workers-ai-provider 3.3.1 → 4.0.1

package/dist/gateway-provider.d.mts

@@ -48,6 +48,12 @@ interface GatewayFetchConfig {
  * A `fetch` that dispatches the wrapped provider's request through AI Gateway.
  * Detects the gateway provider id from the request URL (or uses `config.provider`),
  * strips the provider host to the endpoint path, and forwards the body verbatim.
+ *
+ * @deprecated Route the vendor's model with `createAI` from
+ * `agents/models/ai-sdk`: `ai(model)` for language models, `ai.routed(model)`
+ * for embedding, image, speech and transcription models, on AI Gateway's
+ * provider-native path. `vendorKey: "request"` forwards the vendor key, as
+ * `byok: true` does here.
  */
 declare function createGatewayFetch(config: GatewayFetchConfig): typeof globalThis.fetch;
 /**
@@ -66,6 +72,12 @@ declare function createGatewayFetch(config: GatewayFetchConfig): typeof globalTh
  * });
  * const model = openai("gpt-5");
  * ```
+ *
+ * @deprecated Route the vendor's model with `createAI` from
+ * `agents/models/ai-sdk`: `ai(model)` for language models, `ai.routed(model)`
+ * for embedding, image, speech and transcription models, on AI Gateway's
+ * provider-native path. `vendorKey: "request"` forwards the vendor key, as
+ * `byok: true` does here.
  */
 declare function createGatewayProvider<T>(factory: (opts: {
   apiKey?: string;

package/dist/gateway-provider.mjs

@@ -1,2 +1,2 @@
-import { n as createGatewayProvider, t as createGatewayFetch } from "./gateway-provider-CQU-v2IO.mjs";
+import { n as createGatewayProvider, t as createGatewayFetch } from "./gateway-provider-BQj8ZnKv.mjs";
 export { createGatewayFetch, createGatewayProvider };

package/dist/google.d.mts

@@ -1,4 +1,4 @@
-import { a as ProviderPlugin } from "./gateway-delegate-D_zkIp5r.mjs";
+import { a as ProviderPlugin } from "./gateway-delegate-DYjgGAa-.mjs";
 
 //#region src/google.d.ts
 /**
@@ -7,6 +7,10 @@ import { a as ProviderPlugin } from "./gateway-delegate-D_zkIp5r.mjs";
  * (Google AI Studio) and `"google-vertex/<model>"` slugs.
  *
  * Requires `@ai-sdk/google` (an optional peer dependency — install it yourself).
+ *
+ * @deprecated Pass the vendor's own model to `createAI` from
+ * `agents/models/ai-sdk` instead of a plugin plus a slug:
+ * `ai(createGoogle({ apiKey: "cloudflare" })("gemini-2.5-flash"))`.
  */
 declare const google: ProviderPlugin;
 //#endregion

package/dist/google.mjs

@@ -1,4 +1,4 @@
-import { createGoogleGenerativeAI } from "@ai-sdk/google";
+import { createGoogle } from "@ai-sdk/google";
 //#region src/google.ts
 /**
 * Google (Gemini) provider plugin for the gateway delegate. Pass to
@@ -6,10 +6,14 @@ import { createGoogleGenerativeAI } from "@ai-sdk/google";
 * (Google AI Studio) and `"google-vertex/<model>"` slugs.
 *
 * Requires `@ai-sdk/google` (an optional peer dependency — install it yourself).
+*
+* @deprecated Pass the vendor's own model to `createAI` from
+* `agents/models/ai-sdk` instead of a plugin plus a slug:
+* `ai(createGoogle({ apiKey: "cloudflare" })("gemini-2.5-flash"))`.
 */
 const google = {
 	wireFormat: "google",
-	create: ({ modelId, fetch, baseURL }) => createGoogleGenerativeAI({
+	create: ({ modelId, fetch, baseURL }) => createGoogle({
 		apiKey: "unused",
 		fetch,
 		...baseURL ? { baseURL } : {}

package/dist/google.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"google.mjs","names":[],"sources":["../src/google.ts"],"sourcesContent":["import { createGoogleGenerativeAI } from \"@ai-sdk/google\";\nimport type { ProviderPlugin } from \"./gateway-delegate\";\n\n/**\n * Google (Gemini) provider plugin for the gateway delegate. Pass to\n * `createGatewayDelegate({ providers: [google] })` to handle `\"google/<model>\"`\n * (Google AI Studio) and `\"google-vertex/<model>\"` slugs.\n *\n * Requires `@ai-sdk/google` (an optional peer dependency — install it yourself).\n */\nexport const google: ProviderPlugin = {\n\twireFormat: \"google\",\n\tcreate: ({ modelId, fetch, baseURL }) =>\n\t\t// apiKey is a placeholder — the gateway handles auth (unified billing / BYOK)\n\t\t// and the delegate strips the x-goog-api-key header on the gateway path.\n\t\tcreateGoogleGenerativeAI({ apiKey: \"unused\", fetch, ...(baseURL ? { baseURL } : {}) })(\n\t\t\tmodelId,\n\t\t),\n};\n"],"mappings":";;;;;;;;;AAUA,MAAa,SAAyB;CACrC,YAAY;CACZ,SAAS,EAAE,SAAS,OAAO,cAG1B,yBAAyB;EAAE,QAAQ;EAAU;EAAO,GAAI,UAAU,EAAE,QAAQ,IAAI,CAAC;CAAG,CAAC,CAAC,CACrF,OACD;AACF"}
+{"version":3,"file":"google.mjs","names":[],"sources":["../src/google.ts"],"sourcesContent":["import { createGoogle } from \"@ai-sdk/google\";\nimport type { ProviderPlugin } from \"./gateway-delegate\";\n\n/**\n * Google (Gemini) provider plugin for the gateway delegate. Pass to\n * `createGatewayDelegate({ providers: [google] })` to handle `\"google/<model>\"`\n * (Google AI Studio) and `\"google-vertex/<model>\"` slugs.\n *\n * Requires `@ai-sdk/google` (an optional peer dependency — install it yourself).\n *\n * @deprecated Pass the vendor's own model to `createAI` from\n * `agents/models/ai-sdk` instead of a plugin plus a slug:\n * `ai(createGoogle({ apiKey: \"cloudflare\" })(\"gemini-2.5-flash\"))`.\n */\nexport const google: ProviderPlugin = {\n\twireFormat: \"google\",\n\tcreate: ({ modelId, fetch, baseURL }) =>\n\t\t// apiKey is a placeholder — the gateway handles auth (unified billing / BYOK)\n\t\t// and the delegate strips the x-goog-api-key header on the gateway path.\n\t\tcreateGoogle({ apiKey: \"unused\", fetch, ...(baseURL ? { baseURL } : {}) })(modelId),\n};\n"],"mappings":";;;;;;;;;;;;;AAcA,MAAa,SAAyB;CACrC,YAAY;CACZ,SAAS,EAAE,SAAS,OAAO,cAG1B,aAAa;EAAE,QAAQ;EAAU;EAAO,GAAI,UAAU,EAAE,QAAQ,IAAI,CAAC;CAAG,CAAC,CAAC,CAAC,OAAO;AACpF"}

package/dist/index.d.mts

@@ -1,6 +1,6 @@
-import { C as createResumableStream, E as wireableProviders, S as WireFormat, T as findProviderBySlug, _ as GATEWAY_PROVIDERS, a as ProviderPlugin, b as ResumableStreamOptions, c as selectTransport, d as FallbackAttempt, f as GatewayErrorCode, g as Billing, h as WorkersAIGatewayError, i as ParsedSlug, l as FallbackLeg, m as WorkersAIFallbackError, n as DispatchInfo, o as Transport, p as GatewayErrorContext, r as FallbackOptions, s as parseSlug, t as DelegateCallOptions, u as createClientFallbackModel, v as GatewayDelegateError, w as detectProviderByUrl, x as ResumeExpiredPolicy, y as GatewayProviderInfo } from "./gateway-delegate-D_zkIp5r.mjs";
+import { C as createResumableStream, E as wireableProviders, S as WireFormat, T as findProviderBySlug, _ as GATEWAY_PROVIDERS, a as ProviderPlugin, b as ResumableStreamOptions, c as selectTransport, d as FallbackAttempt, f as GatewayErrorCode, g as Billing, h as WorkersAIGatewayError, i as ParsedSlug, l as FallbackLeg, m as WorkersAIFallbackError, n as DispatchInfo, o as Transport, p as GatewayErrorContext, r as FallbackOptions, s as parseSlug, t as DelegateCallOptions, u as createClientFallbackModel, v as GatewayDelegateError, w as detectProviderByUrl, x as ResumeExpiredPolicy, y as GatewayProviderInfo } from "./gateway-delegate-DYjgGAa-.mjs";
 import { GatewayFetchConfig, createGatewayFetch, createGatewayProvider } from "./gateway-provider.mjs";
-import { EmbeddingModelV3, EmbeddingModelV3CallOptions, EmbeddingModelV3Result, ImageModelV3, LanguageModelV3, RerankingModelV3, SpeechModelV3, TranscriptionModelV3 } from "@ai-sdk/provider";
+import { EmbeddingModelV4, EmbeddingModelV4CallOptions, EmbeddingModelV4Result, ImageModelV4, LanguageModelV4, RerankingModelV4, SpeechModelV4, TranscriptionModelV4 } from "@ai-sdk/provider";
 
 //#region src/aisearch-chat-settings.d.ts
 type AISearchChatSettings = {
@@ -65,8 +65,8 @@ type AISearchChatConfig = {
   binding: AutoRAG;
   gateway?: GatewayOptions;
 };
-declare class AISearchChatLanguageModel implements LanguageModelV3 {
-  readonly specificationVersion = "v3";
+declare class AISearchChatLanguageModel implements LanguageModelV4 {
+  readonly specificationVersion = "v4";
   readonly defaultObjectGenerationMode = "json";
   readonly supportedUrls: Record<string, RegExp[]> | PromiseLike<Record<string, RegExp[]>>;
   readonly modelId: TextGenerationModels;
@@ -80,8 +80,8 @@ declare class AISearchChatLanguageModel implements LanguageModelV3 {
    * Flattens the conversation into a single string for aiSearch.
    */
   private buildQuery;
-  doGenerate(options: Parameters<LanguageModelV3["doGenerate"]>[0]): Promise<Awaited<ReturnType<LanguageModelV3["doGenerate"]>>>;
-  doStream(options: Parameters<LanguageModelV3["doStream"]>[0]): Promise<Awaited<ReturnType<LanguageModelV3["doStream"]>>>;
+  doGenerate(options: Parameters<LanguageModelV4["doGenerate"]>[0]): Promise<Awaited<ReturnType<LanguageModelV4["doGenerate"]>>>;
+  doStream(options: Parameters<LanguageModelV4["doStream"]>[0]): Promise<Awaited<ReturnType<LanguageModelV4["doStream"]>>>;
 }
 //#endregion
 //#region src/workersai-embedding-model.d.ts
@@ -99,8 +99,8 @@ type WorkersAIEmbeddingSettings = {
    */
   [key: string]: unknown;
 };
-declare class WorkersAIEmbeddingModel implements EmbeddingModelV3 {
-  readonly specificationVersion = "v3";
+declare class WorkersAIEmbeddingModel implements EmbeddingModelV4 {
+  readonly specificationVersion = "v4";
   readonly modelId: EmbeddingModels;
   private readonly config;
   private readonly settings;
@@ -111,7 +111,7 @@ declare class WorkersAIEmbeddingModel implements EmbeddingModelV3 {
   doEmbed({
     values,
     abortSignal
-  }: EmbeddingModelV3CallOptions): Promise<EmbeddingModelV3Result>;
+  }: EmbeddingModelV4CallOptions): Promise<EmbeddingModelV4Result>;
 }
 //#endregion
 //#region src/workersai-chat-settings.d.ts
@@ -163,8 +163,8 @@ type WorkersAIChatConfig = {
   gateway?: GatewayOptions; /** True when using a real Workers AI binding (not the REST shim). */
   isBinding: boolean;
 };
-declare class WorkersAIChatLanguageModel implements LanguageModelV3 {
-  readonly specificationVersion = "v3";
+declare class WorkersAIChatLanguageModel implements LanguageModelV4 {
+  readonly specificationVersion = "v4";
   readonly defaultObjectGenerationMode = "json";
   readonly supportedUrls: Record<string, RegExp[]> | PromiseLike<Record<string, RegExp[]>>;
   readonly modelId: TextGenerationModels;
@@ -190,6 +190,11 @@ declare class WorkersAIChatLanguageModel implements LanguageModelV3 {
    *
    * `reasoning_effort: null` is a valid value ("disable reasoning"), so we
    * check `!== undefined` rather than truthiness.
+   *
+   * The unified `reasoning` call option (spec v4) is mapped onto
+   * `reasoning_effort` between the two: an explicit per-call
+   * `providerOptions["workers-ai"].reasoning_effort` wins, then the unified
+   * option, then settings.
    */
   private buildRunInputs;
   /**
@@ -211,8 +216,8 @@ declare class WorkersAIChatLanguageModel implements LanguageModelV3 {
    * place so callers can observe the reinterpretation.
    */
   private extractContent;
-  doGenerate(options: Parameters<LanguageModelV3["doGenerate"]>[0]): Promise<Awaited<ReturnType<LanguageModelV3["doGenerate"]>>>;
-  doStream(options: Parameters<LanguageModelV3["doStream"]>[0]): Promise<Awaited<ReturnType<LanguageModelV3["doStream"]>>>;
+  doGenerate(options: Parameters<LanguageModelV4["doGenerate"]>[0]): Promise<Awaited<ReturnType<LanguageModelV4["doGenerate"]>>>;
+  doStream(options: Parameters<LanguageModelV4["doStream"]>[0]): Promise<Awaited<ReturnType<LanguageModelV4["doStream"]>>>;
 }
 //#endregion
 //#region src/workersai-image-settings.d.ts
@@ -226,11 +231,11 @@ type WorkersAIImageConfig = {
   binding: Ai;
   gateway?: GatewayOptions;
 };
-declare class WorkersAIImageModel implements ImageModelV3 {
+declare class WorkersAIImageModel implements ImageModelV4 {
   readonly modelId: ImageGenerationModels;
   readonly settings: WorkersAIImageSettings;
   readonly config: WorkersAIImageConfig;
-  readonly specificationVersion = "v3";
+  readonly specificationVersion = "v4";
   get maxImagesPerCall(): number;
   get provider(): string;
   constructor(modelId: ImageGenerationModels, settings: WorkersAIImageSettings, config: WorkersAIImageConfig);
@@ -241,7 +246,7 @@ declare class WorkersAIImageModel implements ImageModelV3 {
     aspectRatio,
     seed,
     abortSignal
-  }: Parameters<ImageModelV3["doGenerate"]>[0]): Promise<Awaited<ReturnType<ImageModelV3["doGenerate"]>>>;
+  }: Parameters<ImageModelV4["doGenerate"]>[0]): Promise<Awaited<ReturnType<ImageModelV4["doGenerate"]>>>;
 }
 //#endregion
 //#region src/workersai-transcription-settings.d.ts
@@ -288,20 +293,20 @@ type WorkersAITranscriptionConfig = {
   credentials?: CreateRunConfig;
 };
 /**
- * Workers AI transcription model implementing the AI SDK's `TranscriptionModelV3` interface.
+ * Workers AI transcription model implementing the AI SDK's `TranscriptionModelV4` interface.
  *
  * Supports:
  * - Whisper models (`@cf/openai/whisper`, `whisper-tiny-en`, `whisper-large-v3-turbo`)
  * - Deepgram Nova-3 (`@cf/deepgram/nova-3`) — uses a different input/output format
  */
-declare class WorkersAITranscriptionModel implements TranscriptionModelV3 {
+declare class WorkersAITranscriptionModel implements TranscriptionModelV4 {
   readonly modelId: TranscriptionModels;
   readonly settings: WorkersAITranscriptionSettings;
   readonly config: WorkersAITranscriptionConfig;
-  readonly specificationVersion = "v3";
+  readonly specificationVersion = "v4";
   get provider(): string;
   constructor(modelId: TranscriptionModels, settings: WorkersAITranscriptionSettings, config: WorkersAITranscriptionConfig);
-  doGenerate(options: Parameters<TranscriptionModelV3["doGenerate"]>[0]): Promise<Awaited<ReturnType<TranscriptionModelV3["doGenerate"]>>>;
+  doGenerate(options: Parameters<TranscriptionModelV4["doGenerate"]>[0]): Promise<Awaited<ReturnType<TranscriptionModelV4["doGenerate"]>>>;
   private runWhisper;
   private normalizeWhisperResponse;
   private runNova3;
@@ -318,19 +323,19 @@ type WorkersAISpeechConfig = {
   gateway?: GatewayOptions;
 };
 /**
- * Workers AI speech (text-to-speech) model implementing the AI SDK's `SpeechModelV3` interface.
+ * Workers AI speech (text-to-speech) model implementing the AI SDK's `SpeechModelV4` interface.
  *
  * Currently supports Deepgram Aura-1 (`@cf/deepgram/aura-1`).
  * The model accepts `{ text, voice?, speed? }` and returns raw audio bytes.
  */
-declare class WorkersAISpeechModel implements SpeechModelV3 {
+declare class WorkersAISpeechModel implements SpeechModelV4 {
   readonly modelId: SpeechModels;
   readonly settings: WorkersAISpeechSettings;
   readonly config: WorkersAISpeechConfig;
-  readonly specificationVersion = "v3";
+  readonly specificationVersion = "v4";
   get provider(): string;
   constructor(modelId: SpeechModels, settings: WorkersAISpeechSettings, config: WorkersAISpeechConfig);
-  doGenerate(options: Parameters<SpeechModelV3["doGenerate"]>[0]): Promise<Awaited<ReturnType<SpeechModelV3["doGenerate"]>>>;
+  doGenerate(options: Parameters<SpeechModelV4["doGenerate"]>[0]): Promise<Awaited<ReturnType<SpeechModelV4["doGenerate"]>>>;
 }
 //#endregion
 //#region src/workersai-reranking-settings.d.ts
@@ -343,7 +348,7 @@ type WorkersAIRerankingConfig = {
   gateway?: GatewayOptions;
 };
 /**
- * Workers AI reranking model implementing the AI SDK's `RerankingModelV3` interface.
+ * Workers AI reranking model implementing the AI SDK's `RerankingModelV4` interface.
  *
  * Supports BGE reranker models (`@cf/baai/bge-reranker-base`, `bge-reranker-v2-m3`).
  *
@@ -351,14 +356,14 @@ type WorkersAIRerankingConfig = {
  * - Input: `{ query, contexts: [{ text }], top_k? }`
  * - Output: `{ response: [{ id, score }] }`
  */
-declare class WorkersAIRerankingModel implements RerankingModelV3 {
+declare class WorkersAIRerankingModel implements RerankingModelV4 {
   readonly modelId: RerankingModels;
   readonly settings: WorkersAIRerankingSettings;
   readonly config: WorkersAIRerankingConfig;
-  readonly specificationVersion = "v3";
+  readonly specificationVersion = "v4";
   get provider(): string;
   constructor(modelId: RerankingModels, settings: WorkersAIRerankingSettings, config: WorkersAIRerankingConfig);
-  doRerank(options: Parameters<RerankingModelV3["doRerank"]>[0]): Promise<Awaited<ReturnType<RerankingModelV3["doRerank"]>>>;
+  doRerank(options: Parameters<RerankingModelV4["doRerank"]>[0]): Promise<Awaited<ReturnType<RerankingModelV4["doRerank"]>>>;
 }
 //#endregion
 //#region src/autorag-chat-language-model.d.ts
@@ -479,11 +484,40 @@ interface WorkersAI {
 }
 /**
  * Create a Workers AI provider instance.
+ *
+ * @deprecated Use `createAI` from `agents/models/ai-sdk`, with the binding or
+ * with an account id and API token (credentials mode's `apiKey`):
+ *
+ * ```ts
+ * import { createAI } from "agents/models/ai-sdk";
+ * const ai = createAI({ binding: env.AI }); // or { accountId, apiToken }
+ * ai("@cf/zai-org/glm-4.7-flash"); // was workersai("@cf/…")
+ * ai("anthropic/claude-opus-4.8"); // was "anthropic/…" + providers; the catalog's spelling
+ * ai("dynamic/support"); // dynamic routes, as before
+ * ai(createAnthropic({ apiKey: "cloudflare" })("claude-opus-4-8")); // a vendor's own model
+ * ai.embedding("@cf/baai/bge-base-en-v1.5"); // .image/.speech/.transcription/.reranking alike
+ * ```
+ *
+ * `vendorKey`, `byokAlias` and `zdr` cover BYOK, and through the binding a
+ * catalog stream resumes after a dropped connection. For AI Search, use
+ * `ai-search-provider`.
  */
 declare function createWorkersAI(options: WorkersAISettings): WorkersAI;
+/**
+ * @deprecated Use `createAISearchNamespace` from `ai-search-provider`, over the
+ * `ai_search_namespaces` binding that replaces `env.AI.autorag()`:
+ * `createAISearchNamespace({ binding: env.AI_SEARCH }).get("my-instance").chat()`.
+ * @see https://developers.cloudflare.com/ai-search/api/migration/workers-binding/
+ */
 type AISearchSettings = {
   binding: AutoRAG;
 };
+/**
+ * @deprecated Use `createAISearchNamespace` from `ai-search-provider`, over the
+ * `ai_search_namespaces` binding that replaces `env.AI.autorag()`:
+ * `createAISearchNamespace({ binding: env.AI_SEARCH }).get("my-instance").chat()`.
+ * @see https://developers.cloudflare.com/ai-search/api/migration/workers-binding/
+ */
 interface AISearchProvider {
   (settings?: AISearchChatSettings): AISearchChatLanguageModel;
   /**
@@ -495,7 +529,11 @@ interface AISearchProvider {
  * Create an AI Search provider instance.
  *
  * AI Search (formerly AutoRAG) is Cloudflare's managed search service.
- * @see https://developers.cloudflare.com/ai-search/
+ *
+ * @deprecated Use `createAISearchNamespace` from `ai-search-provider`, over the
+ * `ai_search_namespaces` binding that replaces `env.AI.autorag()`:
+ * `createAISearchNamespace({ binding: env.AI_SEARCH }).get("my-instance").chat()`.
+ * @see https://developers.cloudflare.com/ai-search/api/migration/workers-binding/
  */
 declare function createAISearch(options: AISearchSettings, /** @internal */
 
@@ -511,8 +549,9 @@ type AutoRAGSettings = AISearchSettings;
  */
 type AutoRAGProvider = AISearchProvider;
 /**
- * @deprecated Use `createAISearch` instead. AutoRAG has been renamed to AI Search.
- * @see https://developers.cloudflare.com/ai-search/
+ * @deprecated Use `createAISearchNamespace` from `ai-search-provider`. AutoRAG
+ * has been renamed to AI Search.
+ * @see https://developers.cloudflare.com/ai-search/api/migration/workers-binding/
  */
 declare function createAutoRAG(options: AISearchSettings): AISearchProvider;
 //#endregion