucky](https://github.com/theBucky)).
+- Fixed inherited GitHub Copilot Claude adaptive-thinking effort metadata to match manually checked Copilot model capabilities ([#4637](https://github.com/earendil-works/pi/issues/4637)).
+- Fixed inherited OpenCode/OpenCode Go completion model metadata to omit long-retention cache fields for routes that reject `prompt_cache_retention` ([#5702](https://github.com/earendil-works/pi/issues/5702)).
+- Fixed inherited overlay compositing over CJK wide characters so borders stay aligned when an overlay starts inside a full-width cell ([#5297](https://github.com/earendil-works/pi/issues/5297)).
+- Fixed inherited WezTerm inline Kitty image rendering during full redraw fallbacks so image padding rows are reserved before the placement is drawn without regressing tall-image placement ([#5618](https://github.com/earendil-works/pi/issues/5618), [#4415](https://github.com/earendil-works/pi/issues/4415)).
+- Fixed custom provider config so plain uppercase API key and header values remain literals instead of being treated as legacy environment references; use explicit `$ENV_VAR` syntax for environment variables ([#5661](https://github.com/earendil-works/pi/issues/5661)).
+
+## [0.79.3] - 2026-06-13
+
+### Fixed
+
+- Fixed inherited OpenAI GPT-5.4/GPT-5.5 and OpenAI Codex GPT-5.4/GPT-5.4 mini/GPT-5.5 context window metadata to use the observed 272k-token Codex backend limit, avoiding a billing hazard from prompts above Codex's accepted limit (reported by [@trethore](https://github.com/trethore)).
+
+## [0.79.2] - 2026-06-12
+
+### New Features
+
+- **Clearer Bedrock validation guidance** - Amazon Bedrock data retention validation errors now link to AWS data retention documentation. See [Amazon Bedrock](docs/providers.md#amazon-bedrock).
+
+### Added
+
+- Added an experimental first-time setup flow behind `PI_EXPERIMENTAL=1` that asks for a dark/light theme choice (preselecting the detected appearance) and opt-in analytics data sharing on first launch with the default agent directory; opting in stores a `trackingId` in `settings.json` ([#5587](https://github.com/earendil-works/pi/pull/5587) by [@vegarsti](https://github.com/vegarsti)).
+- Added AWS data retention documentation links to inherited Amazon Bedrock unsupported data retention mode validation errors ([#5561](https://github.com/earendil-works/pi/pull/5561) by [@unexge](https://github.com/unexge)).
+
+### Fixed
+
+- Fixed project trust detection to ignore global `~/.pi/agent` state when running from `$HOME`, and made `pi update` use only saved or explicit project trust without prompting ([#5619](https://github.com/earendil-works/pi/issues/5619)).
+- Fixed experimental first-time setup to skip forked sessions instead of rerunning the setup prompts ([#5627](https://github.com/earendil-works/pi/pull/5627) by [@vegarsti](https://github.com/vegarsti)).
+- Fixed inherited OpenAI-compatible context overflow detection for parenthesized `maximum context length (N)` errors ([#5677](https://github.com/earendil-works/pi/issues/5677)).
+- Fixed inherited OpenAI GPT-5.4/GPT-5.5 and OpenAI Codex GPT-5.4/GPT-5.4 mini/GPT-5.5 context window metadata to match current OpenAI limits ([#5644](https://github.com/earendil-works/pi/issues/5644)).
+- Fixed inherited Anthropic refusal stops to preserve provider `stop_details` explanations in error messages ([#5666](https://github.com/earendil-works/pi/pull/5666) by [@rwachtler](https://github.com/rwachtler)).
+- Increased the inherited OpenAI Codex Responses SSE response-header timeout to 20 seconds to reduce false-positive stalls while retaining the bounded wait introduced for zero-event hangs ([#4945](https://github.com/earendil-works/pi/issues/4945)).
+- Fixed inherited Claude Fable 5 thinking-off requests to omit Anthropic's unsupported `thinking.type: "disabled"` payload ([#5567](https://github.com/earendil-works/pi/pull/5567) by [@tmustier](https://github.com/tmustier)).
+- Fixed inherited late tool progress callbacks after tool settlement to be ignored instead of emitting stale `tool_execution_update` events ([#5573](https://github.com/earendil-works/pi/issues/5573)).
+- Fixed inherited user-message transcript rendering so standalone `+` messages no longer render as `-` ([#5657](https://github.com/earendil-works/pi/issues/5657)).
+- Fixed inherited slash-separated fuzzy queries so provider/model completions remain matchable after insertion.
+- Fixed inherited WezTerm inline Kitty image rendering so reserved row clears do not erase all but the top strip of tool image previews ([#5618](https://github.com/earendil-works/pi/issues/5618)).
+- Fixed inherited editor wrapping for CJK text to break at character boundaries instead of leaving large trailing gaps ([#5585](https://github.com/earendil-works/pi/pull/5585) by [@haoqixu](https://github.com/haoqixu)).
+- Fixed inherited loose Markdown list rendering to preserve blank-line separation between list items ([#5562](https://github.com/earendil-works/pi/pull/5562) by [@Perlence](https://github.com/Perlence)).
+- Fixed `--model` resolution for authenticated custom model IDs whose slash prefix matches an unauthenticated built-in provider ([#5643](https://github.com/earendil-works/pi/issues/5643)).
+- Fixed `/fork` to keep session parent chains connected when the forked path contains labels ([#5669](https://github.com/earendil-works/pi/issues/5669)).
+- Fixed `/share` and `/export` HTML exports to use the active fallback theme when the configured custom theme no longer exists ([#5596](https://github.com/earendil-works/pi/issues/5596)).
+- Fixed custom fallback model IDs with `:<thinking>` suffixes to preserve the requested thinking level when the provider template model does not advertise reasoning ([#5560](https://github.com/earendil-works/pi/pull/5560) by [@haoqixu](https://github.com/haoqixu)).
+
 ## [0.79.1] - 2026-06-09
 
 ### New Features

packages/coding-agent/README.md

@@ -7,11 +7,6 @@
   <a href="https://discord.com/invite/3cU7Bz4UPx"><img alt="Discord" src="https://img.shields.io/badge/discord-community-5865F2?style=flat-square&logo=discord&logoColor=white" /></a>
   <a href="https://www.npmjs.com/package/@earendil-works/pi-coding-agent"><img alt="npm" src="https://img.shields.io/npm/v/@earendil-works/pi-coding-agent?style=flat-square" /></a>
 </p>
-<p align="center">
-  <a href="https://pi.dev">pi.dev</a> domain graciously donated by
-  <br /><br />
-  <a href="https://exe.dev"><img src="docs/images/exy.png" alt="Exy mascot" width="48" /><br />exe.dev</a>
-</p>
 
 > New issues and PRs from new contributors are auto-closed by default. Maintainers review auto-closed issues daily. See [CONTRIBUTING.md](../../CONTRIBUTING.md).
 
@@ -191,7 +186,8 @@ Type `/` in the editor to trigger commands. [Extensions](#extensions) can regist
 | `/clone` | Duplicate the current active branch into a new session |
 | `/compact [prompt]` | Manually compact context, optional custom instructions |
 | `/copy` | Copy last assistant message to clipboard |
-| `/export [file]` | Export session to HTML file |
+| `/export [file]` | Export session to HTML or JSONL file |
+| `/import <file>` | Import and resume a session from a JSONL file |
 | `/share` | Upload as private GitHub gist with shareable HTML link |
 | `/reload` | Reload keybindings, extensions, skills, prompts, and context files (themes hot-reload automatically) |
 | `/hotkeys` | Show all keyboard shortcuts |
@@ -291,15 +287,15 @@ See [docs/settings.md](docs/settings.md) for all options.
 
 ### Project Trust
 
-On interactive startup, pi asks before trusting a project folder that contains project-local extensions or settings and has no saved decision for the folder or a parent folder in `~/.pi/agent/trust.json`. Trusting a project allows pi to load `.pi/settings.json` and `.pi` resources, install missing project packages, and execute project extensions.
+On interactive startup, pi asks before trusting a project folder that contains project-local settings, resources, or project `.agents/skills` and has no saved decision for the folder or a parent folder in `~/.pi/agent/trust.json`. Trusting a project allows pi to load `.pi/settings.json` and `.pi` resources, install missing project packages, and execute project extensions.
 
 Before the trust decision, pi loads only context files, user/global extensions, and CLI `-e` extensions so they can handle the `project_trust` event. Project-local extensions, project package-managed extensions, and project settings are loaded only after the project is trusted. This split also applies when switching to a session from a different cwd whose trust has not been resolved in the current process.
 
-Non-interactive modes (`-p`, `--mode json`, and `--mode rpc`) do not show a trust prompt. Without an applicable saved trust decision, they use `defaultProjectTrust` from global settings: `ask` (default) and `never` ignore trust-gated project inputs, while `always` trusts them. Pass `--approve`/`-a` or `--no-approve`/`-na` to override project trust for one run.
+Non-interactive modes (`-p`, `--mode json`, and `--mode rpc`) do not show a trust prompt. Without an applicable saved trust decision, they use `defaultProjectTrust` from global settings: `ask` (default) and `never` ignore those project resources, while `always` trusts them. Pass `--approve`/`-a` or `--no-approve`/`-na` to override project trust for one run.
 
 If no extension or saved decision applies, `defaultProjectTrust` controls the fallback behavior. Set it to `"ask"`, `"always"`, or `"never"` in `~/.pi/agent/settings.json`, or change it with `/settings`.
 
-`pi config` and package commands use the same project trust flow. Pass `--approve` to trust project-local settings for one command or `--no-approve` to ignore them.
+`pi config` and package commands use the same project trust flow, except `pi update` never prompts. Pass `--approve` to trust project-local settings for one command or `--no-approve` to ignore them.
 
 Use `/trust` in interactive mode to save a project trust decision for future sessions, including trust for the immediate parent folder. It writes `~/.pi/agent/trust.json` only; the current session is not reloaded, so restart pi for changes to take effect.
 
@@ -419,7 +415,8 @@ pi install ssh://git@github.com/user/repo@v1    # tag or commit
 pi remove npm:@foo/pi-tools
 pi uninstall npm:@foo/pi-tools          # alias for remove
 pi list
-pi update                               # update pi and packages (skips pinned packages)
+pi update                               # update pi only
+pi update --all                         # update pi and packages
 pi update --extensions                  # update packages only
 pi update --self                        # update pi only
 pi update --self --force                # reinstall pi even if current
@@ -427,7 +424,7 @@ pi update npm:@foo/pi-tools             # update one package
 pi config                               # enable/disable extensions, skills, prompts, themes
 ```
 
-Packages install to `~/.pi/agent/git/` (git) or `~/.pi/agent/npm/` (npm). Use `-l` for project-local installs (`.pi/git/`, `.pi/npm/`). Git `@ref` values are pinned tags or commits; pinned packages are skipped by `pi update`, so use `pi install git:host/user/repo@new-ref` to move an existing package to a new ref. Git packages install dependencies with `npm install --omit=dev` by default, so runtime deps must be listed under `dependencies`; when `npmCommand` is configured, git packages use plain `install` for compatibility with wrappers. If you use a Node version manager and want package installs to reuse a stable npm context, set `npmCommand` in `settings.json`, for example `["mise", "exec", "node@20", "--", "npm"]`.
+Packages install to `~/.pi/agent/git/` (git) or `~/.pi/agent/npm/` (npm). Use `-l` for project-local installs (`.pi/git/`, `.pi/npm/`). Git `@ref` values are pinned tags or commits; pinned packages are skipped by `pi update --extensions` and `pi update --all`, so use `pi install git:host/user/repo@new-ref` to move an existing package to a new ref. Git packages install dependencies with `npm install --omit=dev` by default, so runtime deps must be listed under `dependencies`; when `npmCommand` is configured, git packages use plain `install` for compatibility with wrappers. If you use a Node version manager and want package installs to reuse a stable npm context, set `npmCommand` in `settings.json`, for example `["mise", "exec", "node@20", "--", "npm"]`.
 
 Create a package by adding a `pi` key to `package.json`:
 
@@ -518,7 +515,8 @@ pi [options] [@files...] [messages...]
 pi install <source> [-l]     # Install package, -l for project-local
 pi remove <source> [-l]      # Remove package
 pi uninstall <source> [-l]   # Alias for remove
-pi update [source|self|pi]   # Update pi and packages (skips pinned packages)
+pi update [source|self|pi]   # Update pi only, or one package source
+pi update --all              # Update pi and packages
 pi update --extensions       # Update packages only
 pi update --self             # Update pi only
 pi update --self --force     # Reinstall pi even if current
@@ -527,7 +525,7 @@ pi list                      # List installed packages
 pi config                    # Enable/disable package resources
 ```
 
-`pi config` and project package commands accept `--approve`/`--no-approve` to trust or ignore project-local settings for one command.
+`pi config` and project package commands accept `--approve`/`--no-approve` to trust or ignore project-local settings for one command. `pi update` never prompts for project trust.
 
 ### Modes
 
@@ -673,8 +671,6 @@ pi --thinking high "Solve this complex problem"
 
 See [CONTRIBUTING.md](../../CONTRIBUTING.md) for guidelines and [docs/development.md](docs/development.md) for setup, forking, and debugging.
 
----
-
 ## License
 
 MIT
@@ -684,3 +680,9 @@ MIT
 - [@earendil-works/pi-ai](https://www.npmjs.com/package/@earendil-works/pi-ai): Core LLM toolkit
 - [@earendil-works/pi-agent-core](https://www.npmjs.com/package/@earendil-works/pi-agent-core): Agent framework
 - [@earendil-works/pi-tui](https://www.npmjs.com/package/@earendil-works/pi-tui): Terminal UI components
+
+<p align="center">
+  <a href="https://pi.dev">pi.dev</a> domain graciously donated by
+  <br /><br />
+  <a href="https://exe.dev"><img src="docs/images/exy.png" alt="Exy mascot" width="48" /><br />exe.dev</a>
+</p>

packages/coding-agent/docs/compaction.md

@@ -276,7 +276,7 @@ Fired before auto-compaction or `/compact`. Can cancel or provide custom summary
 
 ```typescript
 pi.on("session_before_compact", async (event, ctx) => {
-  const { preparation, branchEntries, customInstructions, signal } = event;
+  const { preparation, branchEntries, customInstructions, reason, willRetry, signal } = event;
 
   // preparation.messagesToSummarize - messages to summarize
   // preparation.turnPrefixMessages - split turn prefix (if isSplitTurn)
@@ -287,6 +287,8 @@ pi.on("session_before_compact", async (event, ctx) => {
   // preparation.settings - compaction settings
 
   // branchEntries - all entries on current branch (for custom state)
+  // reason - "manual" (/compact), "threshold", or "overflow"
+  // willRetry - whether the aborted turn is retried after compaction (overflow recovery)
   // signal - AbortSignal (pass to LLM calls)
 
   // Cancel:

packages/coding-agent/docs/containerization.md

@@ -10,46 +10,12 @@ There are two general options. You can either
 
 | Pattern | What is isolated | Best for | Notes |
 | --- | --- | --- | --- |
-| OpenShell | Whole `pi` process in a policy-controlled sandbox | Local or remote managed sandbox | Requires an OpenShell gateway |
 | Gondolin extension | Built-in tools and `!` commands | Local micro-VM isolation while keeping auth on host | See [`examples/extensions/gondolin/`](../examples/extensions/gondolin/). |
 | Plain Docker | Whole `pi` process in a local container | Simple local isolation | Provider API keys enter the container. |
+| OpenShell | Whole `pi` process in a policy-controlled sandbox | Local or remote managed sandbox | Requires an OpenShell gateway |
 
 Extensions run wherever the `pi` process runs. If you run host `pi` with a tool-routing extension, other custom extension tools still run on the host unless they also delegate their operations.
 
-## OpenShell
-
-Use [NVIDIA OpenShell](https://docs.nvidia.com/openshell/about/overview) when you want a policy-controlled sandbox with filesystem, process, network, credential, and inference controls.
-OpenShell can run sandboxes through a local gateway backed by Docker, Podman, or a VM runtime, or through a remote Kubernetes gateway.
-
-Every sandbox requires an active gateway.
-Register and select one before creating a sandbox:
-
-```bash
-openshell gateway add <gateway-url> --name <name>
-openshell gateway select <name>
-```
-
-Launch `pi` inside an OpenShell sandbox:
-
-```bash
-openshell sandbox create --name pi-sandbox --from pi -- pi
-```
-
-In this pattern, the whole `pi` process runs inside the sandbox.
-Built-in tools, `!` commands, and extension tools execute inside the OpenShell boundary.
-
-If the gateway is remote, project files are not bind-mounted from the host, meaning writes in the sandbox are not reflected on your machine.
-Clone the repository inside the sandbox or use OpenShell file transfer commands:
-
-```bash
-openshell sandbox upload pi-sandbox ./repo /workspace
-openshell sandbox download pi-sandbox /workspace/repo ./repo-out
-```
-
-OpenShell providers can keep raw model API keys outside the sandbox.
-When inference routing is configured, code inside the sandbox can call `https://inference.local`, and the gateway injects the configured provider credentials upstream.
-Configure Pi to use the corresponding OpenAI-compatible or Anthropic-compatible endpoint if you want model traffic to use this route.
-
 ## Gondolin
 
 [Gondolin](https://github.com/earendil-works/gondolin) is a local Linux micro-VM.
@@ -109,3 +75,37 @@ docker run --rm -it \
 The `-v "$PWD:/workspace"` mounts your current directory into the container at /workspace such that reads and writes in `/workspace` inside Docker directly affect your host files, like in the Gondolin example.
 
 Use a named volume for `/root/.pi/agent` if you want container-local settings and sessions. Mounting your host `~/.pi/agent` exposes host auth and session files to the container.
+
+## OpenShell
+
+Use [NVIDIA OpenShell](https://docs.nvidia.com/openshell/about/overview) when you want a policy-controlled sandbox with filesystem, process, network, credential, and inference controls.
+OpenShell can run sandboxes through a local gateway backed by Docker, Podman, or a VM runtime, or through a remote Kubernetes gateway.
+
+Every sandbox requires an active gateway.
+Register and select one before creating a sandbox:
+
+```bash
+openshell gateway add <gateway-url> --name <name>
+openshell gateway select <name>
+```
+
+Launch `pi` inside an OpenShell sandbox:
+
+```bash
+openshell sandbox create --name pi-sandbox --from pi -- pi
+```
+
+In this pattern, the whole `pi` process runs inside the sandbox.
+Built-in tools, `!` commands, and extension tools execute inside the OpenShell boundary.
+
+If the gateway is remote, project files are not bind-mounted from the host, meaning writes in the sandbox are not reflected on your machine.
+Clone the repository inside the sandbox or use OpenShell file transfer commands:
+
+```bash
+openshell sandbox upload pi-sandbox ./repo /workspace
+openshell sandbox download pi-sandbox /workspace/repo ./repo-out
+```
+
+OpenShell providers can keep raw model API keys outside the sandbox.
+When inference routing is configured, code inside the sandbox can call `https://inference.local`, and the gateway injects the configured provider credentials upstream.
+Configure Pi to use the corresponding OpenAI-compatible or Anthropic-compatible endpoint if you want model traffic to use this route.

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

@@ -229,7 +229,7 @@ models: [{
 }]
 ```
 
-Use `openrouter` for OpenRouter-style `reasoning: { effort }` controls. Use `together` for Together-style `reasoning: { enabled }` controls; with `supportsReasoningEffort`, it also sends `reasoning_effort`. Use `qwen-chat-template` instead for local Qwen-compatible servers that read `chat_template_kwargs.enable_thinking`.
+Use `openrouter` for OpenRouter-style `reasoning: { effort }` controls. Use `together` for Together-style `reasoning: { enabled }` controls; with `supportsReasoningEffort`, it also sends `reasoning_effort`. Use `qwen-chat-template` for local Qwen-compatible servers that read `chat_template_kwargs.enable_thinking` and need `preserve_thinking`.
 Use `cacheControlFormat: "anthropic"` for OpenAI-compatible providers that expose Anthropic-style prompt caching via `cache_control` on the system prompt, last tool definition, and last user/assistant text content.
 
 For Anthropic-compatible providers using `api: "anthropic-messages"`, set `compat.forceAdaptiveThinking: true` on models or providers whose upstream model requires adaptive thinking (`thinking.type: "adaptive"` plus `output_config.effort`). Built-in adaptive Claude models set this automatically. Set `compat.allowEmptySignature: true` only for providers that emit empty thinking signatures and expect `signature: ""` on replay.
@@ -718,7 +718,8 @@ interface ProviderModelConfig {
     requiresAssistantAfterToolResult?: boolean;
     requiresThinkingAsText?: boolean;
     requiresReasoningContentOnAssistantMessages?: boolean;
-    thinkingFormat?: "openai" | "openrouter" | "deepseek" | "together" | "zai" | "qwen" | "qwen-chat-template";
+    thinkingFormat?: "openai" | "openrouter" | "deepseek" | "together" | "zai" | "qwen" | "chat-template" | "qwen-chat-template" | "string-thinking" | "ant-ling";
+    chatTemplateKwargs?: Record<string, string | number | boolean | null | { "$var": "thinking.enabled" | "thinking.effort"; omitWhenOff?: boolean }>;
     cacheControlFormat?: "anthropic";
 
     // anthropic-messages
@@ -732,5 +733,5 @@ interface ProviderModelConfig {
 }
 ```
 
-`openrouter` sends `reasoning: { effort }`. `deepseek` sends `thinking: { type: "enabled" | "disabled" }` and `reasoning_effort` when enabled. `together` sends `reasoning: { enabled }` and also `reasoning_effort` when `supportsReasoningEffort` is enabled. `qwen` is for DashScope-style top-level `enable_thinking`. Use `qwen-chat-template` for local Qwen-compatible servers that read `chat_template_kwargs.enable_thinking`.
+`openrouter` sends `reasoning: { effort }`. `deepseek` sends `thinking: { type: "enabled" | "disabled" }` and `reasoning_effort` when enabled. `together` sends `reasoning: { enabled }` and also `reasoning_effort` when `supportsReasoningEffort` is enabled. `qwen` is for DashScope-style top-level `enable_thinking`. Use `qwen-chat-template` for local Qwen-compatible servers that read `chat_template_kwargs.enable_thinking` and need `preserve_thinking`. Use `chat-template` for configurable `chat_template_kwargs`, for example DeepSeek V3.x behind vLLM with `chatTemplateKwargs: { "thinking": { "$var": "thinking.enabled" } }`.
 `cacheControlFormat: "anthropic"` applies Anthropic-style `cache_control` markers to the system prompt, last tool definition, and last user/assistant text content.

packages/coding-agent/docs/extensions.md

@@ -216,6 +216,12 @@ export default async function (pi: ExtensionAPI) {
 
 This pattern makes the fetched models available during normal startup and to `pi --list-models`.
 
+### Long-lived resources and shutdown
+
+Extension factories may run in invocations that never start a session. Do not start background resources such as processes, sockets, file watchers, or timers from the factory.
+
+Defer background resource startup until `session_start` or the command/tool/event that needs the resource. Register an idempotent `session_shutdown` handler to close any session-scoped resources you start.
+
 ### Extension Styles
 
 **Single file** - simplest, for small extensions:
@@ -431,7 +437,10 @@ Fired on compaction. See [compaction.md](compaction.md) for details.
 
 ```typescript
 pi.on("session_before_compact", async (event, ctx) => {
-  const { preparation, branchEntries, customInstructions, signal } = event;
+  const { preparation, branchEntries, customInstructions, reason, willRetry, signal } = event;
+
+  // reason - "manual" (/compact), "threshold", or "overflow"
+  // willRetry - whether the aborted turn is retried after compaction (overflow recovery)
 
   // Cancel:
   return { cancel: true };
@@ -449,6 +458,8 @@ pi.on("session_before_compact", async (event, ctx) => {
 pi.on("session_compact", async (event, ctx) => {
   // event.compactionEntry - the saved compaction
   // event.fromExtension - whether extension provided it
+  // event.reason - "manual" (/compact), "threshold", or "overflow"
+  // event.willRetry - whether the aborted turn is retried after compaction (overflow recovery)
 });
 ```
 
@@ -471,7 +482,7 @@ pi.on("session_tree", async (event, ctx) => {
 
 #### session_shutdown
 
-Fired before an extension runtime is torn down.
+Fired before a started session runtime is torn down. Use this to clean up resources opened from `session_start` or other session-scoped hooks.
 
 ```typescript
 pi.on("session_shutdown", async (event, ctx) => {
@@ -892,6 +903,20 @@ Current run mode: `"tui"`, `"rpc"`, `"json"`, or `"print"`. Use `ctx.mode === "t
 
 Current working directory.
 
+Use `CONFIG_DIR_NAME` instead of hardcoding `.pi` when constructing project-local config paths. Rebranded distributions can use a different config directory name.
+
+```typescript
+import { CONFIG_DIR_NAME, type ExtensionAPI } from "@earendil-works/pi-coding-agent";
+import { join } from "node:path";
+
+export default function (pi: ExtensionAPI) {
+  pi.on("session_start", (_event, ctx) => {
+    const projectConfigPath = join(ctx.cwd, CONFIG_DIR_NAME, "my-extension.json");
+    // ...
+  });
+}
+```
+
 ### ctx.isProjectTrusted()
 
 Returns whether project-local trust is active for the current session context. This includes temporary trust decisions and CLI trust overrides, not just saved decisions in the global trust store.
@@ -1528,21 +1553,21 @@ const result = await pi.exec("git", ["status"], { signal, timeout: 5000 });
 
 ### pi.getActiveTools() / pi.getAllTools() / pi.setActiveTools(names)
 
-Manage active tools. This works for both built-in tools and dynamically registered tools.
+Manage active tools. This works for both built-in tools and dynamically registered tools. `pi.getActiveTools()` returns the active tool names as `string[]`; `pi.getAllTools()` returns metadata for all configured tools.
 
 ```typescript
-const active = pi.getActiveTools();
+const active = pi.getActiveTools(); // ["read", "bash", ...]
 const all = pi.getAllTools();
-// [{
+// all = [{
 //   name: "read",
 //   description: "Read file contents...",
 //   parameters: ...,
 //   promptGuidelines: ["Use read to examine files instead of cat or sed."],
 //   sourceInfo: { path: "<builtin:read>", source: "builtin", scope: "temporary", origin: "top-level" }
 // }, ...]
-const names = all.map(t => t.name);
 const builtinTools = all.filter((t) => t.sourceInfo.source === "builtin");
 const extensionTools = all.filter((t) => t.sourceInfo.source !== "builtin" && t.sourceInfo.source !== "sdk");
+pi.setActiveTools([...new Set([...active, "my_custom_tool"])]); // Keep current tools and enable my_custom_tool
 pi.setActiveTools(["read", "bash"]); // Switch to read-only
 ```
 

packages/coding-agent/docs/index.md

@@ -42,7 +42,7 @@ For the full first-run flow, see [Quickstart](quickstart.md).
 - [Using Pi](usage.md) - interactive mode, slash commands, context files, and CLI reference.
 - [Providers](providers.md) - subscription and API-key setup for built-in providers.
 - [Security](security.md) - project trust, sandbox boundaries, and vulnerability reporting.
-- [Containerization](containerization.md) - sandbox pi with OpenShell, Gondolin, or Docker.
+- [Containerization](containerization.md) - sandbox pi with Gondolin, Docker, or OpenShell.
 - [Settings](settings.md) - global and project settings.
 - [Keybindings](keybindings.md) - default shortcuts and custom keybindings.
 - [Sessions](sessions.md) - session management, branching, and tree navigation.

packages/coding-agent/docs/models.md

@@ -161,13 +161,11 @@ The `apiKey` and `headers` fields support command execution, environment interpo
   "apiKey": "$$literal-dollar-prefix"
   "apiKey": "$!literal-bang-prefix"
   ```
-- **Literal value:** Used directly
+- **Literal value:** Used directly. Plain uppercase strings such as `MY_API_KEY` are literals; use `$MY_API_KEY` for environment variables.
   ```json
   "apiKey": "sk-..."
   ```
 
-Legacy uppercase env-var-like values such as `MY_API_KEY` are migrated to `$MY_API_KEY` on startup.
-
 For `models.json`, shell commands are resolved at request time. pi intentionally does not apply built-in TTL, stale reuse, or recovery logic for arbitrary commands. Different commands need different caching and failure strategies, and pi cannot infer the right one.
 
 If your command is slow, expensive, rate-limited, or should keep using a previous value on transient failures, wrap it in your own script or command that implements the caching or TTL behavior you want.
@@ -401,14 +399,15 @@ For providers with partial OpenAI compatibility, use the `compat` field.
 | `requiresAssistantAfterToolResult` | Insert an assistant message before a user message after tool results |
 | `requiresThinkingAsText` | Convert thinking blocks to plain text |
 | `requiresReasoningContentOnAssistantMessages` | Include empty `reasoning_content` on all replayed assistant messages when reasoning is enabled |
-| `thinkingFormat` | Use `reasoning_effort`, `openrouter`, `deepseek`, `together`, `zai`, `qwen`, or `qwen-chat-template` thinking parameters |
+| `thinkingFormat` | Use `reasoning_effort`, `openrouter`, `deepseek`, `together`, `zai`, `qwen`, `chat-template`, or `qwen-chat-template` thinking parameters |
+| `chatTemplateKwargs` | `chat_template_kwargs` values for `thinkingFormat: "chat-template"`; use `{ "$var": "thinking.enabled" }` or `{ "$var": "thinking.effort" }` for pi-controlled thinking values |
 | `cacheControlFormat` | Use Anthropic-style `cache_control` markers on the system prompt, last tool definition, and last user/assistant text content. Currently only `anthropic` is supported. |
 | `supportsStrictMode` | Include the `strict` field in tool definitions |
 | `supportsLongCacheRetention` | Whether the provider accepts long cache retention when cache retention is `long`: `prompt_cache_retention: "24h"` for OpenAI prompt caching, or `cache_control.ttl: "1h"` when `cacheControlFormat` is `anthropic`. Default: `true`. |
 | `openRouterRouting` | OpenRouter provider routing preferences. This object is sent as-is in the `provider` field of the [OpenRouter API request](https://openrouter.ai/docs/guides/routing/provider-selection). |
 | `vercelGatewayRouting` | Vercel AI Gateway routing config for provider selection (`only`, `order`) |
 
-`openrouter` uses `reasoning: { effort }`. `together` uses `reasoning: { enabled }` and also `reasoning_effort` when `supportsReasoningEffort` is enabled. `qwen` uses top-level `enable_thinking`. Use `qwen-chat-template` for local Qwen-compatible servers that require `chat_template_kwargs.enable_thinking`.
+`openrouter` uses `reasoning: { effort }`. `together` uses `reasoning: { enabled }` and also `reasoning_effort` when `supportsReasoningEffort` is enabled. `qwen` uses top-level `enable_thinking`. Use `qwen-chat-template` for local Qwen-compatible servers that require `chat_template_kwargs.enable_thinking` and `preserve_thinking`. Use `chat-template` for vLLM/Hugging Face chat templates that need configurable `chat_template_kwargs`, such as `chatTemplateKwargs: { "thinking": { "$var": "thinking.enabled" } }` for DeepSeek V3.x templates.
 
 `cacheControlFormat: "anthropic"` is for OpenAI-compatible providers that expose Anthropic-style prompt caching through `cache_control` markers on text content and tool definitions.
 

packages/coding-agent/docs/packages.md

@@ -28,7 +28,8 @@ pi install ./relative/path/to/package
 
 pi remove npm:@foo/bar
 pi list                     # show installed packages from settings
-pi update                   # update pi, update packages, and reconcile pinned git refs
+pi update                   # update pi only
+pi update --all             # update pi, update packages, and reconcile pinned git refs
 pi update --extensions      # update packages and reconcile pinned git refs only
 pi update --self            # update pi only
 pi update --self --force    # reinstall pi even if current
@@ -36,7 +37,7 @@ pi update npm:@foo/bar      # update one package
 pi update --extension npm:@foo/bar
 ```
 
-These commands manage pi packages, not the pi CLI installation. To uninstall pi itself, see [Quickstart](quickstart.md#uninstall).
+These commands manage pi packages and `pi update` can update the pi CLI installation. To uninstall pi itself, see [Quickstart](quickstart.md#uninstall).
 
 By default, `install` and `remove` write to user settings (`~/.pi/agent/settings.json`). Use `-l` to write to project settings (`.pi/settings.json`) instead. Project settings can be shared with your team, and pi installs any missing packages automatically on startup after the project is trusted.
 
@@ -58,7 +59,7 @@ npm:@scope/pkg@1.2.3
 npm:pkg
 ```
 
-- Versioned specs are pinned and skipped by package updates (`pi update`, `pi update --extensions`).
+- Versioned specs are pinned and skipped by package updates (`pi update --extensions`, `pi update --all`).
 - User installs go under `~/.pi/agent/npm/`.
 - Project installs go under `.pi/npm/`.
 - Set `npmCommand` in `settings.json` to pin npm package lookup and install operations to a specific wrapper command such as `mise` or `asdf`.
@@ -85,7 +86,7 @@ ssh://git@github.com/user/repo@v1
 - HTTPS and SSH URLs are both supported.
 - SSH URLs use your configured SSH keys automatically (respects `~/.ssh/config`).
 - For non-interactive runs (for example CI), you can set `GIT_TERMINAL_PROMPT=0` to disable credential prompts and set `GIT_SSH_COMMAND` (for example `ssh -o BatchMode=yes -o ConnectTimeout=5`) to fail fast.
-- Refs are pinned tags or commits. `pi update` and `pi update --extensions` do not move them to newer refs, but they do reconcile an existing clone to the configured ref.
+- Refs are pinned tags or commits. `pi update --extensions` and `pi update --all` do not move them to newer refs, but they do reconcile an existing clone to the configured ref.
 - Use `pi install git:host/user/repo@new-ref` to update settings and move an existing package to a new pinned ref.
 - Cloned to `~/.pi/agent/git/<host>/<path>` (global) or `.pi/git/<host>/<path>` (project).
 - When reconciliation changes the checkout, pi resets and cleans the clone, then runs `npm install` if `package.json` exists.

packages/coding-agent/docs/providers.md

@@ -104,6 +104,24 @@ Store credentials in `~/.pi/agent/auth.json`:
 
 The file is created with `0600` permissions (user read/write only). Auth file credentials take priority over environment variables.
 
+API key credentials can also include provider-scoped environment values. These values are used before process environment variables when resolving the credential key, provider/model headers, and provider configuration such as Cloudflare account IDs, Azure OpenAI settings, Vertex project/location, Bedrock settings, `PI_CACHE_RETENTION`, and `HTTP_PROXY`/`HTTPS_PROXY`.
+
+```json
+{
+  "cloudflare-ai-gateway": {
+    "type": "api_key",
+    "key": "$CLOUDFLARE_API_KEY",
+    "env": {
+      "CLOUDFLARE_API_KEY": "...",
+      "CLOUDFLARE_ACCOUNT_ID": "account-id",
+      "CLOUDFLARE_GATEWAY_ID": "gateway-id"
+    }
+  }
+}
+```
+
+Use this when pi should use different provider settings than the project shell environment.
+
 ### Key Resolution
 
 The `key` field supports command execution, environment interpolation, and literals:
@@ -124,13 +142,13 @@ The `key` field supports command execution, environment interpolation, and liter
   { "type": "api_key", "key": "$$literal-dollar-prefix" }
   { "type": "api_key", "key": "$!literal-bang-prefix" }
   ```
-- **Literal value:** Used directly
+- **Literal value:** Used directly. Plain uppercase strings such as `MY_API_KEY` are literals; use `$MY_API_KEY` for environment variables.
   ```json
   { "type": "api_key", "key": "sk-ant-..." }
   { "type": "api_key", "key": "public" }
   ```
 
-Legacy uppercase env-var-like values such as `MY_API_KEY` are migrated to `$MY_API_KEY` on startup. OAuth credentials are also stored here after `/login` and managed automatically.
+OAuth credentials are also stored here after `/login` and managed automatically.
 
 ## Cloud Providers
 
@@ -194,7 +212,7 @@ export AWS_BEDROCK_FORCE_HTTP1=1
 
 ### Cloudflare AI Gateway
 
-`CLOUDFLARE_API_KEY` can be set via `/login`. The account ID and gateway slug must be set as environment variables.
+`CLOUDFLARE_API_KEY` can be set via `/login`. The account ID and gateway slug can be set as environment variables or in the API key credential's `env` object in `auth.json`.
 
 ```bash
 export CLOUDFLARE_API_KEY=...           # or use /login
@@ -218,7 +236,7 @@ For normal pi usage, prefer unified billing or stored BYOK. Inline BYOK requires
 
 ### Cloudflare Workers AI
 
-`CLOUDFLARE_API_KEY` can be set via `/login`. `CLOUDFLARE_ACCOUNT_ID` must be set as an environment variable.
+`CLOUDFLARE_API_KEY` can be set via `/login`. `CLOUDFLARE_ACCOUNT_ID` can be set as an environment variable or in the API key credential's `env` object in `auth.json`.
 
 ```bash
 export CLOUDFLARE_API_KEY=...           # or use /login

packages/coding-agent/docs/rpc.md

@@ -374,11 +374,14 @@ Response:
     "summary": "Summary of conversation...",
     "firstKeptEntryId": "abc123",
     "tokensBefore": 150000,
+    "estimatedTokensAfter": 32000,
     "details": {}
   }
 }
 ```
 
+`estimatedTokensAfter` is a heuristic estimate over the rebuilt message context immediately after compaction, not a provider-exact token count.
+
 #### set_auto_compaction
 
 Enable or disable automatic compaction when context is nearly full.
@@ -924,6 +927,7 @@ The `reason` field is `"manual"`, `"threshold"`, or `"overflow"`.
     "summary": "Summary of conversation...",
     "firstKeptEntryId": "abc123",
     "tokensBefore": 150000,
+    "estimatedTokensAfter": 32000,
     "details": {}
   },
   "aborted": false,

packages/coding-agent/docs/sdk.md

@@ -1110,7 +1110,8 @@ DefaultResourceLoader
 type ResourceLoader
 createEventBus
 
-// Helpers
+// Constants and helpers
+CONFIG_DIR_NAME
 defineTool
 getAgentDir
 getPackageDir

packages/coding-agent/docs/security.md

@@ -6,14 +6,18 @@ Pi is a local coding agent. It runs with the permissions of the user account tha
 
 Project trust controls whether pi loads project-local settings, resources, packages, and extensions. It is not a sandbox and it does not restrict what the model can ask tools to do after you start working in a directory.
 
-Pi considers a project to have trust inputs when it finds any of these from the current working directory:
+Pi considers a project to have resources that require trust when it finds any of these from the current working directory:
 
-- `.pi/` in the current directory
-- `.agents/skills` in the current directory or an ancestor directory
+- `.pi/settings.json`
+- `.pi/extensions`, `.pi/skills`, `.pi/prompts`, or `.pi/themes`
+- `.pi/SYSTEM.md` or `.pi/APPEND_SYSTEM.md`
+- project `.agents/skills` in the current directory or an ancestor directory
 
-When an interactive session starts in a project with configs in `.pi` or `.agents/skills` and no saved decision for the current directory or a parent directory, pi follows `defaultProjectTrust` from global settings. The default value is `"ask"`, which asks whether to trust the project when UI is available. Saved decisions are stored by canonical directory in `~/.pi/agent/trust.json`, and the closest saved decision on the current or parent path applies before the global default.
+A bare `.pi` directory does not count as a project resource that requires trust.
 
-Trusting a project allows pi to load trust-gated project inputs, including:
+When an interactive session starts in a project with resources that require trust and no saved decision for the current directory or a parent directory, pi follows `defaultProjectTrust` from global settings. The default value is `"ask"`, which asks whether to trust the project when UI is available. Saved decisions are stored by canonical directory in `~/.pi/agent/trust.json`, and the closest saved decision on the current or parent path applies before the global default.
+
+Trusting a project allows pi to load project resources that require trust, including:
 
 - `.pi/settings.json`
 - `.pi` resources such as extensions, skills, prompt templates, themes, and system prompt files
@@ -38,7 +42,7 @@ For untrusted repositories, generated code you do not intend to monitor closely,
 
 Common patterns are documented in [Containerization](containerization.md):
 
-- run the whole `pi` process inside OpenShell or Docker
+- run the whole `pi` process inside a container/sandbox
 - run host pi while routing built-in tool execution into a Gondolin micro-VM
 - mount only the workspace paths the agent should access
 - avoid mounting host `~/.pi/agent` unless the container should access host sessions, settings, and credentials

packages/coding-agent/docs/settings.md

@@ -11,13 +11,13 @@ Edit directly or use `/settings` for common options.
 
 ## Project Trust
 
-On interactive startup, pi asks before trusting a project folder that contains trust-gated project inputs and has no saved decision for the folder or a parent folder in `~/.pi/agent/trust.json`. Trusting a project allows pi to load `.pi/settings.json` and `.pi` resources, install missing project packages, and execute project extensions.
+On interactive startup, pi asks before trusting a project folder that contains project-local settings, resources, or project `.agents/skills` and has no saved decision for the folder or a parent folder in `~/.pi/agent/trust.json`. Trusting a project allows pi to load `.pi/settings.json` and `.pi` resources, install missing project packages, and execute project extensions.
 
-Non-interactive modes (`-p`, `--mode json`, and `--mode rpc`) do not show a trust prompt. Without an applicable saved trust decision, they use `defaultProjectTrust` from global settings: `ask` (default) and `never` ignore trust-gated project inputs, while `always` trusts them. Pass `--approve`/`-a` or `--no-approve`/`-na` to override project trust for one run.
+Non-interactive modes (`-p`, `--mode json`, and `--mode rpc`) do not show a trust prompt. Without an applicable saved trust decision, they use `defaultProjectTrust` from global settings: `ask` (default) and `never` ignore those project resources, while `always` trusts them. Pass `--approve`/`-a` or `--no-approve`/`-na` to override project trust for one run.
 
 If no extension or saved decision applies, `defaultProjectTrust` controls the fallback behavior. Set it to `"ask"`, `"always"`, or `"never"` in `~/.pi/agent/settings.json`, or change it with `/settings`.
 
-`pi config` and package commands use the same project trust flow. Pass `--approve` to trust project-local settings for one command or `--no-approve` to ignore them.
+`pi config` and package commands use the same project trust flow, except `pi update` never prompts. Pass `--approve` to trust project-local settings for one command or `--no-approve` to ignore them.
 
 Use `/trust` in interactive mode to save a project trust decision for future sessions, including trust for the immediate parent folder. It writes `~/.pi/agent/trust.json` only; the current session is not reloaded, so restart pi for changes to take effect.
 
@@ -69,6 +69,18 @@ Use `/trust` in interactive mode to save a project trust decision for future ses
 
 Set `PI_SKIP_VERSION_CHECK=1` to disable the Pi version update check. Use `--offline` or `PI_OFFLINE=1` to disable all startup network operations described here, including update checks, package update checks, and install/update telemetry.
 
+### Network
+
+| Setting | Type | Default | Description |
+|---------|------|---------|-------------|
+| `httpProxy` | string | - | HTTP proxy URL applied as `HTTP_PROXY` and `HTTPS_PROXY`. Global setting only. |
+
+```json
+{
+  "httpProxy": "http://127.0.0.1:7890"
+}
+```
+
 ### Warnings
 
 | Setting | Type | Default | Description |

packages/coding-agent/docs/themes.md

@@ -137,7 +137,7 @@ vim ~/.pi/agent/themes/my-theme.json
 }
 ```
 
-- `name` is required and must be unique.
+- `name` is required, must be unique, and must not contain `/`.
 - `vars` is optional. Define reusable colors here, then reference them in `colors`.
 - `colors` must define all 51 required tokens.
 

packages/coding-agent/docs/tui.md

@@ -257,7 +257,7 @@ md.setText("Updated markdown");
 
 ### Image
 
-Renders images in supported terminals (Kitty, iTerm2, Ghostty, WezTerm).
+Renders images in supported terminals (Kitty, iTerm2, Ghostty, WezTerm, Warp).
 
 ```typescript
 const image = new Image(
@@ -742,7 +742,7 @@ ctx.ui.setStatus("my-ext", ctx.ui.theme.fg("accent", "● active"));
 ctx.ui.setStatus("my-ext", undefined);
 ```
 
-**Examples:** [status-line.ts](../examples/extensions/status-line.ts), [plan-mode.ts](../examples/extensions/plan-mode.ts), [preset.ts](../examples/extensions/preset.ts)
+**Examples:** [status-line.ts](../examples/extensions/status-line.ts), [plan-mode/index.ts](../examples/extensions/plan-mode/index.ts), [preset.ts](../examples/extensions/preset.ts)
 
 ### Pattern 4b: Working Indicator Customization
 
@@ -802,7 +802,7 @@ ctx.ui.setWidget("my-widget", (_tui, theme) => {
 ctx.ui.setWidget("my-widget", undefined);
 ```
 
-**Examples:** [plan-mode.ts](../examples/extensions/plan-mode.ts)
+**Examples:** [plan-mode/index.ts](../examples/extensions/plan-mode/index.ts)
 
 ### Pattern 6: Custom Footer
 
@@ -919,7 +919,7 @@ export default function (pi: ExtensionAPI) {
 - **Selection UI**: [examples/extensions/preset.ts](../examples/extensions/preset.ts) - SelectList with DynamicBorder framing
 - **Async with cancel**: [examples/extensions/qna.ts](../examples/extensions/qna.ts) - BorderedLoader for LLM calls
 - **Settings toggles**: [examples/extensions/tools.ts](../examples/extensions/tools.ts) - SettingsList for tool enable/disable
-- **Status indicators**: [examples/extensions/plan-mode.ts](../examples/extensions/plan-mode.ts) - setStatus and setWidget
+- **Status indicators**: [examples/extensions/plan-mode/index.ts](../examples/extensions/plan-mode/index.ts) - setStatus and setWidget
 - **Working indicator**: [examples/extensions/working-indicator.ts](../examples/extensions/working-indicator.ts) - setWorkingIndicator
 - **Custom footer**: [examples/extensions/custom-footer.ts](../examples/extensions/custom-footer.ts) - setFooter with stats
 - **Custom editor**: [examples/extensions/modal-editor.ts](../examples/extensions/modal-editor.ts) - Vim-like modal editing

packages/coding-agent/docs/usage.md

@@ -44,11 +44,13 @@ Type `/` in the editor to open command completion. Extensions can register custo
 | `/name <name>` | Set session display name |
 | `/session` | Show session file, ID, messages, tokens, and cost |
 | `/tree` | Jump to any point in the session and continue from there |
+| `/trust` | Save project trust decision for future sessions |
 | `/fork` | Create a new session from a previous user message |
 | `/clone` | Duplicate the current active branch into a new session |
 | `/compact [prompt]` | Manually compact context, optionally with custom instructions |
 | `/copy` | Copy last assistant message to clipboard |
-| `/export [file]` | Export session to HTML |
+| `/export [file]` | Export session to HTML or JSONL |
+| `/import <file>` | Import and resume a session from a JSONL file |
 | `/share` | Upload as private GitHub gist with shareable HTML link |
 | `/reload` | Reload keybindings, extensions, skills, prompts, and context files |
 | `/hotkeys` | Show all keyboard shortcuts |
@@ -112,15 +114,15 @@ Append to the default prompt without replacing it with `APPEND_SYSTEM.md` in eit
 
 ### Project Trust
 
-On interactive startup, pi asks before trusting a project folder that contains project-local extensions or settings and has no saved decision for the folder or a parent folder in `~/.pi/agent/trust.json`. Trusting a project allows pi to load `.pi/settings.json` and `.pi` resources, install missing project packages, and execute project extensions.
+On interactive startup, pi asks before trusting a project folder that contains project-local settings, resources, or project `.agents/skills` and has no saved decision for the folder or a parent folder in `~/.pi/agent/trust.json`. Trusting a project allows pi to load `.pi/settings.json` and `.pi` resources, install missing project packages, and execute project extensions.
 
 Before the trust decision, pi loads only context files, user/global extensions, and CLI `-e` extensions so they can handle the `project_trust` event. Project-local extensions, project package-managed extensions, and project settings are loaded only after the project is trusted. This split also applies when switching to a session from a different cwd whose trust has not been resolved in the current process.
 
-Non-interactive modes (`-p`, `--mode json`, and `--mode rpc`) do not show a trust prompt. Without an applicable saved trust decision, they use `defaultProjectTrust` from global settings: `ask` (default) and `never` ignore trust-gated project inputs, while `always` trusts them. Pass `--approve`/`-a` or `--no-approve`/`-na` to override project trust for one run.
+Non-interactive modes (`-p`, `--mode json`, and `--mode rpc`) do not show a trust prompt. Without an applicable saved trust decision, they use `defaultProjectTrust` from global settings: `ask` (default) and `never` ignore those project resources, while `always` trusts them. Pass `--approve`/`-a` or `--no-approve`/`-na` to override project trust for one run.
 
 If no extension or saved decision applies, `defaultProjectTrust` controls the fallback behavior. Set it to `"ask"`, `"always"`, or `"never"` in `~/.pi/agent/settings.json`, or change it with `/settings`.
 
-`pi config` and package commands use the same project trust flow. Pass `--approve` to trust project-local settings for one command or `--no-approve` to ignore them.
+`pi config` and package commands use the same project trust flow, except `pi update` never prompts. Pass `--approve` to trust project-local settings for one command or `--no-approve` to ignore them.
 
 Use `/trust` in interactive mode to save a project trust decision for future sessions, including trust for the immediate parent folder. It writes `~/.pi/agent/trust.json` only; the current session is not reloaded, so restart pi for changes to take effect.
 
@@ -145,7 +147,8 @@ pi [options] [@files...] [messages...]
 pi install <source> [-l]     # Install package, -l for project-local
 pi remove <source> [-l]      # Remove package
 pi uninstall <source> [-l]   # Alias for remove
-pi update [source|self|pi]   # Update pi and packages; reconcile pinned git refs
+pi update [source|self|pi]   # Update pi only, or one package source
+pi update --all              # Update pi and packages; reconcile pinned git refs
 pi update --extensions       # Update packages only; reconcile pinned git refs
 pi update --self             # Update pi only
 pi update --extension <src>  # Update one package
@@ -153,7 +156,7 @@ pi list                      # List installed packages
 pi config                    # Enable/disable package resources
 ```
 
-These commands manage pi packages, not the pi CLI installation. To uninstall pi itself, see [Quickstart](quickstart.md#uninstall). `pi config` and project package commands accept `--approve`/`--no-approve` to trust or ignore project-local settings for one command.
+These commands manage pi packages and `pi update` can update the pi CLI installation. To uninstall pi itself, see [Quickstart](quickstart.md#uninstall). `pi config` and project package commands accept `--approve`/`--no-approve` to trust or ignore project-local settings for one command. `pi update` never prompts for project trust.
 
 See [Pi Packages](packages.md) for package sources and security notes.
 

packages/coding-agent/examples/extensions/custom-provider-anthropic/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "pi-extension-custom-provider",
-  "version": "0.79.1",
+  "version": "0.79.10",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "pi-extension-custom-provider",
-      "version": "0.79.1",
+      "version": "0.79.10",
       "dependencies": {
         "@anthropic-ai/sdk": "^0.52.0"
       }

packages/coding-agent/examples/extensions/custom-provider-anthropic/package.json

@@ -1,7 +1,7 @@
 {
   "name": "pi-extension-custom-provider-anthropic",
   "private": true,
-  "version": "0.79.1",
+  "version": "0.79.10",
   "type": "module",
   "scripts": {
     "clean": "echo 'nothing to clean'",

packages/coding-agent/examples/extensions/custom-provider-gitlab-duo/package.json

@@ -1,7 +1,7 @@
 {
   "name": "pi-extension-custom-provider-gitlab-duo",
   "private": true,
-  "version": "0.79.1",
+  "version": "0.79.10",
   "type": "module",
   "scripts": {
     "clean": "echo 'nothing to clean'",