@mariozechner/pi-ai 0.49.2 → 0.50.0

package/dist/providers/register-builtins.js

@@ -0,0 +1,63 @@
+import { clearApiProviders, registerApiProvider } from "../api-registry.js";
+import { streamBedrock, streamSimpleBedrock } from "./amazon-bedrock.js";
+import { streamAnthropic, streamSimpleAnthropic } from "./anthropic.js";
+import { streamAzureOpenAIResponses, streamSimpleAzureOpenAIResponses } from "./azure-openai-responses.js";
+import { streamGoogle, streamSimpleGoogle } from "./google.js";
+import { streamGoogleGeminiCli, streamSimpleGoogleGeminiCli } from "./google-gemini-cli.js";
+import { streamGoogleVertex, streamSimpleGoogleVertex } from "./google-vertex.js";
+import { streamOpenAICodexResponses, streamSimpleOpenAICodexResponses } from "./openai-codex-responses.js";
+import { streamOpenAICompletions, streamSimpleOpenAICompletions } from "./openai-completions.js";
+import { streamOpenAIResponses, streamSimpleOpenAIResponses } from "./openai-responses.js";
+export function registerBuiltInApiProviders() {
+    registerApiProvider({
+        api: "anthropic-messages",
+        stream: streamAnthropic,
+        streamSimple: streamSimpleAnthropic,
+    });
+    registerApiProvider({
+        api: "openai-completions",
+        stream: streamOpenAICompletions,
+        streamSimple: streamSimpleOpenAICompletions,
+    });
+    registerApiProvider({
+        api: "openai-responses",
+        stream: streamOpenAIResponses,
+        streamSimple: streamSimpleOpenAIResponses,
+    });
+    registerApiProvider({
+        api: "azure-openai-responses",
+        stream: streamAzureOpenAIResponses,
+        streamSimple: streamSimpleAzureOpenAIResponses,
+    });
+    registerApiProvider({
+        api: "openai-codex-responses",
+        stream: streamOpenAICodexResponses,
+        streamSimple: streamSimpleOpenAICodexResponses,
+    });
+    registerApiProvider({
+        api: "google-generative-ai",
+        stream: streamGoogle,
+        streamSimple: streamSimpleGoogle,
+    });
+    registerApiProvider({
+        api: "google-gemini-cli",
+        stream: streamGoogleGeminiCli,
+        streamSimple: streamSimpleGoogleGeminiCli,
+    });
+    registerApiProvider({
+        api: "google-vertex",
+        stream: streamGoogleVertex,
+        streamSimple: streamSimpleGoogleVertex,
+    });
+    registerApiProvider({
+        api: "bedrock-converse-stream",
+        stream: streamBedrock,
+        streamSimple: streamSimpleBedrock,
+    });
+}
+export function resetApiProviders() {
+    clearApiProviders();
+    registerBuiltInApiProviders();
+}
+registerBuiltInApiProviders();
+//# sourceMappingURL=register-builtins.js.map

package/dist/providers/register-builtins.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"register-builtins.js","sourceRoot":"","sources":["../../src/providers/register-builtins.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,iBAAiB,EAAE,mBAAmB,EAAE,MAAM,oBAAoB,CAAC;AAC5E,OAAO,EAAE,aAAa,EAAE,mBAAmB,EAAE,MAAM,qBAAqB,CAAC;AACzE,OAAO,EAAE,eAAe,EAAE,qBAAqB,EAAE,MAAM,gBAAgB,CAAC;AACxE,OAAO,EAAE,0BAA0B,EAAE,gCAAgC,EAAE,MAAM,6BAA6B,CAAC;AAC3G,OAAO,EAAE,YAAY,EAAE,kBAAkB,EAAE,MAAM,aAAa,CAAC;AAC/D,OAAO,EAAE,qBAAqB,EAAE,2BAA2B,EAAE,MAAM,wBAAwB,CAAC;AAC5F,OAAO,EAAE,kBAAkB,EAAE,wBAAwB,EAAE,MAAM,oBAAoB,CAAC;AAClF,OAAO,EAAE,0BAA0B,EAAE,gCAAgC,EAAE,MAAM,6BAA6B,CAAC;AAC3G,OAAO,EAAE,uBAAuB,EAAE,6BAA6B,EAAE,MAAM,yBAAyB,CAAC;AACjG,OAAO,EAAE,qBAAqB,EAAE,2BAA2B,EAAE,MAAM,uBAAuB,CAAC;AAE3F,MAAM,UAAU,2BAA2B,GAAS;IACnD,mBAAmB,CAAC;QACnB,GAAG,EAAE,oBAAoB;QACzB,MAAM,EAAE,eAAe;QACvB,YAAY,EAAE,qBAAqB;KACnC,CAAC,CAAC;IAEH,mBAAmB,CAAC;QACnB,GAAG,EAAE,oBAAoB;QACzB,MAAM,EAAE,uBAAuB;QAC/B,YAAY,EAAE,6BAA6B;KAC3C,CAAC,CAAC;IAEH,mBAAmB,CAAC;QACnB,GAAG,EAAE,kBAAkB;QACvB,MAAM,EAAE,qBAAqB;QAC7B,YAAY,EAAE,2BAA2B;KACzC,CAAC,CAAC;IAEH,mBAAmB,CAAC;QACnB,GAAG,EAAE,wBAAwB;QAC7B,MAAM,EAAE,0BAA0B;QAClC,YAAY,EAAE,gCAAgC;KAC9C,CAAC,CAAC;IAEH,mBAAmB,CAAC;QACnB,GAAG,EAAE,wBAAwB;QAC7B,MAAM,EAAE,0BAA0B;QAClC,YAAY,EAAE,gCAAgC;KAC9C,CAAC,CAAC;IAEH,mBAAmB,CAAC;QACnB,GAAG,EAAE,sBAAsB;QAC3B,MAAM,EAAE,YAAY;QACpB,YAAY,EAAE,kBAAkB;KAChC,CAAC,CAAC;IAEH,mBAAmB,CAAC;QACnB,GAAG,EAAE,mBAAmB;QACxB,MAAM,EAAE,qBAAqB;QAC7B,YAAY,EAAE,2BAA2B;KACzC,CAAC,CAAC;IAEH,mBAAmB,CAAC;QACnB,GAAG,EAAE,eAAe;QACpB,MAAM,EAAE,kBAAkB;QAC1B,YAAY,EAAE,wBAAwB;KACtC,CAAC,CAAC;IAEH,mBAAmB,CAAC;QACnB,GAAG,EAAE,yBAAyB;QAC9B,MAAM,EAAE,aAAa;QACrB,YAAY,EAAE,mBAAmB;KACjC,CAAC,CAAC;AAAA,CACH;AAED,MAAM,UAAU,iBAAiB,GAAS;IACzC,iBAAiB,EAAE,CAAC;IACpB,2BAA2B,EAAE,CAAC;AAAA,CAC9B;AAED,2BAA2B,EAAE,CAAC","sourcesContent":["import { clearApiProviders, registerApiProvider } from \"../api-registry.js\";\nimport { streamBedrock, streamSimpleBedrock } from \"./amazon-bedrock.js\";\nimport { streamAnthropic, streamSimpleAnthropic } from \"./anthropic.js\";\nimport { streamAzureOpenAIResponses, streamSimpleAzureOpenAIResponses } from \"./azure-openai-responses.js\";\nimport { streamGoogle, streamSimpleGoogle } from \"./google.js\";\nimport { streamGoogleGeminiCli, streamSimpleGoogleGeminiCli } from \"./google-gemini-cli.js\";\nimport { streamGoogleVertex, streamSimpleGoogleVertex } from \"./google-vertex.js\";\nimport { streamOpenAICodexResponses, streamSimpleOpenAICodexResponses } from \"./openai-codex-responses.js\";\nimport { streamOpenAICompletions, streamSimpleOpenAICompletions } from \"./openai-completions.js\";\nimport { streamOpenAIResponses, streamSimpleOpenAIResponses } from \"./openai-responses.js\";\n\nexport function registerBuiltInApiProviders(): void {\n\tregisterApiProvider({\n\t\tapi: \"anthropic-messages\",\n\t\tstream: streamAnthropic,\n\t\tstreamSimple: streamSimpleAnthropic,\n\t});\n\n\tregisterApiProvider({\n\t\tapi: \"openai-completions\",\n\t\tstream: streamOpenAICompletions,\n\t\tstreamSimple: streamSimpleOpenAICompletions,\n\t});\n\n\tregisterApiProvider({\n\t\tapi: \"openai-responses\",\n\t\tstream: streamOpenAIResponses,\n\t\tstreamSimple: streamSimpleOpenAIResponses,\n\t});\n\n\tregisterApiProvider({\n\t\tapi: \"azure-openai-responses\",\n\t\tstream: streamAzureOpenAIResponses,\n\t\tstreamSimple: streamSimpleAzureOpenAIResponses,\n\t});\n\n\tregisterApiProvider({\n\t\tapi: \"openai-codex-responses\",\n\t\tstream: streamOpenAICodexResponses,\n\t\tstreamSimple: streamSimpleOpenAICodexResponses,\n\t});\n\n\tregisterApiProvider({\n\t\tapi: \"google-generative-ai\",\n\t\tstream: streamGoogle,\n\t\tstreamSimple: streamSimpleGoogle,\n\t});\n\n\tregisterApiProvider({\n\t\tapi: \"google-gemini-cli\",\n\t\tstream: streamGoogleGeminiCli,\n\t\tstreamSimple: streamSimpleGoogleGeminiCli,\n\t});\n\n\tregisterApiProvider({\n\t\tapi: \"google-vertex\",\n\t\tstream: streamGoogleVertex,\n\t\tstreamSimple: streamSimpleGoogleVertex,\n\t});\n\n\tregisterApiProvider({\n\t\tapi: \"bedrock-converse-stream\",\n\t\tstream: streamBedrock,\n\t\tstreamSimple: streamSimpleBedrock,\n\t});\n}\n\nexport function resetApiProviders(): void {\n\tclearApiProviders();\n\tregisterBuiltInApiProviders();\n}\n\nregisterBuiltInApiProviders();\n"]}

package/dist/providers/simple-options.d.ts

@@ -0,0 +1,8 @@
+import type { Api, Model, SimpleStreamOptions, StreamOptions, ThinkingBudgets, ThinkingLevel } from "../types.js";
+export declare function buildBaseOptions(model: Model<Api>, options?: SimpleStreamOptions, apiKey?: string): StreamOptions;
+export declare function clampReasoning(effort: ThinkingLevel | undefined): Exclude<ThinkingLevel, "xhigh"> | undefined;
+export declare function adjustMaxTokensForThinking(baseMaxTokens: number, modelMaxTokens: number, reasoningLevel: ThinkingLevel, customBudgets?: ThinkingBudgets): {
+    maxTokens: number;
+    thinkingBudget: number;
+};
+//# sourceMappingURL=simple-options.d.ts.map

package/dist/providers/simple-options.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"simple-options.d.ts","sourceRoot":"","sources":["../../src/providers/simple-options.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,GAAG,EAAE,KAAK,EAAE,mBAAmB,EAAE,aAAa,EAAE,eAAe,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AAElH,wBAAgB,gBAAgB,CAAC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EAAE,OAAO,CAAC,EAAE,mBAAmB,EAAE,MAAM,CAAC,EAAE,MAAM,GAAG,aAAa,CAUjH;AAED,wBAAgB,cAAc,CAAC,MAAM,EAAE,aAAa,GAAG,SAAS,GAAG,OAAO,CAAC,aAAa,EAAE,OAAO,CAAC,GAAG,SAAS,CAE7G;AAED,wBAAgB,0BAA0B,CACzC,aAAa,EAAE,MAAM,EACrB,cAAc,EAAE,MAAM,EACtB,cAAc,EAAE,aAAa,EAC7B,aAAa,CAAC,EAAE,eAAe,GAC7B;IAAE,SAAS,EAAE,MAAM,CAAC;IAAC,cAAc,EAAE,MAAM,CAAA;CAAE,CAmB/C","sourcesContent":["import type { Api, Model, SimpleStreamOptions, StreamOptions, ThinkingBudgets, ThinkingLevel } from \"../types.js\";\n\nexport function buildBaseOptions(model: Model<Api>, options?: SimpleStreamOptions, apiKey?: string): StreamOptions {\n\treturn {\n\t\ttemperature: options?.temperature,\n\t\tmaxTokens: options?.maxTokens || Math.min(model.maxTokens, 32000),\n\t\tsignal: options?.signal,\n\t\tapiKey: apiKey || options?.apiKey,\n\t\tsessionId: options?.sessionId,\n\t\theaders: options?.headers,\n\t\tonPayload: options?.onPayload,\n\t};\n}\n\nexport function clampReasoning(effort: ThinkingLevel | undefined): Exclude<ThinkingLevel, \"xhigh\"> | undefined {\n\treturn effort === \"xhigh\" ? \"high\" : effort;\n}\n\nexport function adjustMaxTokensForThinking(\n\tbaseMaxTokens: number,\n\tmodelMaxTokens: number,\n\treasoningLevel: ThinkingLevel,\n\tcustomBudgets?: ThinkingBudgets,\n): { maxTokens: number; thinkingBudget: number } {\n\tconst defaultBudgets: ThinkingBudgets = {\n\t\tminimal: 1024,\n\t\tlow: 2048,\n\t\tmedium: 8192,\n\t\thigh: 16384,\n\t};\n\tconst budgets = { ...defaultBudgets, ...customBudgets };\n\n\tconst minOutputTokens = 1024;\n\tconst level = clampReasoning(reasoningLevel)!;\n\tlet thinkingBudget = budgets[level]!;\n\tconst maxTokens = Math.min(baseMaxTokens + thinkingBudget, modelMaxTokens);\n\n\tif (maxTokens <= thinkingBudget) {\n\t\tthinkingBudget = Math.max(0, maxTokens - minOutputTokens);\n\t}\n\n\treturn { maxTokens, thinkingBudget };\n}\n"]}

package/dist/providers/simple-options.js

@@ -0,0 +1,32 @@
+export function buildBaseOptions(model, options, apiKey) {
+    return {
+        temperature: options?.temperature,
+        maxTokens: options?.maxTokens || Math.min(model.maxTokens, 32000),
+        signal: options?.signal,
+        apiKey: apiKey || options?.apiKey,
+        sessionId: options?.sessionId,
+        headers: options?.headers,
+        onPayload: options?.onPayload,
+    };
+}
+export function clampReasoning(effort) {
+    return effort === "xhigh" ? "high" : effort;
+}
+export function adjustMaxTokensForThinking(baseMaxTokens, modelMaxTokens, reasoningLevel, customBudgets) {
+    const defaultBudgets = {
+        minimal: 1024,
+        low: 2048,
+        medium: 8192,
+        high: 16384,
+    };
+    const budgets = { ...defaultBudgets, ...customBudgets };
+    const minOutputTokens = 1024;
+    const level = clampReasoning(reasoningLevel);
+    let thinkingBudget = budgets[level];
+    const maxTokens = Math.min(baseMaxTokens + thinkingBudget, modelMaxTokens);
+    if (maxTokens <= thinkingBudget) {
+        thinkingBudget = Math.max(0, maxTokens - minOutputTokens);
+    }
+    return { maxTokens, thinkingBudget };
+}
+//# sourceMappingURL=simple-options.js.map

package/dist/providers/simple-options.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"simple-options.js","sourceRoot":"","sources":["../../src/providers/simple-options.ts"],"names":[],"mappings":"AAEA,MAAM,UAAU,gBAAgB,CAAC,KAAiB,EAAE,OAA6B,EAAE,MAAe,EAAiB;IAClH,OAAO;QACN,WAAW,EAAE,OAAO,EAAE,WAAW;QACjC,SAAS,EAAE,OAAO,EAAE,SAAS,IAAI,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,SAAS,EAAE,KAAK,CAAC;QACjE,MAAM,EAAE,OAAO,EAAE,MAAM;QACvB,MAAM,EAAE,MAAM,IAAI,OAAO,EAAE,MAAM;QACjC,SAAS,EAAE,OAAO,EAAE,SAAS;QAC7B,OAAO,EAAE,OAAO,EAAE,OAAO;QACzB,SAAS,EAAE,OAAO,EAAE,SAAS;KAC7B,CAAC;AAAA,CACF;AAED,MAAM,UAAU,cAAc,CAAC,MAAiC,EAA+C;IAC9G,OAAO,MAAM,KAAK,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC;AAAA,CAC5C;AAED,MAAM,UAAU,0BAA0B,CACzC,aAAqB,EACrB,cAAsB,EACtB,cAA6B,EAC7B,aAA+B,EACiB;IAChD,MAAM,cAAc,GAAoB;QACvC,OAAO,EAAE,IAAI;QACb,GAAG,EAAE,IAAI;QACT,MAAM,EAAE,IAAI;QACZ,IAAI,EAAE,KAAK;KACX,CAAC;IACF,MAAM,OAAO,GAAG,EAAE,GAAG,cAAc,EAAE,GAAG,aAAa,EAAE,CAAC;IAExD,MAAM,eAAe,GAAG,IAAI,CAAC;IAC7B,MAAM,KAAK,GAAG,cAAc,CAAC,cAAc,CAAE,CAAC;IAC9C,IAAI,cAAc,GAAG,OAAO,CAAC,KAAK,CAAE,CAAC;IACrC,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,CAAC,aAAa,GAAG,cAAc,EAAE,cAAc,CAAC,CAAC;IAE3E,IAAI,SAAS,IAAI,cAAc,EAAE,CAAC;QACjC,cAAc,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,SAAS,GAAG,eAAe,CAAC,CAAC;IAC3D,CAAC;IAED,OAAO,EAAE,SAAS,EAAE,cAAc,EAAE,CAAC;AAAA,CACrC","sourcesContent":["import type { Api, Model, SimpleStreamOptions, StreamOptions, ThinkingBudgets, ThinkingLevel } from \"../types.js\";\n\nexport function buildBaseOptions(model: Model<Api>, options?: SimpleStreamOptions, apiKey?: string): StreamOptions {\n\treturn {\n\t\ttemperature: options?.temperature,\n\t\tmaxTokens: options?.maxTokens || Math.min(model.maxTokens, 32000),\n\t\tsignal: options?.signal,\n\t\tapiKey: apiKey || options?.apiKey,\n\t\tsessionId: options?.sessionId,\n\t\theaders: options?.headers,\n\t\tonPayload: options?.onPayload,\n\t};\n}\n\nexport function clampReasoning(effort: ThinkingLevel | undefined): Exclude<ThinkingLevel, \"xhigh\"> | undefined {\n\treturn effort === \"xhigh\" ? \"high\" : effort;\n}\n\nexport function adjustMaxTokensForThinking(\n\tbaseMaxTokens: number,\n\tmodelMaxTokens: number,\n\treasoningLevel: ThinkingLevel,\n\tcustomBudgets?: ThinkingBudgets,\n): { maxTokens: number; thinkingBudget: number } {\n\tconst defaultBudgets: ThinkingBudgets = {\n\t\tminimal: 1024,\n\t\tlow: 2048,\n\t\tmedium: 8192,\n\t\thigh: 16384,\n\t};\n\tconst budgets = { ...defaultBudgets, ...customBudgets };\n\n\tconst minOutputTokens = 1024;\n\tconst level = clampReasoning(reasoningLevel)!;\n\tlet thinkingBudget = budgets[level]!;\n\tconst maxTokens = Math.min(baseMaxTokens + thinkingBudget, modelMaxTokens);\n\n\tif (maxTokens <= thinkingBudget) {\n\t\tthinkingBudget = Math.max(0, maxTokens - minOutputTokens);\n\t}\n\n\treturn { maxTokens, thinkingBudget };\n}\n"]}

package/dist/stream.d.ts

@@ -1,13 +1,8 @@
-import type { Api, AssistantMessage, AssistantMessageEventStream, Context, KnownProvider, Model, OptionsForApi, SimpleStreamOptions } from "./types.js";
-/**
- * Get API key for provider from known environment variables, e.g. OPENAI_API_KEY.
- *
- * Will not return API keys for providers that require OAuth tokens.
- */
-export declare function getEnvApiKey(provider: KnownProvider): string | undefined;
-export declare function getEnvApiKey(provider: string): string | undefined;
-export declare function stream<TApi extends Api>(model: Model<TApi>, context: Context, options?: OptionsForApi<TApi>): AssistantMessageEventStream;
-export declare function complete<TApi extends Api>(model: Model<TApi>, context: Context, options?: OptionsForApi<TApi>): Promise<AssistantMessage>;
+import "./providers/register-builtins.js";
+import type { Api, AssistantMessage, AssistantMessageEventStream, Context, Model, ProviderStreamOptions, SimpleStreamOptions } from "./types.js";
+export { getEnvApiKey } from "./env-api-keys.js";
+export declare function stream<TApi extends Api>(model: Model<TApi>, context: Context, options?: ProviderStreamOptions): AssistantMessageEventStream;
+export declare function complete<TApi extends Api>(model: Model<TApi>, context: Context, options?: ProviderStreamOptions): Promise<AssistantMessage>;
 export declare function streamSimple<TApi extends Api>(model: Model<TApi>, context: Context, options?: SimpleStreamOptions): AssistantMessageEventStream;
 export declare function completeSimple<TApi extends Api>(model: Model<TApi>, context: Context, options?: SimpleStreamOptions): Promise<AssistantMessage>;
 //# sourceMappingURL=stream.d.ts.map

package/dist/stream.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"stream.d.ts","sourceRoot":"","sources":["../src/stream.ts"],"names":[],"mappings":"AAgBA,OAAO,KAAK,EACX,GAAG,EACH,gBAAgB,EAChB,2BAA2B,EAC3B,OAAO,EACP,aAAa,EACb,KAAK,EACL,aAAa,EACb,mBAAmB,EAGnB,MAAM,YAAY,CAAC;AAoBpB;;;;GAIG;AACH,wBAAgB,YAAY,CAAC,QAAQ,EAAE,aAAa,GAAG,MAAM,GAAG,SAAS,CAAC;AAC1E,wBAAgB,YAAY,CAAC,QAAQ,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS,CAAC;AA+DnE,wBAAgB,MAAM,CAAC,IAAI,SAAS,GAAG,EACtC,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,EAClB,OAAO,EAAE,OAAO,EAChB,OAAO,CAAC,EAAE,aAAa,CAAC,IAAI,CAAC,GAC3B,2BAA2B,CA6C7B;AAED,wBAAsB,QAAQ,CAAC,IAAI,SAAS,GAAG,EAC9C,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,EAClB,OAAO,EAAE,OAAO,EAChB,OAAO,CAAC,EAAE,aAAa,CAAC,IAAI,CAAC,GAC3B,OAAO,CAAC,gBAAgB,CAAC,CAG3B;AAED,wBAAgB,YAAY,CAAC,IAAI,SAAS,GAAG,EAC5C,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,EAClB,OAAO,EAAE,OAAO,EAChB,OAAO,CAAC,EAAE,mBAAmB,GAC3B,2BAA2B,CAkB7B;AAED,wBAAsB,cAAc,CAAC,IAAI,SAAS,GAAG,EACpD,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,EAClB,OAAO,EAAE,OAAO,EAChB,OAAO,CAAC,EAAE,mBAAmB,GAC3B,OAAO,CAAC,gBAAgB,CAAC,CAG3B","sourcesContent":["import { existsSync } from \"node:fs\";\nimport { homedir } from \"node:os\";\nimport { join } from \"node:path\";\nimport { supportsXhigh } from \"./models.js\";\nimport { type BedrockOptions, streamBedrock } from \"./providers/amazon-bedrock.js\";\nimport { type AnthropicOptions, streamAnthropic } from \"./providers/anthropic.js\";\nimport { type GoogleOptions, streamGoogle } from \"./providers/google.js\";\nimport {\n\ttype GoogleGeminiCliOptions,\n\ttype GoogleThinkingLevel,\n\tstreamGoogleGeminiCli,\n} from \"./providers/google-gemini-cli.js\";\nimport { type GoogleVertexOptions, streamGoogleVertex } from \"./providers/google-vertex.js\";\nimport { type OpenAICodexResponsesOptions, streamOpenAICodexResponses } from \"./providers/openai-codex-responses.js\";\nimport { type OpenAICompletionsOptions, streamOpenAICompletions } from \"./providers/openai-completions.js\";\nimport { type OpenAIResponsesOptions, streamOpenAIResponses } from \"./providers/openai-responses.js\";\nimport type {\n\tApi,\n\tAssistantMessage,\n\tAssistantMessageEventStream,\n\tContext,\n\tKnownProvider,\n\tModel,\n\tOptionsForApi,\n\tSimpleStreamOptions,\n\tThinkingBudgets,\n\tThinkingLevel,\n} from \"./types.js\";\n\nlet cachedVertexAdcCredentialsExists: boolean | null = null;\n\nfunction hasVertexAdcCredentials(): boolean {\n\tif (cachedVertexAdcCredentialsExists === null) {\n\t\t// Check GOOGLE_APPLICATION_CREDENTIALS env var first (standard way)\n\t\tconst gacPath = process.env.GOOGLE_APPLICATION_CREDENTIALS;\n\t\tif (gacPath) {\n\t\t\tcachedVertexAdcCredentialsExists = existsSync(gacPath);\n\t\t} else {\n\t\t\t// Fall back to default ADC path (lazy evaluation)\n\t\t\tcachedVertexAdcCredentialsExists = existsSync(\n\t\t\t\tjoin(homedir(), \".config\", \"gcloud\", \"application_default_credentials.json\"),\n\t\t\t);\n\t\t}\n\t}\n\treturn cachedVertexAdcCredentialsExists;\n}\n\n/**\n * Get API key for provider from known environment variables, e.g. OPENAI_API_KEY.\n *\n * Will not return API keys for providers that require OAuth tokens.\n */\nexport function getEnvApiKey(provider: KnownProvider): string | undefined;\nexport function getEnvApiKey(provider: string): string | undefined;\nexport function getEnvApiKey(provider: any): string | undefined {\n\t// Fall back to environment variables\n\tif (provider === \"github-copilot\") {\n\t\treturn process.env.COPILOT_GITHUB_TOKEN || process.env.GH_TOKEN || process.env.GITHUB_TOKEN;\n\t}\n\n\t// ANTHROPIC_OAUTH_TOKEN takes precedence over ANTHROPIC_API_KEY\n\tif (provider === \"anthropic\") {\n\t\treturn process.env.ANTHROPIC_OAUTH_TOKEN || process.env.ANTHROPIC_API_KEY;\n\t}\n\n\t// Vertex AI uses Application Default Credentials, not API keys.\n\t// Auth is configured via `gcloud auth application-default login`.\n\tif (provider === \"google-vertex\") {\n\t\tconst hasCredentials = hasVertexAdcCredentials();\n\t\tconst hasProject = !!(process.env.GOOGLE_CLOUD_PROJECT || process.env.GCLOUD_PROJECT);\n\t\tconst hasLocation = !!process.env.GOOGLE_CLOUD_LOCATION;\n\n\t\tif (hasCredentials && hasProject && hasLocation) {\n\t\t\treturn \"<authenticated>\";\n\t\t}\n\t}\n\n\tif (provider === \"amazon-bedrock\") {\n\t\t// Amazon Bedrock supports multiple credential sources:\n\t\t// 1. AWS_PROFILE - named profile from ~/.aws/credentials\n\t\t// 2. AWS_ACCESS_KEY_ID + AWS_SECRET_ACCESS_KEY - standard IAM keys\n\t\t// 3. AWS_BEARER_TOKEN_BEDROCK - Bedrock API keys (bearer token)\n\t\t// 4. AWS_CONTAINER_CREDENTIALS_RELATIVE_URI - ECS task roles\n\t\t// 5. AWS_CONTAINER_CREDENTIALS_FULL_URI - ECS task roles (full URI)\n\t\t// 6. AWS_WEB_IDENTITY_TOKEN_FILE - IRSA (IAM Roles for Service Accounts)\n\t\tif (\n\t\t\tprocess.env.AWS_PROFILE ||\n\t\t\t(process.env.AWS_ACCESS_KEY_ID && process.env.AWS_SECRET_ACCESS_KEY) ||\n\t\t\tprocess.env.AWS_BEARER_TOKEN_BEDROCK ||\n\t\t\tprocess.env.AWS_CONTAINER_CREDENTIALS_RELATIVE_URI ||\n\t\t\tprocess.env.AWS_CONTAINER_CREDENTIALS_FULL_URI ||\n\t\t\tprocess.env.AWS_WEB_IDENTITY_TOKEN_FILE\n\t\t) {\n\t\t\treturn \"<authenticated>\";\n\t\t}\n\t}\n\n\tconst envMap: Record<string, string> = {\n\t\topenai: \"OPENAI_API_KEY\",\n\t\tgoogle: \"GEMINI_API_KEY\",\n\t\tgroq: \"GROQ_API_KEY\",\n\t\tcerebras: \"CEREBRAS_API_KEY\",\n\t\txai: \"XAI_API_KEY\",\n\t\topenrouter: \"OPENROUTER_API_KEY\",\n\t\t\"vercel-ai-gateway\": \"AI_GATEWAY_API_KEY\",\n\t\tzai: \"ZAI_API_KEY\",\n\t\tmistral: \"MISTRAL_API_KEY\",\n\t\tminimax: \"MINIMAX_API_KEY\",\n\t\t\"minimax-cn\": \"MINIMAX_CN_API_KEY\",\n\t\topencode: \"OPENCODE_API_KEY\",\n\t};\n\n\tconst envVar = envMap[provider];\n\treturn envVar ? process.env[envVar] : undefined;\n}\n\nexport function stream<TApi extends Api>(\n\tmodel: Model<TApi>,\n\tcontext: Context,\n\toptions?: OptionsForApi<TApi>,\n): AssistantMessageEventStream {\n\t// Vertex AI uses Application Default Credentials, not API keys\n\tif (model.api === \"google-vertex\") {\n\t\treturn streamGoogleVertex(model as Model<\"google-vertex\">, context, options as GoogleVertexOptions);\n\t} else if (model.api === \"bedrock-converse-stream\") {\n\t\t// Bedrock doesn't have any API keys instead it sources credentials from standard AWS env variables or from given AWS profile.\n\t\treturn streamBedrock(model as Model<\"bedrock-converse-stream\">, context, (options || {}) as BedrockOptions);\n\t}\n\n\tconst apiKey = options?.apiKey || getEnvApiKey(model.provider);\n\tif (!apiKey) {\n\t\tthrow new Error(`No API key for provider: ${model.provider}`);\n\t}\n\tconst providerOptions = { ...options, apiKey };\n\n\tconst api: Api = model.api;\n\tswitch (api) {\n\t\tcase \"anthropic-messages\":\n\t\t\treturn streamAnthropic(model as Model<\"anthropic-messages\">, context, providerOptions);\n\n\t\tcase \"openai-completions\":\n\t\t\treturn streamOpenAICompletions(model as Model<\"openai-completions\">, context, providerOptions as any);\n\n\t\tcase \"openai-responses\":\n\t\t\treturn streamOpenAIResponses(model as Model<\"openai-responses\">, context, providerOptions as any);\n\n\t\tcase \"openai-codex-responses\":\n\t\t\treturn streamOpenAICodexResponses(model as Model<\"openai-codex-responses\">, context, providerOptions as any);\n\n\t\tcase \"google-generative-ai\":\n\t\t\treturn streamGoogle(model as Model<\"google-generative-ai\">, context, providerOptions);\n\n\t\tcase \"google-gemini-cli\":\n\t\t\treturn streamGoogleGeminiCli(\n\t\t\t\tmodel as Model<\"google-gemini-cli\">,\n\t\t\t\tcontext,\n\t\t\t\tproviderOptions as GoogleGeminiCliOptions,\n\t\t\t);\n\n\t\tdefault: {\n\t\t\t// This should never be reached if all Api cases are handled\n\t\t\tconst _exhaustive: never = api;\n\t\t\tthrow new Error(`Unhandled API: ${_exhaustive}`);\n\t\t}\n\t}\n}\n\nexport async function complete<TApi extends Api>(\n\tmodel: Model<TApi>,\n\tcontext: Context,\n\toptions?: OptionsForApi<TApi>,\n): Promise<AssistantMessage> {\n\tconst s = stream(model, context, options);\n\treturn s.result();\n}\n\nexport function streamSimple<TApi extends Api>(\n\tmodel: Model<TApi>,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream {\n\t// Vertex AI uses Application Default Credentials, not API keys\n\tif (model.api === \"google-vertex\") {\n\t\tconst providerOptions = mapOptionsForApi(model, options, undefined);\n\t\treturn stream(model, context, providerOptions);\n\t} else if (model.api === \"bedrock-converse-stream\") {\n\t\t// Bedrock doesn't have any API keys instead it sources credentials from standard AWS env variables or from given AWS profile.\n\t\tconst providerOptions = mapOptionsForApi(model, options, undefined);\n\t\treturn stream(model, context, providerOptions);\n\t}\n\n\tconst apiKey = options?.apiKey || getEnvApiKey(model.provider);\n\tif (!apiKey) {\n\t\tthrow new Error(`No API key for provider: ${model.provider}`);\n\t}\n\n\tconst providerOptions = mapOptionsForApi(model, options, apiKey);\n\treturn stream(model, context, providerOptions);\n}\n\nexport async function completeSimple<TApi extends Api>(\n\tmodel: Model<TApi>,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): Promise<AssistantMessage> {\n\tconst s = streamSimple(model, context, options);\n\treturn s.result();\n}\n\nfunction mapOptionsForApi<TApi extends Api>(\n\tmodel: Model<TApi>,\n\toptions?: SimpleStreamOptions,\n\tapiKey?: string,\n): OptionsForApi<TApi> {\n\tconst base = {\n\t\ttemperature: options?.temperature,\n\t\tmaxTokens: options?.maxTokens || Math.min(model.maxTokens, 32000),\n\t\tsignal: options?.signal,\n\t\tapiKey: apiKey || options?.apiKey,\n\t\tsessionId: options?.sessionId,\n\t};\n\n\t// Helper to clamp xhigh to high for providers that don't support it\n\tconst clampReasoning = (effort: ThinkingLevel | undefined) => (effort === \"xhigh\" ? \"high\" : effort);\n\n\t/**\n\t * Adjust maxTokens to account for thinking budget.\n\t * APIs like Anthropic and Bedrock require max_tokens > thinking.budget_tokens.\n\t * Returns { adjustedMaxTokens, adjustedThinkingBudget }\n\t */\n\tconst adjustMaxTokensForThinking = (\n\t\tbaseMaxTokens: number,\n\t\tmodelMaxTokens: number,\n\t\treasoningLevel: ThinkingLevel,\n\t\tcustomBudgets?: ThinkingBudgets,\n\t): { maxTokens: number; thinkingBudget: number } => {\n\t\tconst defaultBudgets: ThinkingBudgets = {\n\t\t\tminimal: 1024,\n\t\t\tlow: 2048,\n\t\t\tmedium: 8192,\n\t\t\thigh: 16384,\n\t\t};\n\t\tconst budgets = { ...defaultBudgets, ...customBudgets };\n\n\t\tconst minOutputTokens = 1024;\n\t\tconst level = clampReasoning(reasoningLevel)!;\n\t\tlet thinkingBudget = budgets[level]!;\n\t\t// Caller's maxTokens is the desired output; add thinking budget on top, capped at model limit\n\t\tconst maxTokens = Math.min(baseMaxTokens + thinkingBudget, modelMaxTokens);\n\n\t\t// If not enough room for thinking + output, reduce thinking budget\n\t\tif (maxTokens <= thinkingBudget) {\n\t\t\tthinkingBudget = Math.max(0, maxTokens - minOutputTokens);\n\t\t}\n\n\t\treturn { maxTokens, thinkingBudget };\n\t};\n\n\tswitch (model.api) {\n\t\tcase \"anthropic-messages\": {\n\t\t\t// Explicitly disable thinking when reasoning is not specified\n\t\t\tif (!options?.reasoning) {\n\t\t\t\treturn { ...base, thinkingEnabled: false } satisfies AnthropicOptions;\n\t\t\t}\n\n\t\t\t// Claude requires max_tokens > thinking.budget_tokens\n\t\t\t// So we need to ensure maxTokens accounts for both thinking and output\n\t\t\tconst adjusted = adjustMaxTokensForThinking(\n\t\t\t\tbase.maxTokens || 0,\n\t\t\t\tmodel.maxTokens,\n\t\t\t\toptions.reasoning,\n\t\t\t\toptions?.thinkingBudgets,\n\t\t\t);\n\n\t\t\treturn {\n\t\t\t\t...base,\n\t\t\t\tmaxTokens: adjusted.maxTokens,\n\t\t\t\tthinkingEnabled: true,\n\t\t\t\tthinkingBudgetTokens: adjusted.thinkingBudget,\n\t\t\t} satisfies AnthropicOptions;\n\t\t}\n\n\t\tcase \"bedrock-converse-stream\": {\n\t\t\t// Explicitly disable thinking when reasoning is not specified\n\t\t\tif (!options?.reasoning) {\n\t\t\t\treturn { ...base, reasoning: undefined } satisfies BedrockOptions;\n\t\t\t}\n\n\t\t\t// Claude requires max_tokens > thinking.budget_tokens (same as Anthropic direct API)\n\t\t\t// So we need to ensure maxTokens accounts for both thinking and output\n\t\t\tif (model.id.includes(\"anthropic.claude\") || model.id.includes(\"anthropic/claude\")) {\n\t\t\t\tconst adjusted = adjustMaxTokensForThinking(\n\t\t\t\t\tbase.maxTokens || 0,\n\t\t\t\t\tmodel.maxTokens,\n\t\t\t\t\toptions.reasoning,\n\t\t\t\t\toptions?.thinkingBudgets,\n\t\t\t\t);\n\n\t\t\t\treturn {\n\t\t\t\t\t...base,\n\t\t\t\t\tmaxTokens: adjusted.maxTokens,\n\t\t\t\t\treasoning: options.reasoning,\n\t\t\t\t\tthinkingBudgets: {\n\t\t\t\t\t\t...(options?.thinkingBudgets || {}),\n\t\t\t\t\t\t[clampReasoning(options.reasoning)!]: adjusted.thinkingBudget,\n\t\t\t\t\t},\n\t\t\t\t} satisfies BedrockOptions;\n\t\t\t}\n\n\t\t\t// Non-Claude models - pass through\n\t\t\treturn {\n\t\t\t\t...base,\n\t\t\t\treasoning: options?.reasoning,\n\t\t\t\tthinkingBudgets: options?.thinkingBudgets,\n\t\t\t} satisfies BedrockOptions;\n\t\t}\n\n\t\tcase \"openai-completions\":\n\t\t\treturn {\n\t\t\t\t...base,\n\t\t\t\treasoningEffort: supportsXhigh(model) ? options?.reasoning : clampReasoning(options?.reasoning),\n\t\t\t} satisfies OpenAICompletionsOptions;\n\n\t\tcase \"openai-responses\":\n\t\t\treturn {\n\t\t\t\t...base,\n\t\t\t\treasoningEffort: supportsXhigh(model) ? options?.reasoning : clampReasoning(options?.reasoning),\n\t\t\t} satisfies OpenAIResponsesOptions;\n\n\t\tcase \"openai-codex-responses\":\n\t\t\treturn {\n\t\t\t\t...base,\n\t\t\t\treasoningEffort: supportsXhigh(model) ? options?.reasoning : clampReasoning(options?.reasoning),\n\t\t\t} satisfies OpenAICodexResponsesOptions;\n\n\t\tcase \"google-generative-ai\": {\n\t\t\t// Explicitly disable thinking when reasoning is not specified\n\t\t\t// This is needed because Gemini has \"dynamic thinking\" enabled by default\n\t\t\tif (!options?.reasoning) {\n\t\t\t\treturn { ...base, thinking: { enabled: false } } satisfies GoogleOptions;\n\t\t\t}\n\n\t\t\tconst googleModel = model as Model<\"google-generative-ai\">;\n\t\t\tconst effort = clampReasoning(options.reasoning)!;\n\n\t\t\t// Gemini 3 models use thinkingLevel exclusively instead of thinkingBudget.\n\t\t\t// https://ai.google.dev/gemini-api/docs/thinking#set-budget\n\t\t\tif (isGemini3ProModel(googleModel) || isGemini3FlashModel(googleModel)) {\n\t\t\t\treturn {\n\t\t\t\t\t...base,\n\t\t\t\t\tthinking: {\n\t\t\t\t\t\tenabled: true,\n\t\t\t\t\t\tlevel: getGemini3ThinkingLevel(effort, googleModel),\n\t\t\t\t\t},\n\t\t\t\t} satisfies GoogleOptions;\n\t\t\t}\n\n\t\t\treturn {\n\t\t\t\t...base,\n\t\t\t\tthinking: {\n\t\t\t\t\tenabled: true,\n\t\t\t\t\tbudgetTokens: getGoogleBudget(googleModel, effort, options?.thinkingBudgets),\n\t\t\t\t},\n\t\t\t} satisfies GoogleOptions;\n\t\t}\n\n\t\tcase \"google-gemini-cli\": {\n\t\t\tif (!options?.reasoning) {\n\t\t\t\treturn { ...base, thinking: { enabled: false } } satisfies GoogleGeminiCliOptions;\n\t\t\t}\n\n\t\t\tconst effort = clampReasoning(options.reasoning)!;\n\n\t\t\t// Gemini 3 models use thinkingLevel instead of thinkingBudget\n\t\t\tif (model.id.includes(\"3-pro\") || model.id.includes(\"3-flash\")) {\n\t\t\t\treturn {\n\t\t\t\t\t...base,\n\t\t\t\t\tthinking: {\n\t\t\t\t\t\tenabled: true,\n\t\t\t\t\t\tlevel: getGeminiCliThinkingLevel(effort, model.id),\n\t\t\t\t\t},\n\t\t\t\t} satisfies GoogleGeminiCliOptions;\n\t\t\t}\n\n\t\t\t// Models using thinkingBudget (Gemini 2.x, Claude via Antigravity)\n\t\t\t// Claude requires max_tokens > thinking.budget_tokens\n\t\t\t// So we need to ensure maxTokens accounts for both thinking and output\n\t\t\tconst defaultBudgets: ThinkingBudgets = {\n\t\t\t\tminimal: 1024,\n\t\t\t\tlow: 2048,\n\t\t\t\tmedium: 8192,\n\t\t\t\thigh: 16384,\n\t\t\t};\n\t\t\tconst budgets = { ...defaultBudgets, ...options?.thinkingBudgets };\n\n\t\t\tconst minOutputTokens = 1024;\n\t\t\tlet thinkingBudget = budgets[effort]!;\n\t\t\t// Caller's maxTokens is the desired output; add thinking budget on top, capped at model limit\n\t\t\tconst maxTokens = Math.min((base.maxTokens || 0) + thinkingBudget, model.maxTokens);\n\n\t\t\t// If not enough room for thinking + output, reduce thinking budget\n\t\t\tif (maxTokens <= thinkingBudget) {\n\t\t\t\tthinkingBudget = Math.max(0, maxTokens - minOutputTokens);\n\t\t\t}\n\n\t\t\treturn {\n\t\t\t\t...base,\n\t\t\t\tmaxTokens,\n\t\t\t\tthinking: {\n\t\t\t\t\tenabled: true,\n\t\t\t\t\tbudgetTokens: thinkingBudget,\n\t\t\t\t},\n\t\t\t} satisfies GoogleGeminiCliOptions;\n\t\t}\n\n\t\tcase \"google-vertex\": {\n\t\t\t// Explicitly disable thinking when reasoning is not specified\n\t\t\tif (!options?.reasoning) {\n\t\t\t\treturn { ...base, thinking: { enabled: false } } satisfies GoogleVertexOptions;\n\t\t\t}\n\n\t\t\tconst vertexModel = model as Model<\"google-vertex\">;\n\t\t\tconst effort = clampReasoning(options.reasoning)!;\n\t\t\tconst geminiModel = vertexModel as unknown as Model<\"google-generative-ai\">;\n\n\t\t\tif (isGemini3ProModel(geminiModel) || isGemini3FlashModel(geminiModel)) {\n\t\t\t\treturn {\n\t\t\t\t\t...base,\n\t\t\t\t\tthinking: {\n\t\t\t\t\t\tenabled: true,\n\t\t\t\t\t\tlevel: getGemini3ThinkingLevel(effort, geminiModel),\n\t\t\t\t\t},\n\t\t\t\t} satisfies GoogleVertexOptions;\n\t\t\t}\n\n\t\t\treturn {\n\t\t\t\t...base,\n\t\t\t\tthinking: {\n\t\t\t\t\tenabled: true,\n\t\t\t\t\tbudgetTokens: getGoogleBudget(geminiModel, effort, options?.thinkingBudgets),\n\t\t\t\t},\n\t\t\t} satisfies GoogleVertexOptions;\n\t\t}\n\n\t\tdefault: {\n\t\t\t// Exhaustiveness check\n\t\t\tconst _exhaustive: never = model.api;\n\t\t\tthrow new Error(`Unhandled API in mapOptionsForApi: ${_exhaustive}`);\n\t\t}\n\t}\n}\n\ntype ClampedThinkingLevel = Exclude<ThinkingLevel, \"xhigh\">;\n\nfunction isGemini3ProModel(model: Model<\"google-generative-ai\">): boolean {\n\t// Covers gemini-3-pro, gemini-3-pro-preview, and possible other prefixed ids in the future\n\treturn model.id.includes(\"3-pro\");\n}\n\nfunction isGemini3FlashModel(model: Model<\"google-generative-ai\">): boolean {\n\t// Covers gemini-3-flash, gemini-3-flash-preview, and possible other prefixed ids in the future\n\treturn model.id.includes(\"3-flash\");\n}\n\nfunction getGemini3ThinkingLevel(\n\teffort: ClampedThinkingLevel,\n\tmodel: Model<\"google-generative-ai\">,\n): GoogleThinkingLevel {\n\tif (isGemini3ProModel(model)) {\n\t\t// Gemini 3 Pro only supports LOW/HIGH (for now)\n\t\tswitch (effort) {\n\t\t\tcase \"minimal\":\n\t\t\tcase \"low\":\n\t\t\t\treturn \"LOW\";\n\t\t\tcase \"medium\":\n\t\t\tcase \"high\":\n\t\t\t\treturn \"HIGH\";\n\t\t}\n\t}\n\t// Gemini 3 Flash supports all four levels\n\tswitch (effort) {\n\t\tcase \"minimal\":\n\t\t\treturn \"MINIMAL\";\n\t\tcase \"low\":\n\t\t\treturn \"LOW\";\n\t\tcase \"medium\":\n\t\t\treturn \"MEDIUM\";\n\t\tcase \"high\":\n\t\t\treturn \"HIGH\";\n\t}\n}\n\nfunction getGeminiCliThinkingLevel(effort: ClampedThinkingLevel, modelId: string): GoogleThinkingLevel {\n\tif (modelId.includes(\"3-pro\")) {\n\t\t// Gemini 3 Pro only supports LOW/HIGH (for now)\n\t\tswitch (effort) {\n\t\t\tcase \"minimal\":\n\t\t\tcase \"low\":\n\t\t\t\treturn \"LOW\";\n\t\t\tcase \"medium\":\n\t\t\tcase \"high\":\n\t\t\t\treturn \"HIGH\";\n\t\t}\n\t}\n\t// Gemini 3 Flash supports all four levels\n\tswitch (effort) {\n\t\tcase \"minimal\":\n\t\t\treturn \"MINIMAL\";\n\t\tcase \"low\":\n\t\t\treturn \"LOW\";\n\t\tcase \"medium\":\n\t\t\treturn \"MEDIUM\";\n\t\tcase \"high\":\n\t\t\treturn \"HIGH\";\n\t}\n}\n\nfunction getGoogleBudget(\n\tmodel: Model<\"google-generative-ai\">,\n\teffort: ClampedThinkingLevel,\n\tcustomBudgets?: ThinkingBudgets,\n): number {\n\t// Custom budgets take precedence if provided for this level\n\tif (customBudgets?.[effort] !== undefined) {\n\t\treturn customBudgets[effort]!;\n\t}\n\n\t// See https://ai.google.dev/gemini-api/docs/thinking#set-budget\n\tif (model.id.includes(\"2.5-pro\")) {\n\t\tconst budgets: Record<ClampedThinkingLevel, number> = {\n\t\t\tminimal: 128,\n\t\t\tlow: 2048,\n\t\t\tmedium: 8192,\n\t\t\thigh: 32768,\n\t\t};\n\t\treturn budgets[effort];\n\t}\n\n\tif (model.id.includes(\"2.5-flash\")) {\n\t\t// Covers 2.5-flash-lite as well\n\t\tconst budgets: Record<ClampedThinkingLevel, number> = {\n\t\t\tminimal: 128,\n\t\t\tlow: 2048,\n\t\t\tmedium: 8192,\n\t\t\thigh: 24576,\n\t\t};\n\t\treturn budgets[effort];\n\t}\n\n\t// Unknown model - use dynamic\n\treturn -1;\n}\n"]}
+{"version":3,"file":"stream.d.ts","sourceRoot":"","sources":["../src/stream.ts"],"names":[],"mappings":"AAAA,OAAO,kCAAkC,CAAC;AAG1C,OAAO,KAAK,EACX,GAAG,EACH,gBAAgB,EAChB,2BAA2B,EAC3B,OAAO,EACP,KAAK,EACL,qBAAqB,EACrB,mBAAmB,EAEnB,MAAM,YAAY,CAAC;AAEpB,OAAO,EAAE,YAAY,EAAE,MAAM,mBAAmB,CAAC;AAmBjD,wBAAgB,MAAM,CAAC,IAAI,SAAS,GAAG,EACtC,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,EAClB,OAAO,EAAE,OAAO,EAChB,OAAO,CAAC,EAAE,qBAAqB,GAC7B,2BAA2B,CAG7B;AAED,wBAAsB,QAAQ,CAAC,IAAI,SAAS,GAAG,EAC9C,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,EAClB,OAAO,EAAE,OAAO,EAChB,OAAO,CAAC,EAAE,qBAAqB,GAC7B,OAAO,CAAC,gBAAgB,CAAC,CAG3B;AAED,wBAAgB,YAAY,CAAC,IAAI,SAAS,GAAG,EAC5C,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,EAClB,OAAO,EAAE,OAAO,EAChB,OAAO,CAAC,EAAE,mBAAmB,GAC3B,2BAA2B,CAG7B;AAED,wBAAsB,cAAc,CAAC,IAAI,SAAS,GAAG,EACpD,KAAK,EAAE,KAAK,CAAC,IAAI,CAAC,EAClB,OAAO,EAAE,OAAO,EAChB,OAAO,CAAC,EAAE,mBAAmB,GAC3B,OAAO,CAAC,gBAAgB,CAAC,CAG3B","sourcesContent":["import \"./providers/register-builtins.js\";\n\nimport { getApiProvider } from \"./api-registry.js\";\nimport type {\n\tApi,\n\tAssistantMessage,\n\tAssistantMessageEventStream,\n\tContext,\n\tModel,\n\tProviderStreamOptions,\n\tSimpleStreamOptions,\n\tStreamOptions,\n} from \"./types.js\";\n\nexport { getEnvApiKey } from \"./env-api-keys.js\";\n\n// Set up http proxy according to env variables for `fetch` based SDKs in Node.js.\n// Bun has builtin support for this.\nif (typeof process !== \"undefined\" && process.versions?.node) {\n\timport(\"undici\").then((m) => {\n\t\tconst { EnvHttpProxyAgent, setGlobalDispatcher } = m;\n\t\tsetGlobalDispatcher(new EnvHttpProxyAgent());\n\t});\n}\n\nfunction resolveApiProvider(api: Api) {\n\tconst provider = getApiProvider(api);\n\tif (!provider) {\n\t\tthrow new Error(`No API provider registered for api: ${api}`);\n\t}\n\treturn provider;\n}\n\nexport function stream<TApi extends Api>(\n\tmodel: Model<TApi>,\n\tcontext: Context,\n\toptions?: ProviderStreamOptions,\n): AssistantMessageEventStream {\n\tconst provider = resolveApiProvider(model.api);\n\treturn provider.stream(model, context, options as StreamOptions);\n}\n\nexport async function complete<TApi extends Api>(\n\tmodel: Model<TApi>,\n\tcontext: Context,\n\toptions?: ProviderStreamOptions,\n): Promise<AssistantMessage> {\n\tconst s = stream(model, context, options);\n\treturn s.result();\n}\n\nexport function streamSimple<TApi extends Api>(\n\tmodel: Model<TApi>,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream {\n\tconst provider = resolveApiProvider(model.api);\n\treturn provider.streamSimple(model, context, options);\n}\n\nexport async function completeSimple<TApi extends Api>(\n\tmodel: Model<TApi>,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): Promise<AssistantMessage> {\n\tconst s = streamSimple(model, context, options);\n\treturn s.result();\n}\n"]}

package/dist/stream.js

@@ -1,418 +1,35 @@
-import { existsSync } from "node:fs";
-import { homedir } from "node:os";
-import { join } from "node:path";
-import { supportsXhigh } from "./models.js";
-import { streamBedrock } from "./providers/amazon-bedrock.js";
-import { streamAnthropic } from "./providers/anthropic.js";
-import { streamGoogle } from "./providers/google.js";
-import { streamGoogleGeminiCli, } from "./providers/google-gemini-cli.js";
-import { streamGoogleVertex } from "./providers/google-vertex.js";
-import { streamOpenAICodexResponses } from "./providers/openai-codex-responses.js";
-import { streamOpenAICompletions } from "./providers/openai-completions.js";
-import { streamOpenAIResponses } from "./providers/openai-responses.js";
-let cachedVertexAdcCredentialsExists = null;
-function hasVertexAdcCredentials() {
-    if (cachedVertexAdcCredentialsExists === null) {
-        // Check GOOGLE_APPLICATION_CREDENTIALS env var first (standard way)
-        const gacPath = process.env.GOOGLE_APPLICATION_CREDENTIALS;
-        if (gacPath) {
-            cachedVertexAdcCredentialsExists = existsSync(gacPath);
-        }
-        else {
-            // Fall back to default ADC path (lazy evaluation)
-            cachedVertexAdcCredentialsExists = existsSync(join(homedir(), ".config", "gcloud", "application_default_credentials.json"));
-        }
-    }
-    return cachedVertexAdcCredentialsExists;
-}
-export function getEnvApiKey(provider) {
-    // Fall back to environment variables
-    if (provider === "github-copilot") {
-        return process.env.COPILOT_GITHUB_TOKEN || process.env.GH_TOKEN || process.env.GITHUB_TOKEN;
-    }
-    // ANTHROPIC_OAUTH_TOKEN takes precedence over ANTHROPIC_API_KEY
-    if (provider === "anthropic") {
-        return process.env.ANTHROPIC_OAUTH_TOKEN || process.env.ANTHROPIC_API_KEY;
-    }
-    // Vertex AI uses Application Default Credentials, not API keys.
-    // Auth is configured via `gcloud auth application-default login`.
-    if (provider === "google-vertex") {
-        const hasCredentials = hasVertexAdcCredentials();
-        const hasProject = !!(process.env.GOOGLE_CLOUD_PROJECT || process.env.GCLOUD_PROJECT);
-        const hasLocation = !!process.env.GOOGLE_CLOUD_LOCATION;
-        if (hasCredentials && hasProject && hasLocation) {
-            return "<authenticated>";
-        }
-    }
-    if (provider === "amazon-bedrock") {
-        // Amazon Bedrock supports multiple credential sources:
-        // 1. AWS_PROFILE - named profile from ~/.aws/credentials
-        // 2. AWS_ACCESS_KEY_ID + AWS_SECRET_ACCESS_KEY - standard IAM keys
-        // 3. AWS_BEARER_TOKEN_BEDROCK - Bedrock API keys (bearer token)
-        // 4. AWS_CONTAINER_CREDENTIALS_RELATIVE_URI - ECS task roles
-        // 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) {
-            return "<authenticated>";
-        }
-    }
-    const envMap = {
-        openai: "OPENAI_API_KEY",
-        google: "GEMINI_API_KEY",
-        groq: "GROQ_API_KEY",
-        cerebras: "CEREBRAS_API_KEY",
-        xai: "XAI_API_KEY",
-        openrouter: "OPENROUTER_API_KEY",
-        "vercel-ai-gateway": "AI_GATEWAY_API_KEY",
-        zai: "ZAI_API_KEY",
-        mistral: "MISTRAL_API_KEY",
-        minimax: "MINIMAX_API_KEY",
-        "minimax-cn": "MINIMAX_CN_API_KEY",
-        opencode: "OPENCODE_API_KEY",
-    };
-    const envVar = envMap[provider];
-    return envVar ? process.env[envVar] : undefined;
+import "./providers/register-builtins.js";
+import { getApiProvider } from "./api-registry.js";
+export { getEnvApiKey } from "./env-api-keys.js";
+// Set up http proxy according to env variables for `fetch` based SDKs in Node.js.
+// Bun has builtin support for this.
+if (typeof process !== "undefined" && process.versions?.node) {
+    import("undici").then((m) => {
+        const { EnvHttpProxyAgent, setGlobalDispatcher } = m;
+        setGlobalDispatcher(new EnvHttpProxyAgent());
+    });
+}
+function resolveApiProvider(api) {
+    const provider = getApiProvider(api);
+    if (!provider) {
+        throw new Error(`No API provider registered for api: ${api}`);
+    }
+    return provider;
 }
 export function stream(model, context, options) {
-    // Vertex AI uses Application Default Credentials, not API keys
-    if (model.api === "google-vertex") {
-        return streamGoogleVertex(model, context, options);
-    }
-    else if (model.api === "bedrock-converse-stream") {
-        // Bedrock doesn't have any API keys instead it sources credentials from standard AWS env variables or from given AWS profile.
-        return streamBedrock(model, context, (options || {}));
-    }
-    const apiKey = options?.apiKey || getEnvApiKey(model.provider);
-    if (!apiKey) {
-        throw new Error(`No API key for provider: ${model.provider}`);
-    }
-    const providerOptions = { ...options, apiKey };
-    const api = model.api;
-    switch (api) {
-        case "anthropic-messages":
-            return streamAnthropic(model, context, providerOptions);
-        case "openai-completions":
-            return streamOpenAICompletions(model, context, providerOptions);
-        case "openai-responses":
-            return streamOpenAIResponses(model, context, providerOptions);
-        case "openai-codex-responses":
-            return streamOpenAICodexResponses(model, context, providerOptions);
-        case "google-generative-ai":
-            return streamGoogle(model, context, providerOptions);
-        case "google-gemini-cli":
-            return streamGoogleGeminiCli(model, context, providerOptions);
-        default: {
-            // This should never be reached if all Api cases are handled
-            const _exhaustive = api;
-            throw new Error(`Unhandled API: ${_exhaustive}`);
-        }
-    }
+    const provider = resolveApiProvider(model.api);
+    return provider.stream(model, context, options);
 }
 export async function complete(model, context, options) {
     const s = stream(model, context, options);
     return s.result();
 }
 export function streamSimple(model, context, options) {
-    // Vertex AI uses Application Default Credentials, not API keys
-    if (model.api === "google-vertex") {
-        const providerOptions = mapOptionsForApi(model, options, undefined);
-        return stream(model, context, providerOptions);
-    }
-    else if (model.api === "bedrock-converse-stream") {
-        // Bedrock doesn't have any API keys instead it sources credentials from standard AWS env variables or from given AWS profile.
-        const providerOptions = mapOptionsForApi(model, options, undefined);
-        return stream(model, context, providerOptions);
-    }
-    const apiKey = options?.apiKey || getEnvApiKey(model.provider);
-    if (!apiKey) {
-        throw new Error(`No API key for provider: ${model.provider}`);
-    }
-    const providerOptions = mapOptionsForApi(model, options, apiKey);
-    return stream(model, context, providerOptions);
+    const provider = resolveApiProvider(model.api);
+    return provider.streamSimple(model, context, options);
 }
 export async function completeSimple(model, context, options) {
     const s = streamSimple(model, context, options);
     return s.result();
 }
-function mapOptionsForApi(model, options, apiKey) {
-    const base = {
-        temperature: options?.temperature,
-        maxTokens: options?.maxTokens || Math.min(model.maxTokens, 32000),
-        signal: options?.signal,
-        apiKey: apiKey || options?.apiKey,
-        sessionId: options?.sessionId,
-    };
-    // Helper to clamp xhigh to high for providers that don't support it
-    const clampReasoning = (effort) => (effort === "xhigh" ? "high" : effort);
-    /**
-     * Adjust maxTokens to account for thinking budget.
-     * APIs like Anthropic and Bedrock require max_tokens > thinking.budget_tokens.
-     * Returns { adjustedMaxTokens, adjustedThinkingBudget }
-     */
-    const adjustMaxTokensForThinking = (baseMaxTokens, modelMaxTokens, reasoningLevel, customBudgets) => {
-        const defaultBudgets = {
-            minimal: 1024,
-            low: 2048,
-            medium: 8192,
-            high: 16384,
-        };
-        const budgets = { ...defaultBudgets, ...customBudgets };
-        const minOutputTokens = 1024;
-        const level = clampReasoning(reasoningLevel);
-        let thinkingBudget = budgets[level];
-        // Caller's maxTokens is the desired output; add thinking budget on top, capped at model limit
-        const maxTokens = Math.min(baseMaxTokens + thinkingBudget, modelMaxTokens);
-        // If not enough room for thinking + output, reduce thinking budget
-        if (maxTokens <= thinkingBudget) {
-            thinkingBudget = Math.max(0, maxTokens - minOutputTokens);
-        }
-        return { maxTokens, thinkingBudget };
-    };
-    switch (model.api) {
-        case "anthropic-messages": {
-            // Explicitly disable thinking when reasoning is not specified
-            if (!options?.reasoning) {
-                return { ...base, thinkingEnabled: false };
-            }
-            // Claude requires max_tokens > thinking.budget_tokens
-            // So we need to ensure maxTokens accounts for both thinking and output
-            const adjusted = adjustMaxTokensForThinking(base.maxTokens || 0, model.maxTokens, options.reasoning, options?.thinkingBudgets);
-            return {
-                ...base,
-                maxTokens: adjusted.maxTokens,
-                thinkingEnabled: true,
-                thinkingBudgetTokens: adjusted.thinkingBudget,
-            };
-        }
-        case "bedrock-converse-stream": {
-            // Explicitly disable thinking when reasoning is not specified
-            if (!options?.reasoning) {
-                return { ...base, reasoning: undefined };
-            }
-            // Claude requires max_tokens > thinking.budget_tokens (same as Anthropic direct API)
-            // So we need to ensure maxTokens accounts for both thinking and output
-            if (model.id.includes("anthropic.claude") || model.id.includes("anthropic/claude")) {
-                const adjusted = adjustMaxTokensForThinking(base.maxTokens || 0, model.maxTokens, options.reasoning, options?.thinkingBudgets);
-                return {
-                    ...base,
-                    maxTokens: adjusted.maxTokens,
-                    reasoning: options.reasoning,
-                    thinkingBudgets: {
-                        ...(options?.thinkingBudgets || {}),
-                        [clampReasoning(options.reasoning)]: adjusted.thinkingBudget,
-                    },
-                };
-            }
-            // Non-Claude models - pass through
-            return {
-                ...base,
-                reasoning: options?.reasoning,
-                thinkingBudgets: options?.thinkingBudgets,
-            };
-        }
-        case "openai-completions":
-            return {
-                ...base,
-                reasoningEffort: supportsXhigh(model) ? options?.reasoning : clampReasoning(options?.reasoning),
-            };
-        case "openai-responses":
-            return {
-                ...base,
-                reasoningEffort: supportsXhigh(model) ? options?.reasoning : clampReasoning(options?.reasoning),
-            };
-        case "openai-codex-responses":
-            return {
-                ...base,
-                reasoningEffort: supportsXhigh(model) ? options?.reasoning : clampReasoning(options?.reasoning),
-            };
-        case "google-generative-ai": {
-            // Explicitly disable thinking when reasoning is not specified
-            // This is needed because Gemini has "dynamic thinking" enabled by default
-            if (!options?.reasoning) {
-                return { ...base, thinking: { enabled: false } };
-            }
-            const googleModel = model;
-            const effort = clampReasoning(options.reasoning);
-            // Gemini 3 models use thinkingLevel exclusively instead of thinkingBudget.
-            // https://ai.google.dev/gemini-api/docs/thinking#set-budget
-            if (isGemini3ProModel(googleModel) || isGemini3FlashModel(googleModel)) {
-                return {
-                    ...base,
-                    thinking: {
-                        enabled: true,
-                        level: getGemini3ThinkingLevel(effort, googleModel),
-                    },
-                };
-            }
-            return {
-                ...base,
-                thinking: {
-                    enabled: true,
-                    budgetTokens: getGoogleBudget(googleModel, effort, options?.thinkingBudgets),
-                },
-            };
-        }
-        case "google-gemini-cli": {
-            if (!options?.reasoning) {
-                return { ...base, thinking: { enabled: false } };
-            }
-            const effort = clampReasoning(options.reasoning);
-            // Gemini 3 models use thinkingLevel instead of thinkingBudget
-            if (model.id.includes("3-pro") || model.id.includes("3-flash")) {
-                return {
-                    ...base,
-                    thinking: {
-                        enabled: true,
-                        level: getGeminiCliThinkingLevel(effort, model.id),
-                    },
-                };
-            }
-            // Models using thinkingBudget (Gemini 2.x, Claude via Antigravity)
-            // Claude requires max_tokens > thinking.budget_tokens
-            // So we need to ensure maxTokens accounts for both thinking and output
-            const defaultBudgets = {
-                minimal: 1024,
-                low: 2048,
-                medium: 8192,
-                high: 16384,
-            };
-            const budgets = { ...defaultBudgets, ...options?.thinkingBudgets };
-            const minOutputTokens = 1024;
-            let thinkingBudget = budgets[effort];
-            // Caller's maxTokens is the desired output; add thinking budget on top, capped at model limit
-            const maxTokens = Math.min((base.maxTokens || 0) + thinkingBudget, model.maxTokens);
-            // If not enough room for thinking + output, reduce thinking budget
-            if (maxTokens <= thinkingBudget) {
-                thinkingBudget = Math.max(0, maxTokens - minOutputTokens);
-            }
-            return {
-                ...base,
-                maxTokens,
-                thinking: {
-                    enabled: true,
-                    budgetTokens: thinkingBudget,
-                },
-            };
-        }
-        case "google-vertex": {
-            // Explicitly disable thinking when reasoning is not specified
-            if (!options?.reasoning) {
-                return { ...base, thinking: { enabled: false } };
-            }
-            const vertexModel = model;
-            const effort = clampReasoning(options.reasoning);
-            const geminiModel = vertexModel;
-            if (isGemini3ProModel(geminiModel) || isGemini3FlashModel(geminiModel)) {
-                return {
-                    ...base,
-                    thinking: {
-                        enabled: true,
-                        level: getGemini3ThinkingLevel(effort, geminiModel),
-                    },
-                };
-            }
-            return {
-                ...base,
-                thinking: {
-                    enabled: true,
-                    budgetTokens: getGoogleBudget(geminiModel, effort, options?.thinkingBudgets),
-                },
-            };
-        }
-        default: {
-            // Exhaustiveness check
-            const _exhaustive = model.api;
-            throw new Error(`Unhandled API in mapOptionsForApi: ${_exhaustive}`);
-        }
-    }
-}
-function isGemini3ProModel(model) {
-    // Covers gemini-3-pro, gemini-3-pro-preview, and possible other prefixed ids in the future
-    return model.id.includes("3-pro");
-}
-function isGemini3FlashModel(model) {
-    // Covers gemini-3-flash, gemini-3-flash-preview, and possible other prefixed ids in the future
-    return model.id.includes("3-flash");
-}
-function getGemini3ThinkingLevel(effort, model) {
-    if (isGemini3ProModel(model)) {
-        // Gemini 3 Pro only supports LOW/HIGH (for now)
-        switch (effort) {
-            case "minimal":
-            case "low":
-                return "LOW";
-            case "medium":
-            case "high":
-                return "HIGH";
-        }
-    }
-    // Gemini 3 Flash supports all four levels
-    switch (effort) {
-        case "minimal":
-            return "MINIMAL";
-        case "low":
-            return "LOW";
-        case "medium":
-            return "MEDIUM";
-        case "high":
-            return "HIGH";
-    }
-}
-function getGeminiCliThinkingLevel(effort, modelId) {
-    if (modelId.includes("3-pro")) {
-        // Gemini 3 Pro only supports LOW/HIGH (for now)
-        switch (effort) {
-            case "minimal":
-            case "low":
-                return "LOW";
-            case "medium":
-            case "high":
-                return "HIGH";
-        }
-    }
-    // Gemini 3 Flash supports all four levels
-    switch (effort) {
-        case "minimal":
-            return "MINIMAL";
-        case "low":
-            return "LOW";
-        case "medium":
-            return "MEDIUM";
-        case "high":
-            return "HIGH";
-    }
-}
-function getGoogleBudget(model, effort, customBudgets) {
-    // Custom budgets take precedence if provided for this level
-    if (customBudgets?.[effort] !== undefined) {
-        return customBudgets[effort];
-    }
-    // See https://ai.google.dev/gemini-api/docs/thinking#set-budget
-    if (model.id.includes("2.5-pro")) {
-        const budgets = {
-            minimal: 128,
-            low: 2048,
-            medium: 8192,
-            high: 32768,
-        };
-        return budgets[effort];
-    }
-    if (model.id.includes("2.5-flash")) {
-        // Covers 2.5-flash-lite as well
-        const budgets = {
-            minimal: 128,
-            low: 2048,
-            medium: 8192,
-            high: 24576,
-        };
-        return budgets[effort];
-    }
-    // Unknown model - use dynamic
-    return -1;
-}
 //# sourceMappingURL=stream.js.map