ton/pi_harness
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat(ai): complete models runtime migration

Browse Source
This commit is contained in:
Mario Zechner
2026-06-23 15:29:17 +02:00
parent 470a4736a3
commit 129eb460cd
47 changed files with 1502 additions and 576 deletions
Download Patch File Download Diff File Expand all files Collapse all files
packages/ai/src/api/anthropic-messages.ts
+49 -62
View File
@@ -18,6 +18,7 @@ import type {
Message,
Model,
ProviderEnv,
ProviderHeaders,
SimpleStreamOptions,
StopReason,
StreamFunction,
@@ -34,7 +35,6 @@ 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";
import { buildCopilotDynamicHeaders, hasCopilotVisionInput } from "./github-copilot-headers.ts";
import { adjustMaxTokensForThinking, buildBaseOptions } from "./simple-options.ts";
import { transformMessages } from "./transform-messages.ts";
@@ -170,16 +170,11 @@ const INTERLEAVED_THINKING_BETA = "interleaved-thinking-2025-05-14";
function getAnthropicCompat(
model: Model<"anthropic-messages">,
): Required<Omit<AnthropicMessagesCompat, "forceAdaptiveThinking">> {
// Auto-detect session affinity and cache control support from provider
const isFireworks = model.provider === "fireworks";
const isCloudflareAiGatewayAnthropic =
model.provider === "cloudflare-ai-gateway" && model.baseUrl.includes("anthropic");
return {
supportsEagerToolInputStreaming: model.compat?.supportsEagerToolInputStreaming ?? !isFireworks,
supportsLongCacheRetention: model.compat?.supportsLongCacheRetention ?? !isFireworks,
sendSessionAffinityHeaders:
model.compat?.sendSessionAffinityHeaders ?? !!(isFireworks || isCloudflareAiGatewayAnthropic),
supportsCacheControlOnTools: model.compat?.supportsCacheControlOnTools ?? !isFireworks,
supportsEagerToolInputStreaming: model.compat?.supportsEagerToolInputStreaming ?? true,
supportsLongCacheRetention: model.compat?.supportsLongCacheRetention ?? true,
sendSessionAffinityHeaders: model.compat?.sendSessionAffinityHeaders ?? false,
supportsCacheControlOnTools: model.compat?.supportsCacheControlOnTools ?? true,
supportsTemperature: model.compat?.supportsTemperature ?? true,
allowEmptySignature: model.compat?.allowEmptySignature ?? false,
};
@@ -247,8 +242,8 @@ export interface AnthropicOptions extends StreamOptions {
client?: Anthropic;
}
function mergeHeaders(...headerSources: (Record<string, string | null> | undefined)[]): Record<string, string | null> {
const merged: Record<string, string | null> = {};
function mergeHeaders(...headerSources: (ProviderHeaders | undefined)[]): ProviderHeaders {
const merged: ProviderHeaders = {};
for (const headers of headerSources) {
if (headers) {
Object.assign(merged, headers);
@@ -257,6 +252,27 @@ function mergeHeaders(...headerSources: (Record<string, string | null> | undefin
return merged;
}
function hasHeader(headers: ProviderHeaders | undefined, name: string): boolean {
if (!headers) return false;
const expected = name.toLowerCase();
for (const [key, value] of Object.entries(headers)) {
if (key.toLowerCase() === expected && value !== null && value.trim().length > 0) return true;
}
return false;
}
function assertRequestAuth(provider: string, apiKey: string | undefined, headers: ProviderHeaders | undefined): void {
if (apiKey) return;
if (
hasHeader(headers, "authorization") ||
hasHeader(headers, "x-api-key") ||
hasHeader(headers, "cf-aig-authorization")
) {
return;
}
throw new Error(`No API key for provider: ${provider}`);
}
interface ServerSentEvent {
event: string | null;
data: string;
@@ -484,9 +500,7 @@ export const stream: StreamFunction<"anthropic-messages", AnthropicOptions> = (
isOAuth = false;
} else {
const apiKey = options?.apiKey;
if (!apiKey) {
throw new Error(`No API key for provider: ${model.provider}`);
}
assertRequestAuth(model.provider, apiKey, options?.headers);
let copilotDynamicHeaders: Record<string, string> | undefined;
if (model.provider === "github-copilot") {
@@ -508,7 +522,6 @@ export const stream: StreamFunction<"anthropic-messages", AnthropicOptions> = (
options?.headers,
copilotDynamicHeaders,
cacheSessionId,
options?.env,
);
client = created.client;
isOAuth = created.isOAuthToken;
@@ -748,12 +761,9 @@ export const streamSimple: StreamFunction<"anthropic-messages", SimpleStreamOpti
context: Context,
options?: SimpleStreamOptions,
): AssistantMessageEventStream => {
const apiKey = options?.apiKey;
if (!apiKey) {
throw new Error(`No API key for provider: ${model.provider}`);
}
assertRequestAuth(model.provider, options?.apiKey, options?.headers);
const base = buildBaseOptions(model, options, apiKey);
const base = buildBaseOptions(model, options, options?.apiKey);
if (!options?.reasoning) {
return stream(model, context, { ...base, thinkingEnabled: false } satisfies AnthropicOptions);
}
@@ -792,13 +802,12 @@ function isOAuthToken(apiKey: string): boolean {
function createClient(
model: Model<"anthropic-messages">,
apiKey: string,
apiKey: string | undefined,
interleavedThinking: boolean,
useFineGrainedToolStreamingBeta: boolean,
optionsHeaders?: Record<string, string>,
optionsHeaders?: ProviderHeaders,
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;
@@ -810,34 +819,11 @@ function createClient(
betaFeatures.push(INTERLEAVED_THINKING_BETA);
}
if (model.provider === "cloudflare-ai-gateway") {
const client = new Anthropic({
apiKey: null,
authToken: null,
baseURL: resolveCloudflareBaseUrl(model, env),
dangerouslyAllowBrowser: true,
defaultHeaders: mergeHeaders(
{
accept: "application/json",
"anthropic-dangerous-direct-browser-access": "true",
"cf-aig-authorization": `Bearer ${apiKey}`,
"x-api-key": null,
Authorization: null,
...(betaFeatures.length > 0 ? { "anthropic-beta": betaFeatures.join(",") } : {}),
},
model.headers,
optionsHeaders,
),
});
return { client, isOAuthToken: false };
}
// Copilot: Bearer auth, selective betas.
if (model.provider === "github-copilot") {
const client = new Anthropic({
apiKey: null,
authToken: apiKey,
authToken: apiKey ?? null,
baseURL: model.baseUrl,
dangerouslyAllowBrowser: true,
defaultHeaders: mergeHeaders(
@@ -856,7 +842,7 @@ function createClient(
}
// OAuth: Bearer auth, Claude Code identity headers
if (isOAuthToken(apiKey)) {
if (apiKey && isOAuthToken(apiKey)) {
const client = new Anthropic({
apiKey: null,
authToken: apiKey,
@@ -878,24 +864,25 @@ function createClient(
return { client, isOAuthToken: true };
}
// API key auth
const sessionAffinityHeaders: Record<string, string | null> =
// API key or header-owned auth.
const sessionAffinityHeaders: ProviderHeaders =
sessionId && getAnthropicCompat(model).sendSessionAffinityHeaders ? { "x-session-affinity": sessionId } : {};
const defaultHeaders = mergeHeaders(
{
accept: "application/json",
"anthropic-dangerous-direct-browser-access": "true",
...(betaFeatures.length > 0 ? { "anthropic-beta": betaFeatures.join(",") } : {}),
},
sessionAffinityHeaders,
model.headers,
optionsHeaders,
);
const client = new Anthropic({
apiKey,
apiKey: apiKey ?? null,
authToken: null,
baseURL: model.baseUrl,
dangerouslyAllowBrowser: true,
defaultHeaders: mergeHeaders(
{
accept: "application/json",
"anthropic-dangerous-direct-browser-access": "true",
...(betaFeatures.length > 0 ? { "anthropic-beta": betaFeatures.join(",") } : {}),
},
sessionAffinityHeaders,
model.headers,
optionsHeaders,
),
defaultHeaders,
});
return { client, isOAuthToken: false };
packages/ai/src/api/bedrock-converse-stream.ts
+4 -2
View File
@@ -48,6 +48,7 @@ import type {
ToolResultMessage,
} from "../types.ts";
import { AssistantMessageEventStream } from "../utils/event-stream.ts";
import { providerHeadersToRecord } from "../utils/headers.ts";
import { parseStreamingJson } from "../utils/json-parse.ts";
import { resolveHttpProxyUrlForTarget } from "../utils/node-http-proxy.ts";
import { getProviderEnvValue } from "../utils/provider-env.ts";
@@ -204,8 +205,9 @@ export const stream: StreamFunction<"bedrock-converse-stream", BedrockOptions> =
try {
const client = new BedrockRuntimeClient(config);
if (options.headers && Object.keys(options.headers).length > 0) {
addCustomHeadersMiddleware(client, options.headers);
const customHeaders = providerHeadersToRecord(options.headers);
if (customHeaders) {
addCustomHeadersMiddleware(client, customHeaders);
}
const cacheRetention = resolveCacheRetention(options.cacheRetention, options.env);
const inferenceMaxTokens = options.maxTokens ?? (isAnthropicClaudeModel(model) ? model.maxTokens : undefined);
packages/ai/src/api/cloudflare.ts
-21
View File
@@ -1,6 +1,3 @@
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 =
"https://api.cloudflare.com/client/v4/accounts/{CLOUDFLARE_ACCOUNT_ID}/ai/v1";
@@ -16,21 +13,3 @@ export const CLOUDFLARE_AI_GATEWAY_OPENAI_BASE_URL =
/** AI Gateway → Anthropic passthrough. */
export const CLOUDFLARE_AI_GATEWAY_ANTHROPIC_BASE_URL =
"https://gateway.ai.cloudflare.com/v1/{CLOUDFLARE_ACCOUNT_ID}/{CLOUDFLARE_GATEWAY_ID}/anthropic";
export function isCloudflareProvider(provider: string): boolean {
return provider === "cloudflare-workers-ai" || provider === "cloudflare-ai-gateway";
}
/** 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 = getProviderEnvValue(name, env);
if (!value) {
throw new Error(`${name} is required for provider ${model.provider} but is not set.`);
}
return value;
});
return baseUrl;
}
packages/ai/src/api/google-generative-ai.ts
+6 -3
View File
@@ -10,6 +10,7 @@ import type {
AssistantMessage,
Context,
Model,
ProviderHeaders,
SimpleStreamOptions,
StreamFunction,
StreamOptions,
@@ -20,6 +21,7 @@ import type {
ToolCall,
} from "../types.ts";
import { AssistantMessageEventStream } from "../utils/event-stream.ts";
import { providerHeadersToRecord } from "../utils/headers.ts";
import { sanitizeSurrogates } from "../utils/sanitize-unicode.ts";
import type { GoogleThinkingLevel } from "./google-shared.ts";
import {
@@ -318,15 +320,16 @@ export const streamSimple: StreamFunction<"google-generative-ai", SimpleStreamOp
function createClient(
model: Model<"google-generative-ai">,
apiKey?: string,
optionsHeaders?: Record<string, string>,
optionsHeaders?: ProviderHeaders,
): GoogleGenAI {
const httpOptions: { baseUrl?: string; apiVersion?: string; headers?: Record<string, string> } = {};
if (model.baseUrl) {
httpOptions.baseUrl = model.baseUrl;
httpOptions.apiVersion = ""; // baseUrl already includes version path, don't append
}
if (model.headers || optionsHeaders) {
httpOptions.headers = { ...model.headers, ...optionsHeaders };
const headers = providerHeadersToRecord({ ...model.headers, ...optionsHeaders });
if (headers) {
httpOptions.headers = headers;
}
return new GoogleGenAI({
packages/ai/src/api/google-vertex.ts
+8 -8
View File
@@ -15,6 +15,7 @@ import type {
Model,
ThinkingLevel as PiThinkingLevel,
ProviderEnv,
ProviderHeaders,
SimpleStreamOptions,
StreamFunction,
StreamOptions,
@@ -24,6 +25,7 @@ import type {
ToolCall,
} from "../types.ts";
import { AssistantMessageEventStream } from "../utils/event-stream.ts";
import { providerHeadersToRecord } from "../utils/headers.ts";
import { getProviderEnvValue } from "../utils/provider-env.ts";
import { sanitizeSurrogates } from "../utils/sanitize-unicode.ts";
import type { GoogleThinkingLevel } from "./google-shared.ts";
@@ -334,7 +336,7 @@ function createClient(
model: Model<"google-vertex">,
project: string,
location: string,
optionsHeaders?: Record<string, string>,
optionsHeaders?: ProviderHeaders,
env?: ProviderEnv,
): GoogleGenAI {
const googleAuthOptions = buildGoogleAuthOptions(env);
@@ -351,7 +353,7 @@ function createClient(
function createClientWithApiKey(
model: Model<"google-vertex">,
apiKey: string,
optionsHeaders?: Record<string, string>,
optionsHeaders?: ProviderHeaders,
): GoogleGenAI {
return new GoogleGenAI({
vertexai: true,
@@ -361,10 +363,7 @@ function createClientWithApiKey(
});
}
function buildHttpOptions(
model: Model<"google-vertex">,
optionsHeaders?: Record<string, string>,
): HttpOptions | undefined {
function buildHttpOptions(model: Model<"google-vertex">, optionsHeaders?: ProviderHeaders): HttpOptions | undefined {
const httpOptions: HttpOptions = {};
const baseUrl = resolveCustomBaseUrl(model.baseUrl);
if (baseUrl) {
@@ -375,8 +374,9 @@ function buildHttpOptions(
}
}
if (model.headers || optionsHeaders) {
httpOptions.headers = { ...model.headers, ...optionsHeaders };
const headers = providerHeadersToRecord({ ...model.headers, ...optionsHeaders });
if (headers) {
httpOptions.headers = headers;
}
return Object.keys(httpOptions).length > 0 ? httpOptions : undefined;
packages/ai/src/api/openai-codex-responses.ts
+9 -4
View File
@@ -28,6 +28,7 @@ import type {
Context,
Model,
ProviderEnv,
ProviderHeaders,
SimpleStreamOptions,
StreamFunction,
StreamOptions,
@@ -1467,13 +1468,17 @@ function createCodexRequestId(): string {
function buildBaseCodexHeaders(
initHeaders: Record<string, string> | undefined,
additionalHeaders: Record<string, string> | undefined,
additionalHeaders: ProviderHeaders | undefined,
accountId: string,
token: string,
): Headers {
const headers = new Headers(initHeaders);
for (const [key, value] of Object.entries(additionalHeaders || {})) {
headers.set(key, value);
if (value === null) {
headers.delete(key);
} else {
headers.set(key, value);
}
}
headers.set("Authorization", `Bearer ${token}`);
headers.set("chatgpt-account-id", accountId);
@@ -1485,7 +1490,7 @@ function buildBaseCodexHeaders(
function buildSSEHeaders(
initHeaders: Record<string, string> | undefined,
additionalHeaders: Record<string, string> | undefined,
additionalHeaders: ProviderHeaders | undefined,
accountId: string,
token: string,
sessionId?: string,
@@ -1505,7 +1510,7 @@ function buildSSEHeaders(
function buildWebSocketHeaders(
initHeaders: Record<string, string> | undefined,
additionalHeaders: Record<string, string> | undefined,
additionalHeaders: ProviderHeaders | undefined,
accountId: string,
token: string,
requestId: string,
packages/ai/src/api/openai-completions.ts
+65 -134
View File
@@ -21,6 +21,7 @@ import type {
Model,
OpenAICompletionsCompat,
ProviderEnv,
ProviderHeaders,
SimpleStreamOptions,
StopReason,
StreamFunction,
@@ -36,7 +37,6 @@ 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";
import { clampOpenAIPromptCacheKey } from "./openai-prompt-cache.ts";
import { buildBaseOptions } from "./simple-options.ts";
@@ -47,6 +47,21 @@ import { transformMessages } from "./transform-messages.ts";
* This is needed because Anthropic (via proxy) requires the tools param
* to be present when messages include tool_calls or tool role messages.
*/
function hasHeader(headers: ProviderHeaders | undefined, name: string): boolean {
if (!headers) return false;
const expected = name.toLowerCase();
for (const [key, value] of Object.entries(headers)) {
if (key.toLowerCase() === expected && value !== null && value.trim().length > 0) return true;
}
return false;
}
function getClientApiKey(provider: string, apiKey: string | undefined, headers: ProviderHeaders | undefined): string {
if (apiKey) return apiKey;
if (hasHeader(headers, "authorization") || hasHeader(headers, "cf-aig-authorization")) return "unused";
throw new Error(`No API key for provider: ${provider}`);
}
function hasToolHistory(messages: Message[]): boolean {
for (const msg of messages) {
if (msg.role === "toolResult") {
@@ -160,14 +175,11 @@ export const stream: StreamFunction<"openai-completions", OpenAICompletionsOptio
};
try {
const apiKey = options?.apiKey;
if (!apiKey) {
throw new Error(`No API key for provider: ${model.provider}`);
}
const apiKey = getClientApiKey(model.provider, options?.apiKey, options?.headers);
const compat = getCompat(model);
const cacheRetention = resolveCacheRetention(options?.cacheRetention, options?.env);
const cacheSessionId = cacheRetention === "none" ? undefined : options?.sessionId;
const client = createClient(model, context, apiKey, options?.headers, cacheSessionId, compat, options?.env);
const client = createClient(model, context, apiKey, options?.headers, cacheSessionId, compat);
let params = buildParams(model, context, options, compat, cacheRetention);
const nextParams = await options?.onPayload?.(params, model);
if (nextParams !== undefined) {
@@ -468,12 +480,9 @@ export const streamSimple: StreamFunction<"openai-completions", SimpleStreamOpti
context: Context,
options?: SimpleStreamOptions,
): AssistantMessageEventStream => {
const apiKey = options?.apiKey;
if (!apiKey) {
throw new Error(`No API key for provider: ${model.provider}`);
}
getClientApiKey(model.provider, options?.apiKey, options?.headers);
const base = buildBaseOptions(model, options, apiKey);
const base = buildBaseOptions(model, options, options?.apiKey);
const clampedReasoning = options?.reasoning ? clampThinkingLevel(model, options.reasoning) : undefined;
const reasoningEffort = clampedReasoning === "off" ? undefined : clampedReasoning;
const toolChoice = (options as OpenAICompletionsOptions | undefined)?.toolChoice;
@@ -489,12 +498,11 @@ function createClient(
model: Model<"openai-completions">,
context: Context,
apiKey: string,
optionsHeaders?: Record<string, string>,
optionsHeaders?: ProviderHeaders,
sessionId?: string,
compat: ResolvedOpenAICompletionsCompat = getCompat(model),
env?: ProviderEnv,
) {
const headers = { ...model.headers };
const headers: ProviderHeaders = { ...model.headers };
if (model.provider === "github-copilot") {
const hasImages = hasCopilotVisionInput(context.messages);
const copilotHeaders = buildCopilotDynamicHeaders({
@@ -515,20 +523,11 @@ function createClient(
Object.assign(headers, optionsHeaders);
}
const defaultHeaders =
model.provider === "cloudflare-ai-gateway"
? {
...headers,
Authorization: headers.Authorization ?? null,
"cf-aig-authorization": `Bearer ${apiKey}`,
}
: headers;
return new OpenAI({
apiKey,
baseURL: isCloudflareProvider(model.provider) ? resolveCloudflareBaseUrl(model, env) : model.baseUrl,
baseURL: model.baseUrl,
dangerouslyAllowBrowser: true,
defaultHeaders,
defaultHeaders: headers,
});
}
@@ -674,7 +673,7 @@ function buildParams(
}
// Vercel AI Gateway provider routing preferences
if (model.baseUrl.includes("ai-gateway.vercel.sh") && model.compat?.vercelGatewayRouting) {
if (model.compat?.vercelGatewayRouting) {
const routing = model.compat.vercelGatewayRouting;
if (routing.only || routing.order) {
const gatewayOptions: Record<string, string[]> = {};
@@ -1164,123 +1163,55 @@ function mapStopReason(reason: ChatCompletionChunk.Choice["finish_reason"] | str
}
}
/**
* Detect compatibility settings from provider and baseUrl for known providers.
* Provider takes precedence over URL-based detection since it's explicitly configured.
* Returns a fully resolved OpenAICompletionsCompat object with all fields set.
*/
function detectCompat(model: Model<"openai-completions">): ResolvedOpenAICompletionsCompat {
const provider = model.provider;
const baseUrl = model.baseUrl;
const isZai =
provider === "zai" ||
provider === "zai-coding-cn" ||
baseUrl.includes("api.z.ai") ||
baseUrl.includes("open.bigmodel.cn");
const isTogether =
provider === "together" || baseUrl.includes("api.together.ai") || baseUrl.includes("api.together.xyz");
const isMoonshot = provider === "moonshotai" || provider === "moonshotai-cn" || baseUrl.includes("api.moonshot.");
const isOpenRouter = provider === "openrouter" || baseUrl.includes("openrouter.ai");
const isCloudflareWorkersAI = provider === "cloudflare-workers-ai" || baseUrl.includes("api.cloudflare.com");
const isCloudflareAiGateway = provider === "cloudflare-ai-gateway" || baseUrl.includes("gateway.ai.cloudflare.com");
const isNvidia = provider === "nvidia" || baseUrl.includes("integrate.api.nvidia.com");
const isAntLing = provider === "ant-ling" || baseUrl.includes("api.ant-ling.com");
const isNonStandard =
isNvidia ||
provider === "cerebras" ||
baseUrl.includes("cerebras.ai") ||
provider === "xai" ||
baseUrl.includes("api.x.ai") ||
isTogether ||
baseUrl.includes("chutes.ai") ||
baseUrl.includes("deepseek.com") ||
isZai ||
isMoonshot ||
provider === "opencode" ||
baseUrl.includes("opencode.ai") ||
isCloudflareWorkersAI ||
isCloudflareAiGateway ||
isAntLing;
const useMaxTokens =
baseUrl.includes("chutes.ai") || isMoonshot || isCloudflareAiGateway || isTogether || isNvidia || isAntLing;
const isGrok = provider === "xai" || baseUrl.includes("api.x.ai");
const isDeepSeek = provider === "deepseek" || baseUrl.includes("deepseek.com");
const isOpenRouterDeveloperRoleModel =
isOpenRouter && (model.id.startsWith("anthropic/") || model.id.startsWith("openai/"));
const cacheControlFormat = provider === "openrouter" && model.id.startsWith("anthropic/") ? "anthropic" : undefined;
return {
supportsStore: !isNonStandard,
supportsDeveloperRole: isOpenRouterDeveloperRoleModel || (!isNonStandard && !isOpenRouter),
supportsReasoningEffort:
!isGrok && !isZai && !isMoonshot && !isTogether && !isCloudflareAiGateway && !isNvidia && !isAntLing,
supportsUsageInStreaming: true,
maxTokensField: useMaxTokens ? "max_tokens" : "max_completion_tokens",
requiresToolResultName: false,
requiresAssistantAfterToolResult: false,
requiresThinkingAsText: false,
requiresReasoningContentOnAssistantMessages: isDeepSeek,
thinkingFormat: isDeepSeek
? "deepseek"
: isZai
? "zai"
: isTogether
? "together"
: isAntLing
? "ant-ling"
: isOpenRouter
? "openrouter"
: "openai",
openRouterRouting: {},
vercelGatewayRouting: {},
chatTemplateKwargs: {},
zaiToolStream: false,
supportsStrictMode: !isMoonshot && !isTogether && !isCloudflareAiGateway && !isNvidia,
cacheControlFormat,
sendSessionAffinityHeaders: false,
supportsLongCacheRetention: !(
isTogether ||
isCloudflareWorkersAI ||
isCloudflareAiGateway ||
isNvidia ||
isAntLing
),
};
}
const DEFAULT_COMPAT: ResolvedOpenAICompletionsCompat = {
supportsStore: true,
supportsDeveloperRole: true,
supportsReasoningEffort: true,
supportsUsageInStreaming: true,
maxTokensField: "max_completion_tokens",
requiresToolResultName: false,
requiresAssistantAfterToolResult: false,
requiresThinkingAsText: false,
requiresReasoningContentOnAssistantMessages: false,
thinkingFormat: "openai",
openRouterRouting: {},
vercelGatewayRouting: {},
chatTemplateKwargs: {},
zaiToolStream: false,
supportsStrictMode: true,
cacheControlFormat: undefined,
sendSessionAffinityHeaders: false,
supportsLongCacheRetention: true,
};
/**
* Get resolved compatibility settings for a model.
* Uses explicit model.compat if provided, otherwise auto-detects from provider/URL.
* Uses explicit generated/custom model.compat over OpenAI-standard defaults.
*/
function getCompat(model: Model<"openai-completions">): ResolvedOpenAICompletionsCompat {
const detected = detectCompat(model);
if (!model.compat) return detected;
if (!model.compat) return DEFAULT_COMPAT;
return {
supportsStore: model.compat.supportsStore ?? detected.supportsStore,
supportsDeveloperRole: model.compat.supportsDeveloperRole ?? detected.supportsDeveloperRole,
supportsReasoningEffort: model.compat.supportsReasoningEffort ?? detected.supportsReasoningEffort,
supportsUsageInStreaming: model.compat.supportsUsageInStreaming ?? detected.supportsUsageInStreaming,
maxTokensField: model.compat.maxTokensField ?? detected.maxTokensField,
requiresToolResultName: model.compat.requiresToolResultName ?? detected.requiresToolResultName,
supportsStore: model.compat.supportsStore ?? DEFAULT_COMPAT.supportsStore,
supportsDeveloperRole: model.compat.supportsDeveloperRole ?? DEFAULT_COMPAT.supportsDeveloperRole,
supportsReasoningEffort: model.compat.supportsReasoningEffort ?? DEFAULT_COMPAT.supportsReasoningEffort,
supportsUsageInStreaming: model.compat.supportsUsageInStreaming ?? DEFAULT_COMPAT.supportsUsageInStreaming,
maxTokensField: model.compat.maxTokensField ?? DEFAULT_COMPAT.maxTokensField,
requiresToolResultName: model.compat.requiresToolResultName ?? DEFAULT_COMPAT.requiresToolResultName,
requiresAssistantAfterToolResult:
model.compat.requiresAssistantAfterToolResult ?? detected.requiresAssistantAfterToolResult,
requiresThinkingAsText: model.compat.requiresThinkingAsText ?? detected.requiresThinkingAsText,
model.compat.requiresAssistantAfterToolResult ?? DEFAULT_COMPAT.requiresAssistantAfterToolResult,
requiresThinkingAsText: model.compat.requiresThinkingAsText ?? DEFAULT_COMPAT.requiresThinkingAsText,
requiresReasoningContentOnAssistantMessages:
model.compat.requiresReasoningContentOnAssistantMessages ??
detected.requiresReasoningContentOnAssistantMessages,
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,
sendSessionAffinityHeaders: model.compat.sendSessionAffinityHeaders ?? detected.sendSessionAffinityHeaders,
supportsLongCacheRetention: model.compat.supportsLongCacheRetention ?? detected.supportsLongCacheRetention,
DEFAULT_COMPAT.requiresReasoningContentOnAssistantMessages,
thinkingFormat: model.compat.thinkingFormat ?? DEFAULT_COMPAT.thinkingFormat,
openRouterRouting: model.compat.openRouterRouting ?? DEFAULT_COMPAT.openRouterRouting,
vercelGatewayRouting: model.compat.vercelGatewayRouting ?? DEFAULT_COMPAT.vercelGatewayRouting,
chatTemplateKwargs: model.compat.chatTemplateKwargs ?? DEFAULT_COMPAT.chatTemplateKwargs,
zaiToolStream: model.compat.zaiToolStream ?? DEFAULT_COMPAT.zaiToolStream,
supportsStrictMode: model.compat.supportsStrictMode ?? DEFAULT_COMPAT.supportsStrictMode,
cacheControlFormat: model.compat.cacheControlFormat ?? DEFAULT_COMPAT.cacheControlFormat,
sendSessionAffinityHeaders: model.compat.sendSessionAffinityHeaders ?? DEFAULT_COMPAT.sendSessionAffinityHeaders,
supportsLongCacheRetention: model.compat.supportsLongCacheRetention ?? DEFAULT_COMPAT.supportsLongCacheRetention,
};
}
packages/ai/src/api/openai-responses.ts
+24 -25
View File
@@ -9,6 +9,7 @@ import type {
Model,
OpenAIResponsesCompat,
ProviderEnv,
ProviderHeaders,
SimpleStreamOptions,
StreamFunction,
StreamOptions,
@@ -17,7 +18,6 @@ import type {
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";
import { convertResponsesMessages, convertResponsesTools, processResponsesStream } from "./openai-responses-shared.ts";
@@ -25,6 +25,21 @@ import { buildBaseOptions } from "./simple-options.ts";
const OPENAI_TOOL_CALL_PROVIDERS = new Set(["openai", "openai-codex", "opencode"]);
function hasHeader(headers: ProviderHeaders | undefined, name: string): boolean {
if (!headers) return false;
const expected = name.toLowerCase();
for (const [key, value] of Object.entries(headers)) {
if (key.toLowerCase() === expected && value !== null && value.trim().length > 0) return true;
}
return false;
}
function getClientApiKey(provider: string, apiKey: string | undefined, headers: ProviderHeaders | undefined): string {
if (apiKey) return apiKey;
if (hasHeader(headers, "authorization") || hasHeader(headers, "cf-aig-authorization")) return "unused";
throw new Error(`No API key for provider: ${provider}`);
}
/**
* Resolve cache retention preference.
* Defaults to "short" and uses PI_CACHE_RETENTION for backward compatibility.
@@ -109,13 +124,10 @@ export const stream: StreamFunction<"openai-responses", OpenAIResponsesOptions>
try {
// Create OpenAI client
const apiKey = options?.apiKey;
if (!apiKey) {
throw new Error(`No API key for provider: ${model.provider}`);
}
const apiKey = getClientApiKey(model.provider, options?.apiKey, options?.headers);
const cacheRetention = resolveCacheRetention(options?.cacheRetention, options?.env);
const cacheSessionId = cacheRetention === "none" ? undefined : options?.sessionId;
const client = createClient(model, context, apiKey, options?.headers, cacheSessionId, options?.env);
const client = createClient(model, context, apiKey, options?.headers, cacheSessionId);
let params = buildParams(model, context, options);
const nextParams = await options?.onPayload?.(params, model);
if (nextParams !== undefined) {
@@ -166,12 +178,9 @@ export const streamSimple: StreamFunction<"openai-responses", SimpleStreamOption
context: Context,
options?: SimpleStreamOptions,
): AssistantMessageEventStream => {
const apiKey = options?.apiKey;
if (!apiKey) {
throw new Error(`No API key for provider: ${model.provider}`);
}
getClientApiKey(model.provider, options?.apiKey, options?.headers);
const base = buildBaseOptions(model, options, apiKey);
const base = buildBaseOptions(model, options, options?.apiKey);
const clampedReasoning = options?.reasoning ? clampThinkingLevel(model, options.reasoning) : undefined;
const reasoningEffort = clampedReasoning === "off" ? undefined : clampedReasoning;
@@ -185,12 +194,11 @@ function createClient(
model: Model<"openai-responses">,
context: Context,
apiKey: string,
optionsHeaders?: Record<string, string>,
optionsHeaders?: ProviderHeaders,
sessionId?: string,
env?: ProviderEnv,
) {
const compat = getCompat(model);
const headers = { ...model.headers };
const headers: ProviderHeaders = { ...model.headers };
if (model.provider === "github-copilot") {
const hasImages = hasCopilotVisionInput(context.messages);
const copilotHeaders = buildCopilotDynamicHeaders({
@@ -212,20 +220,11 @@ function createClient(
Object.assign(headers, optionsHeaders);
}
const defaultHeaders =
model.provider === "cloudflare-ai-gateway"
? {
...headers,
Authorization: headers.Authorization ?? null,
"cf-aig-authorization": `Bearer ${apiKey}`,
}
: headers;
return new OpenAI({
apiKey,
baseURL: isCloudflareProvider(model.provider) ? resolveCloudflareBaseUrl(model, env) : model.baseUrl,
baseURL: model.baseUrl,
dangerouslyAllowBrowser: true,
defaultHeaders,
defaultHeaders: headers,
});
}
packages/ai/src/api/openrouter-images.ts
+4 -6
View File
@@ -13,9 +13,10 @@ import type {
ImagesFunction,
ImagesModel,
ImagesOptions,
ProviderHeaders,
TextContent,
} from "../types.ts";
import { headersToRecord } from "../utils/headers.ts";
import { headersToRecord, providerHeadersToRecord } from "../utils/headers.ts";
import { sanitizeSurrogates } from "../utils/sanitize-unicode.ts";
interface OpenRouterGeneratedImage {
@@ -106,16 +107,13 @@ export const generateImages: ImagesFunction<"openrouter-images", ImagesOptions>
function createClient(
model: ImagesModel<"openrouter-images">,
apiKey: string,
optionsHeaders?: Record<string, string>,
optionsHeaders?: ProviderHeaders,
): OpenAI {
return new OpenAI({
apiKey,
baseURL: model.baseUrl,
dangerouslyAllowBrowser: true,
defaultHeaders: {
...model.headers,
...optionsHeaders,
},
defaultHeaders: providerHeadersToRecord({ ...model.headers, ...optionsHeaders }),
});
}