npm - @mariozechner/pi-ai - Versions diffs - 0.32.1 → 0.32.3 - Mend

@mariozechner/pi-ai 0.32.1 → 0.32.3

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (4) hide show

package/dist/stream.d.ts.map CHANGED Viewed

	@@ -1 +1 @@
1	- {"version":3,"file":"stream.d.ts","sourceRoot":"","sources":["../src/stream.ts"],"names":[],"mappings":"AAWA,OAAO,KAAK,EACX,GAAG,EACH,gBAAgB,EAChB,2BAA2B,EAC3B,OAAO,EACP,aAAa,EACb,KAAK,EACL,aAAa,EAEb,mBAAmB,EACnB,MAAM,YAAY,CAAC;AAEpB;;;;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;AAmCnE,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,CAqC7B;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,CAQ7B;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 { supportsXhigh } from \"./models.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 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\tReasoningEffort,\n\tSimpleStreamOptions,\n} from \"./types.js\";\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 doesn't use API keys.\n\t// It relies on Google Cloud auth: `gcloud auth application-default login`.\n\t// @google/genai library picks up and manages the auth automatically.\n\t// Return a dummy value to maintain consistency.\n\tif (provider === \"google-vertex\") {\n\t\treturn \"vertex-ai-authenticated\";\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\tzai: \"ZAI_API_KEY\",\n\t\tmistral: \"MISTRAL_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\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 \"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\tcase \"google-vertex\":\n\t\t\treturn streamGoogleVertex(model as Model<\"google-vertex\">, context, providerOptions as GoogleVertexOptions);\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\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};\n\n\t// Helper to clamp xhigh to high for providers that don't support it\n\tconst clampReasoning = (effort: ReasoningEffort \| undefined) => (effort === \"xhigh\" ? \"high\" : effort);\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 anthropicBudgets = {\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\n\t\t\tconst minOutputTokens = 1024;\n\t\t\tlet thinkingBudget = anthropicBudgets[clampReasoning(options.reasoning)!];\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\tthinkingEnabled: true,\n\t\t\t\tthinkingBudgetTokens: thinkingBudget,\n\t\t\t} satisfies AnthropicOptions;\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 \"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),\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 budgets: Record<ClampedReasoningEffort, number> = {\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\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),\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 ClampedReasoningEffort = Exclude<ReasoningEffort, \"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: ClampedReasoningEffort,\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: ClampedReasoningEffort, 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(model: Model<\"google-generative-ai\">, effort: ClampedReasoningEffort): number {\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<ClampedReasoningEffort, 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<ClampedReasoningEffort, 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"]}
1	+ {"version":3,"file":"stream.d.ts","sourceRoot":"","sources":["../src/stream.ts"],"names":[],"mappings":"AAWA,OAAO,KAAK,EACX,GAAG,EACH,gBAAgB,EAChB,2BAA2B,EAC3B,OAAO,EACP,aAAa,EACb,KAAK,EACL,aAAa,EAEb,mBAAmB,EACnB,MAAM,YAAY,CAAC;AAEpB;;;;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+BnE,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,CAuC7B;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,CAc7B;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 { supportsXhigh } from \"./models.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 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\tReasoningEffort,\n\tSimpleStreamOptions,\n} from \"./types.js\";\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\t// Don't return a dummy value - require explicit auth.json configuration.\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\tzai: \"ZAI_API_KEY\",\n\t\tmistral: \"MISTRAL_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}\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 \"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}\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};\n\n\t// Helper to clamp xhigh to high for providers that don't support it\n\tconst clampReasoning = (effort: ReasoningEffort \| undefined) => (effort === \"xhigh\" ? \"high\" : effort);\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 anthropicBudgets = {\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\n\t\t\tconst minOutputTokens = 1024;\n\t\t\tlet thinkingBudget = anthropicBudgets[clampReasoning(options.reasoning)!];\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\tthinkingEnabled: true,\n\t\t\t\tthinkingBudgetTokens: thinkingBudget,\n\t\t\t} satisfies AnthropicOptions;\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 \"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),\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 budgets: Record<ClampedReasoningEffort, number> = {\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\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),\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 ClampedReasoningEffort = Exclude<ReasoningEffort, \"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: ClampedReasoningEffort,\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: ClampedReasoningEffort, 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(model: Model<\"google-generative-ai\">, effort: ClampedReasoningEffort): number {\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<ClampedReasoningEffort, 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<ClampedReasoningEffort, 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"]}

package/dist/stream.js CHANGED Viewed

@@ -14,13 +14,9 @@ export function getEnvApiKey(provider) {
     if (provider === "anthropic") {
         return process.env.ANTHROPIC_OAUTH_TOKEN || process.env.ANTHROPIC_API_KEY;
     }
-    // Vertex AI doesn't use API keys.
-    // It relies on Google Cloud auth: `gcloud auth application-default login`.
-    // @google/genai library picks up and manages the auth automatically.
-    // Return a dummy value to maintain consistency.
-    if (provider === "google-vertex") {
-        return "vertex-ai-authenticated";
-    }
+    // Vertex AI uses Application Default Credentials, not API keys.
+    // Auth is configured via `gcloud auth application-default login`.
+    // Don't return a dummy value - require explicit auth.json configuration.
     const envMap = {
         openai: "OPENAI_API_KEY",
         google: "GEMINI_API_KEY",
@@ -35,6 +31,10 @@ export function getEnvApiKey(provider) {
     return envVar ? process.env[envVar] : undefined;
 }
 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);
+    }
     const apiKey = options?.apiKey || getEnvApiKey(model.provider);
     if (!apiKey) {
         throw new Error(`No API key for provider: ${model.provider}`);
@@ -52,8 +52,6 @@ export function stream(model, context, options) {
             return streamGoogle(model, context, providerOptions);
         case "google-gemini-cli":
             return streamGoogleGeminiCli(model, context, providerOptions);
-        case "google-vertex":
-            return streamGoogleVertex(model, context, providerOptions);
         default: {
             // This should never be reached if all Api cases are handled
             const _exhaustive = api;
@@ -66,6 +64,11 @@ export async function complete(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);
+    }
     const apiKey = options?.apiKey || getEnvApiKey(model.provider);
     if (!apiKey) {
         throw new Error(`No API key for provider: ${model.provider}`);

package/dist/stream.js.map CHANGED Viewed

	@@ -1 +1 @@
1	- {"version":3,"file":"stream.js","sourceRoot":"","sources":["../src/stream.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AAC5C,OAAO,EAAyB,eAAe,EAAE,MAAM,0BAA0B,CAAC;AAClF,OAAO,EAAsB,YAAY,EAAE,MAAM,uBAAuB,CAAC;AACzE,OAAO,EAGN,qBAAqB,GACrB,MAAM,kCAAkC,CAAC;AAC1C,OAAO,EAA4B,kBAAkB,EAAE,MAAM,8BAA8B,CAAC;AAC5F,OAAO,EAAiC,uBAAuB,EAAE,MAAM,mCAAmC,CAAC;AAC3G,OAAO,EAA+B,qBAAqB,EAAE,MAAM,iCAAiC,CAAC;AAoBrG,MAAM,UAAU,YAAY,CAAC,QAAa,EAAsB;IAC/D,qCAAqC;IACrC,IAAI,QAAQ,KAAK,gBAAgB,EAAE,CAAC;QACnC,OAAO,OAAO,CAAC,GAAG,CAAC,oBAAoB,IAAI,OAAO,CAAC,GAAG,CAAC,QAAQ,IAAI,OAAO,CAAC,GAAG,CAAC,YAAY,CAAC;IAC7F,CAAC;IAED,gEAAgE;IAChE,IAAI,QAAQ,KAAK,WAAW,EAAE,CAAC;QAC9B,OAAO,OAAO,CAAC,GAAG,CAAC,qBAAqB,IAAI,OAAO,CAAC,GAAG,CAAC,iBAAiB,CAAC;IAC3E,CAAC;IAED,kCAAkC;IAClC,2EAA2E;IAC3E,qEAAqE;IACrE,gDAAgD;IAChD,IAAI,QAAQ,KAAK,eAAe,EAAE,CAAC;QAClC,OAAO,yBAAyB,CAAC;IAClC,CAAC;IAED,MAAM,MAAM,GAA2B;QACtC,MAAM,EAAE,gBAAgB;QACxB,MAAM,EAAE,gBAAgB;QACxB,IAAI,EAAE,cAAc;QACpB,QAAQ,EAAE,kBAAkB;QAC5B,GAAG,EAAE,aAAa;QAClB,UAAU,EAAE,oBAAoB;QAChC,GAAG,EAAE,aAAa;QAClB,OAAO,EAAE,iBAAiB;KAC1B,CAAC;IAEF,MAAM,MAAM,GAAG,MAAM,CAAC,QAAQ,CAAC,CAAC;IAChC,OAAO,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;AAAA,CAChD;AAED,MAAM,UAAU,MAAM,CACrB,KAAkB,EAClB,OAAgB,EAChB,OAA6B,EACC;IAC9B,MAAM,MAAM,GAAG,OAAO,EAAE,MAAM,IAAI,YAAY,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;IAC/D,IAAI,CAAC,MAAM,EAAE,CAAC;QACb,MAAM,IAAI,KAAK,CAAC,4BAA4B,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC;IAC/D,CAAC;IACD,MAAM,eAAe,GAAG,EAAE,GAAG,OAAO,EAAE,MAAM,EAAE,CAAC;IAE/C,MAAM,GAAG,GAAQ,KAAK,CAAC,GAAG,CAAC;IAC3B,QAAQ,GAAG,EAAE,CAAC;QACb,KAAK,oBAAoB;YACxB,OAAO,eAAe,CAAC,KAAoC,EAAE,OAAO,EAAE,eAAe,CAAC,CAAC;QAExF,KAAK,oBAAoB;YACxB,OAAO,uBAAuB,CAAC,KAAoC,EAAE,OAAO,EAAE,eAAsB,CAAC,CAAC;QAEvG,KAAK,kBAAkB;YACtB,OAAO,qBAAqB,CAAC,KAAkC,EAAE,OAAO,EAAE,eAAsB,CAAC,CAAC;QAEnG,KAAK,sBAAsB;YAC1B,OAAO,YAAY,CAAC,KAAsC,EAAE,OAAO,EAAE,eAAe,CAAC,CAAC;QAEvF,KAAK,mBAAmB;YACvB,OAAO,qBAAqB,CAC3B,KAAmC,EACnC,OAAO,EACP,eAAyC,CACzC,CAAC;QAEH,KAAK,eAAe;YACnB,OAAO,kBAAkB,CAAC,KAA+B,EAAE,OAAO,EAAE,eAAsC,CAAC,CAAC;QAE7G,SAAS,CAAC;YACT,4DAA4D;YAC5D,MAAM,WAAW,GAAU,GAAG,CAAC;YAC/B,MAAM,IAAI,KAAK,CAAC,kBAAkB,WAAW,EAAE,CAAC,CAAC;QAClD,CAAC;IACF,CAAC;AAAA,CACD;AAED,MAAM,CAAC,KAAK,UAAU,QAAQ,CAC7B,KAAkB,EAClB,OAAgB,EAChB,OAA6B,EACD;IAC5B,MAAM,CAAC,GAAG,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;IAC1C,OAAO,CAAC,CAAC,MAAM,EAAE,CAAC;AAAA,CAClB;AAED,MAAM,UAAU,YAAY,CAC3B,KAAkB,EAClB,OAAgB,EAChB,OAA6B,EACC;IAC9B,MAAM,MAAM,GAAG,OAAO,EAAE,MAAM,IAAI,YAAY,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;IAC/D,IAAI,CAAC,MAAM,EAAE,CAAC;QACb,MAAM,IAAI,KAAK,CAAC,4BAA4B,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC;IAC/D,CAAC;IAED,MAAM,eAAe,GAAG,gBAAgB,CAAC,KAAK,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC;IACjE,OAAO,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,eAAe,CAAC,CAAC;AAAA,CAC/C;AAED,MAAM,CAAC,KAAK,UAAU,cAAc,CACnC,KAAkB,EAClB,OAAgB,EAChB,OAA6B,EACD;IAC5B,MAAM,CAAC,GAAG,YAAY,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;IAChD,OAAO,CAAC,CAAC,MAAM,EAAE,CAAC;AAAA,CAClB;AAED,SAAS,gBAAgB,CACxB,KAAkB,EAClB,OAA6B,EAC7B,MAAe,EACO;IACtB,MAAM,IAAI,GAAG;QACZ,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;KACjC,CAAC;IAEF,oEAAoE;IACpE,MAAM,cAAc,GAAG,CAAC,MAAmC,EAAE,EAAE,CAAC,CAAC,MAAM,KAAK,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC;IAEvG,QAAQ,KAAK,CAAC,GAAG,EAAE,CAAC;QACnB,KAAK,oBAAoB,EAAE,CAAC;YAC3B,8DAA8D;YAC9D,IAAI,CAAC,OAAO,EAAE,SAAS,EAAE,CAAC;gBACzB,OAAO,EAAE,GAAG,IAAI,EAAE,eAAe,EAAE,KAAK,EAA6B,CAAC;YACvE,CAAC;YAED,sDAAsD;YACtD,uEAAuE;YACvE,MAAM,gBAAgB,GAAG;gBACxB,OAAO,EAAE,IAAI;gBACb,GAAG,EAAE,IAAI;gBACT,MAAM,EAAE,IAAI;gBACZ,IAAI,EAAE,KAAK;aACX,CAAC;YAEF,MAAM,eAAe,GAAG,IAAI,CAAC;YAC7B,IAAI,cAAc,GAAG,gBAAgB,CAAC,cAAc,CAAC,OAAO,CAAC,SAAS,CAAE,CAAC,CAAC;YAC1E,8FAA8F;YAC9F,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,SAAS,IAAI,CAAC,CAAC,GAAG,cAAc,EAAE,KAAK,CAAC,SAAS,CAAC,CAAC;YAEpF,mEAAmE;YACnE,IAAI,SAAS,IAAI,cAAc,EAAE,CAAC;gBACjC,cAAc,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,SAAS,GAAG,eAAe,CAAC,CAAC;YAC3D,CAAC;YAED,OAAO;gBACN,GAAG,IAAI;gBACP,SAAS;gBACT,eAAe,EAAE,IAAI;gBACrB,oBAAoB,EAAE,cAAc;aACT,CAAC;QAC9B,CAAC;QAED,KAAK,oBAAoB;YACxB,OAAO;gBACN,GAAG,IAAI;gBACP,eAAe,EAAE,aAAa,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC,CAAC,cAAc,CAAC,OAAO,EAAE,SAAS,CAAC;aAC5D,CAAC;QAEtC,KAAK,kBAAkB;YACtB,OAAO;gBACN,GAAG,IAAI;gBACP,eAAe,EAAE,aAAa,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC,CAAC,cAAc,CAAC,OAAO,EAAE,SAAS,CAAC;aAC9D,CAAC;QAEpC,KAAK,sBAAsB,EAAE,CAAC;YAC7B,8DAA8D;YAC9D,0EAA0E;YAC1E,IAAI,CAAC,OAAO,EAAE,SAAS,EAAE,CAAC;gBACzB,OAAO,EAAE,GAAG,IAAI,EAAE,QAAQ,EAAE,EAAE,OAAO,EAAE,KAAK,EAAE,EAA0B,CAAC;YAC1E,CAAC;YAED,MAAM,WAAW,GAAG,KAAsC,CAAC;YAC3D,MAAM,MAAM,GAAG,cAAc,CAAC,OAAO,CAAC,SAAS,CAAE,CAAC;YAElD,2EAA2E;YAC3E,4DAA4D;YAC5D,IAAI,iBAAiB,CAAC,WAAW,CAAC,IAAI,mBAAmB,CAAC,WAAW,CAAC,EAAE,CAAC;gBACxE,OAAO;oBACN,GAAG,IAAI;oBACP,QAAQ,EAAE;wBACT,OAAO,EAAE,IAAI;wBACb,KAAK,EAAE,uBAAuB,CAAC,MAAM,EAAE,WAAW,CAAC;qBACnD;iBACuB,CAAC;YAC3B,CAAC;YAED,OAAO;gBACN,GAAG,IAAI;gBACP,QAAQ,EAAE;oBACT,OAAO,EAAE,IAAI;oBACb,YAAY,EAAE,eAAe,CAAC,WAAW,EAAE,MAAM,CAAC;iBAClD;aACuB,CAAC;QAC3B,CAAC;QAED,KAAK,mBAAmB,EAAE,CAAC;YAC1B,IAAI,CAAC,OAAO,EAAE,SAAS,EAAE,CAAC;gBACzB,OAAO,EAAE,GAAG,IAAI,EAAE,QAAQ,EAAE,EAAE,OAAO,EAAE,KAAK,EAAE,EAAmC,CAAC;YACnF,CAAC;YAED,MAAM,MAAM,GAAG,cAAc,CAAC,OAAO,CAAC,SAAS,CAAE,CAAC;YAElD,8DAA8D;YAC9D,IAAI,KAAK,CAAC,EAAE,CAAC,QAAQ,CAAC,OAAO,CAAC,IAAI,KAAK,CAAC,EAAE,CAAC,QAAQ,CAAC,SAAS,CAAC,EAAE,CAAC;gBAChE,OAAO;oBACN,GAAG,IAAI;oBACP,QAAQ,EAAE;wBACT,OAAO,EAAE,IAAI;wBACb,KAAK,EAAE,yBAAyB,CAAC,MAAM,EAAE,KAAK,CAAC,EAAE,CAAC;qBAClD;iBACgC,CAAC;YACpC,CAAC;YAED,mEAAmE;YACnE,sDAAsD;YACtD,uEAAuE;YACvE,MAAM,OAAO,GAA2C;gBACvD,OAAO,EAAE,IAAI;gBACb,GAAG,EAAE,IAAI;gBACT,MAAM,EAAE,IAAI;gBACZ,IAAI,EAAE,KAAK;aACX,CAAC;YAEF,MAAM,eAAe,GAAG,IAAI,CAAC;YAC7B,IAAI,cAAc,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC;YACrC,8FAA8F;YAC9F,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,SAAS,IAAI,CAAC,CAAC,GAAG,cAAc,EAAE,KAAK,CAAC,SAAS,CAAC,CAAC;YAEpF,mEAAmE;YACnE,IAAI,SAAS,IAAI,cAAc,EAAE,CAAC;gBACjC,cAAc,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,SAAS,GAAG,eAAe,CAAC,CAAC;YAC3D,CAAC;YAED,OAAO;gBACN,GAAG,IAAI;gBACP,SAAS;gBACT,QAAQ,EAAE;oBACT,OAAO,EAAE,IAAI;oBACb,YAAY,EAAE,cAAc;iBAC5B;aACgC,CAAC;QACpC,CAAC;QAED,KAAK,eAAe,EAAE,CAAC;YACtB,8DAA8D;YAC9D,IAAI,CAAC,OAAO,EAAE,SAAS,EAAE,CAAC;gBACzB,OAAO,EAAE,GAAG,IAAI,EAAE,QAAQ,EAAE,EAAE,OAAO,EAAE,KAAK,EAAE,EAAgC,CAAC;YAChF,CAAC;YAED,MAAM,WAAW,GAAG,KAA+B,CAAC;YACpD,MAAM,MAAM,GAAG,cAAc,CAAC,OAAO,CAAC,SAAS,CAAE,CAAC;YAClD,MAAM,WAAW,GAAG,WAAuD,CAAC;YAE5E,IAAI,iBAAiB,CAAC,WAAW,CAAC,IAAI,mBAAmB,CAAC,WAAW,CAAC,EAAE,CAAC;gBACxE,OAAO;oBACN,GAAG,IAAI;oBACP,QAAQ,EAAE;wBACT,OAAO,EAAE,IAAI;wBACb,KAAK,EAAE,uBAAuB,CAAC,MAAM,EAAE,WAAW,CAAC;qBACnD;iBAC6B,CAAC;YACjC,CAAC;YAED,OAAO;gBACN,GAAG,IAAI;gBACP,QAAQ,EAAE;oBACT,OAAO,EAAE,IAAI;oBACb,YAAY,EAAE,eAAe,CAAC,WAAW,EAAE,MAAM,CAAC;iBAClD;aAC6B,CAAC;QACjC,CAAC;QAED,SAAS,CAAC;YACT,uBAAuB;YACvB,MAAM,WAAW,GAAU,KAAK,CAAC,GAAG,CAAC;YACrC,MAAM,IAAI,KAAK,CAAC,sCAAsC,WAAW,EAAE,CAAC,CAAC;QACtE,CAAC;IACF,CAAC;AAAA,CACD;AAID,SAAS,iBAAiB,CAAC,KAAoC,EAAW;IACzE,2FAA2F;IAC3F,OAAO,KAAK,CAAC,EAAE,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;AAAA,CAClC;AAED,SAAS,mBAAmB,CAAC,KAAoC,EAAW;IAC3E,+FAA+F;IAC/F,OAAO,KAAK,CAAC,EAAE,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC;AAAA,CACpC;AAED,SAAS,uBAAuB,CAC/B,MAA8B,EAC9B,KAAoC,EACd;IACtB,IAAI,iBAAiB,CAAC,KAAK,CAAC,EAAE,CAAC;QAC9B,gDAAgD;QAChD,QAAQ,MAAM,EAAE,CAAC;YAChB,KAAK,SAAS,CAAC;YACf,KAAK,KAAK;gBACT,OAAO,KAAK,CAAC;YACd,KAAK,QAAQ,CAAC;YACd,KAAK,MAAM;gBACV,OAAO,MAAM,CAAC;QAChB,CAAC;IACF,CAAC;IACD,0CAA0C;IAC1C,QAAQ,MAAM,EAAE,CAAC;QAChB,KAAK,SAAS;YACb,OAAO,SAAS,CAAC;QAClB,KAAK,KAAK;YACT,OAAO,KAAK,CAAC;QACd,KAAK,QAAQ;YACZ,OAAO,QAAQ,CAAC;QACjB,KAAK,MAAM;YACV,OAAO,MAAM,CAAC;IAChB,CAAC;AAAA,CACD;AAED,SAAS,yBAAyB,CAAC,MAA8B,EAAE,OAAe,EAAuB;IACxG,IAAI,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,CAAC;QAC/B,gDAAgD;QAChD,QAAQ,MAAM,EAAE,CAAC;YAChB,KAAK,SAAS,CAAC;YACf,KAAK,KAAK;gBACT,OAAO,KAAK,CAAC;YACd,KAAK,QAAQ,CAAC;YACd,KAAK,MAAM;gBACV,OAAO,MAAM,CAAC;QAChB,CAAC;IACF,CAAC;IACD,0CAA0C;IAC1C,QAAQ,MAAM,EAAE,CAAC;QAChB,KAAK,SAAS;YACb,OAAO,SAAS,CAAC;QAClB,KAAK,KAAK;YACT,OAAO,KAAK,CAAC;QACd,KAAK,QAAQ;YACZ,OAAO,QAAQ,CAAC;QACjB,KAAK,MAAM;YACV,OAAO,MAAM,CAAC;IAChB,CAAC;AAAA,CACD;AAED,SAAS,eAAe,CAAC,KAAoC,EAAE,MAA8B,EAAU;IACtG,gEAAgE;IAChE,IAAI,KAAK,CAAC,EAAE,CAAC,QAAQ,CAAC,SAAS,CAAC,EAAE,CAAC;QAClC,MAAM,OAAO,GAA2C;YACvD,OAAO,EAAE,GAAG;YACZ,GAAG,EAAE,IAAI;YACT,MAAM,EAAE,IAAI;YACZ,IAAI,EAAE,KAAK;SACX,CAAC;QACF,OAAO,OAAO,CAAC,MAAM,CAAC,CAAC;IACxB,CAAC;IAED,IAAI,KAAK,CAAC,EAAE,CAAC,QAAQ,CAAC,WAAW,CAAC,EAAE,CAAC;QACpC,gCAAgC;QAChC,MAAM,OAAO,GAA2C;YACvD,OAAO,EAAE,GAAG;YACZ,GAAG,EAAE,IAAI;YACT,MAAM,EAAE,IAAI;YACZ,IAAI,EAAE,KAAK;SACX,CAAC;QACF,OAAO,OAAO,CAAC,MAAM,CAAC,CAAC;IACxB,CAAC;IAED,8BAA8B;IAC9B,OAAO,CAAC,CAAC,CAAC;AAAA,CACV","sourcesContent":["import { supportsXhigh } from \"./models.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 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\tReasoningEffort,\n\tSimpleStreamOptions,\n} from \"./types.js\";\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 doesn't use API keys.\n\t// It relies on Google Cloud auth: `gcloud auth application-default login`.\n\t// @google/genai library picks up and manages the auth automatically.\n\t// Return a dummy value to maintain consistency.\n\tif (provider === \"google-vertex\") {\n\t\treturn \"vertex-ai-authenticated\";\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\tzai: \"ZAI_API_KEY\",\n\t\tmistral: \"MISTRAL_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\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 \"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\tcase \"google-vertex\":\n\t\t\treturn streamGoogleVertex(model as Model<\"google-vertex\">, context, providerOptions as GoogleVertexOptions);\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\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};\n\n\t// Helper to clamp xhigh to high for providers that don't support it\n\tconst clampReasoning = (effort: ReasoningEffort \| undefined) => (effort === \"xhigh\" ? \"high\" : effort);\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 anthropicBudgets = {\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\n\t\t\tconst minOutputTokens = 1024;\n\t\t\tlet thinkingBudget = anthropicBudgets[clampReasoning(options.reasoning)!];\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\tthinkingEnabled: true,\n\t\t\t\tthinkingBudgetTokens: thinkingBudget,\n\t\t\t} satisfies AnthropicOptions;\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 \"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),\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 budgets: Record<ClampedReasoningEffort, number> = {\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\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),\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 ClampedReasoningEffort = Exclude<ReasoningEffort, \"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: ClampedReasoningEffort,\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: ClampedReasoningEffort, 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(model: Model<\"google-generative-ai\">, effort: ClampedReasoningEffort): number {\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<ClampedReasoningEffort, 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<ClampedReasoningEffort, 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"]}
1	+ {"version":3,"file":"stream.js","sourceRoot":"","sources":["../src/stream.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AAC5C,OAAO,EAAyB,eAAe,EAAE,MAAM,0BAA0B,CAAC;AAClF,OAAO,EAAsB,YAAY,EAAE,MAAM,uBAAuB,CAAC;AACzE,OAAO,EAGN,qBAAqB,GACrB,MAAM,kCAAkC,CAAC;AAC1C,OAAO,EAA4B,kBAAkB,EAAE,MAAM,8BAA8B,CAAC;AAC5F,OAAO,EAAiC,uBAAuB,EAAE,MAAM,mCAAmC,CAAC;AAC3G,OAAO,EAA+B,qBAAqB,EAAE,MAAM,iCAAiC,CAAC;AAoBrG,MAAM,UAAU,YAAY,CAAC,QAAa,EAAsB;IAC/D,qCAAqC;IACrC,IAAI,QAAQ,KAAK,gBAAgB,EAAE,CAAC;QACnC,OAAO,OAAO,CAAC,GAAG,CAAC,oBAAoB,IAAI,OAAO,CAAC,GAAG,CAAC,QAAQ,IAAI,OAAO,CAAC,GAAG,CAAC,YAAY,CAAC;IAC7F,CAAC;IAED,gEAAgE;IAChE,IAAI,QAAQ,KAAK,WAAW,EAAE,CAAC;QAC9B,OAAO,OAAO,CAAC,GAAG,CAAC,qBAAqB,IAAI,OAAO,CAAC,GAAG,CAAC,iBAAiB,CAAC;IAC3E,CAAC;IAED,gEAAgE;IAChE,kEAAkE;IAClE,yEAAyE;IAEzE,MAAM,MAAM,GAA2B;QACtC,MAAM,EAAE,gBAAgB;QACxB,MAAM,EAAE,gBAAgB;QACxB,IAAI,EAAE,cAAc;QACpB,QAAQ,EAAE,kBAAkB;QAC5B,GAAG,EAAE,aAAa;QAClB,UAAU,EAAE,oBAAoB;QAChC,GAAG,EAAE,aAAa;QAClB,OAAO,EAAE,iBAAiB;KAC1B,CAAC;IAEF,MAAM,MAAM,GAAG,MAAM,CAAC,QAAQ,CAAC,CAAC;IAChC,OAAO,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;AAAA,CAChD;AAED,MAAM,UAAU,MAAM,CACrB,KAAkB,EAClB,OAAgB,EAChB,OAA6B,EACC;IAC9B,+DAA+D;IAC/D,IAAI,KAAK,CAAC,GAAG,KAAK,eAAe,EAAE,CAAC;QACnC,OAAO,kBAAkB,CAAC,KAA+B,EAAE,OAAO,EAAE,OAA8B,CAAC,CAAC;IACrG,CAAC;IAED,MAAM,MAAM,GAAG,OAAO,EAAE,MAAM,IAAI,YAAY,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;IAC/D,IAAI,CAAC,MAAM,EAAE,CAAC;QACb,MAAM,IAAI,KAAK,CAAC,4BAA4B,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC;IAC/D,CAAC;IACD,MAAM,eAAe,GAAG,EAAE,GAAG,OAAO,EAAE,MAAM,EAAE,CAAC;IAE/C,MAAM,GAAG,GAAQ,KAAK,CAAC,GAAG,CAAC;IAC3B,QAAQ,GAAG,EAAE,CAAC;QACb,KAAK,oBAAoB;YACxB,OAAO,eAAe,CAAC,KAAoC,EAAE,OAAO,EAAE,eAAe,CAAC,CAAC;QAExF,KAAK,oBAAoB;YACxB,OAAO,uBAAuB,CAAC,KAAoC,EAAE,OAAO,EAAE,eAAsB,CAAC,CAAC;QAEvG,KAAK,kBAAkB;YACtB,OAAO,qBAAqB,CAAC,KAAkC,EAAE,OAAO,EAAE,eAAsB,CAAC,CAAC;QAEnG,KAAK,sBAAsB;YAC1B,OAAO,YAAY,CAAC,KAAsC,EAAE,OAAO,EAAE,eAAe,CAAC,CAAC;QAEvF,KAAK,mBAAmB;YACvB,OAAO,qBAAqB,CAC3B,KAAmC,EACnC,OAAO,EACP,eAAyC,CACzC,CAAC;QAEH,SAAS,CAAC;YACT,4DAA4D;YAC5D,MAAM,WAAW,GAAU,GAAG,CAAC;YAC/B,MAAM,IAAI,KAAK,CAAC,kBAAkB,WAAW,EAAE,CAAC,CAAC;QAClD,CAAC;IACF,CAAC;AAAA,CACD;AAED,MAAM,CAAC,KAAK,UAAU,QAAQ,CAC7B,KAAkB,EAClB,OAAgB,EAChB,OAA6B,EACD;IAC5B,MAAM,CAAC,GAAG,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;IAC1C,OAAO,CAAC,CAAC,MAAM,EAAE,CAAC;AAAA,CAClB;AAED,MAAM,UAAU,YAAY,CAC3B,KAAkB,EAClB,OAAgB,EAChB,OAA6B,EACC;IAC9B,+DAA+D;IAC/D,IAAI,KAAK,CAAC,GAAG,KAAK,eAAe,EAAE,CAAC;QACnC,MAAM,eAAe,GAAG,gBAAgB,CAAC,KAAK,EAAE,OAAO,EAAE,SAAS,CAAC,CAAC;QACpE,OAAO,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,eAAe,CAAC,CAAC;IAChD,CAAC;IAED,MAAM,MAAM,GAAG,OAAO,EAAE,MAAM,IAAI,YAAY,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;IAC/D,IAAI,CAAC,MAAM,EAAE,CAAC;QACb,MAAM,IAAI,KAAK,CAAC,4BAA4B,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC;IAC/D,CAAC;IAED,MAAM,eAAe,GAAG,gBAAgB,CAAC,KAAK,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC;IACjE,OAAO,MAAM,CAAC,KAAK,EAAE,OAAO,EAAE,eAAe,CAAC,CAAC;AAAA,CAC/C;AAED,MAAM,CAAC,KAAK,UAAU,cAAc,CACnC,KAAkB,EAClB,OAAgB,EAChB,OAA6B,EACD;IAC5B,MAAM,CAAC,GAAG,YAAY,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;IAChD,OAAO,CAAC,CAAC,MAAM,EAAE,CAAC;AAAA,CAClB;AAED,SAAS,gBAAgB,CACxB,KAAkB,EAClB,OAA6B,EAC7B,MAAe,EACO;IACtB,MAAM,IAAI,GAAG;QACZ,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;KACjC,CAAC;IAEF,oEAAoE;IACpE,MAAM,cAAc,GAAG,CAAC,MAAmC,EAAE,EAAE,CAAC,CAAC,MAAM,KAAK,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC;IAEvG,QAAQ,KAAK,CAAC,GAAG,EAAE,CAAC;QACnB,KAAK,oBAAoB,EAAE,CAAC;YAC3B,8DAA8D;YAC9D,IAAI,CAAC,OAAO,EAAE,SAAS,EAAE,CAAC;gBACzB,OAAO,EAAE,GAAG,IAAI,EAAE,eAAe,EAAE,KAAK,EAA6B,CAAC;YACvE,CAAC;YAED,sDAAsD;YACtD,uEAAuE;YACvE,MAAM,gBAAgB,GAAG;gBACxB,OAAO,EAAE,IAAI;gBACb,GAAG,EAAE,IAAI;gBACT,MAAM,EAAE,IAAI;gBACZ,IAAI,EAAE,KAAK;aACX,CAAC;YAEF,MAAM,eAAe,GAAG,IAAI,CAAC;YAC7B,IAAI,cAAc,GAAG,gBAAgB,CAAC,cAAc,CAAC,OAAO,CAAC,SAAS,CAAE,CAAC,CAAC;YAC1E,8FAA8F;YAC9F,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,SAAS,IAAI,CAAC,CAAC,GAAG,cAAc,EAAE,KAAK,CAAC,SAAS,CAAC,CAAC;YAEpF,mEAAmE;YACnE,IAAI,SAAS,IAAI,cAAc,EAAE,CAAC;gBACjC,cAAc,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,SAAS,GAAG,eAAe,CAAC,CAAC;YAC3D,CAAC;YAED,OAAO;gBACN,GAAG,IAAI;gBACP,SAAS;gBACT,eAAe,EAAE,IAAI;gBACrB,oBAAoB,EAAE,cAAc;aACT,CAAC;QAC9B,CAAC;QAED,KAAK,oBAAoB;YACxB,OAAO;gBACN,GAAG,IAAI;gBACP,eAAe,EAAE,aAAa,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC,CAAC,cAAc,CAAC,OAAO,EAAE,SAAS,CAAC;aAC5D,CAAC;QAEtC,KAAK,kBAAkB;YACtB,OAAO;gBACN,GAAG,IAAI;gBACP,eAAe,EAAE,aAAa,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC,CAAC,cAAc,CAAC,OAAO,EAAE,SAAS,CAAC;aAC9D,CAAC;QAEpC,KAAK,sBAAsB,EAAE,CAAC;YAC7B,8DAA8D;YAC9D,0EAA0E;YAC1E,IAAI,CAAC,OAAO,EAAE,SAAS,EAAE,CAAC;gBACzB,OAAO,EAAE,GAAG,IAAI,EAAE,QAAQ,EAAE,EAAE,OAAO,EAAE,KAAK,EAAE,EAA0B,CAAC;YAC1E,CAAC;YAED,MAAM,WAAW,GAAG,KAAsC,CAAC;YAC3D,MAAM,MAAM,GAAG,cAAc,CAAC,OAAO,CAAC,SAAS,CAAE,CAAC;YAElD,2EAA2E;YAC3E,4DAA4D;YAC5D,IAAI,iBAAiB,CAAC,WAAW,CAAC,IAAI,mBAAmB,CAAC,WAAW,CAAC,EAAE,CAAC;gBACxE,OAAO;oBACN,GAAG,IAAI;oBACP,QAAQ,EAAE;wBACT,OAAO,EAAE,IAAI;wBACb,KAAK,EAAE,uBAAuB,CAAC,MAAM,EAAE,WAAW,CAAC;qBACnD;iBACuB,CAAC;YAC3B,CAAC;YAED,OAAO;gBACN,GAAG,IAAI;gBACP,QAAQ,EAAE;oBACT,OAAO,EAAE,IAAI;oBACb,YAAY,EAAE,eAAe,CAAC,WAAW,EAAE,MAAM,CAAC;iBAClD;aACuB,CAAC;QAC3B,CAAC;QAED,KAAK,mBAAmB,EAAE,CAAC;YAC1B,IAAI,CAAC,OAAO,EAAE,SAAS,EAAE,CAAC;gBACzB,OAAO,EAAE,GAAG,IAAI,EAAE,QAAQ,EAAE,EAAE,OAAO,EAAE,KAAK,EAAE,EAAmC,CAAC;YACnF,CAAC;YAED,MAAM,MAAM,GAAG,cAAc,CAAC,OAAO,CAAC,SAAS,CAAE,CAAC;YAElD,8DAA8D;YAC9D,IAAI,KAAK,CAAC,EAAE,CAAC,QAAQ,CAAC,OAAO,CAAC,IAAI,KAAK,CAAC,EAAE,CAAC,QAAQ,CAAC,SAAS,CAAC,EAAE,CAAC;gBAChE,OAAO;oBACN,GAAG,IAAI;oBACP,QAAQ,EAAE;wBACT,OAAO,EAAE,IAAI;wBACb,KAAK,EAAE,yBAAyB,CAAC,MAAM,EAAE,KAAK,CAAC,EAAE,CAAC;qBAClD;iBACgC,CAAC;YACpC,CAAC;YAED,mEAAmE;YACnE,sDAAsD;YACtD,uEAAuE;YACvE,MAAM,OAAO,GAA2C;gBACvD,OAAO,EAAE,IAAI;gBACb,GAAG,EAAE,IAAI;gBACT,MAAM,EAAE,IAAI;gBACZ,IAAI,EAAE,KAAK;aACX,CAAC;YAEF,MAAM,eAAe,GAAG,IAAI,CAAC;YAC7B,IAAI,cAAc,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC;YACrC,8FAA8F;YAC9F,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,SAAS,IAAI,CAAC,CAAC,GAAG,cAAc,EAAE,KAAK,CAAC,SAAS,CAAC,CAAC;YAEpF,mEAAmE;YACnE,IAAI,SAAS,IAAI,cAAc,EAAE,CAAC;gBACjC,cAAc,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,SAAS,GAAG,eAAe,CAAC,CAAC;YAC3D,CAAC;YAED,OAAO;gBACN,GAAG,IAAI;gBACP,SAAS;gBACT,QAAQ,EAAE;oBACT,OAAO,EAAE,IAAI;oBACb,YAAY,EAAE,cAAc;iBAC5B;aACgC,CAAC;QACpC,CAAC;QAED,KAAK,eAAe,EAAE,CAAC;YACtB,8DAA8D;YAC9D,IAAI,CAAC,OAAO,EAAE,SAAS,EAAE,CAAC;gBACzB,OAAO,EAAE,GAAG,IAAI,EAAE,QAAQ,EAAE,EAAE,OAAO,EAAE,KAAK,EAAE,EAAgC,CAAC;YAChF,CAAC;YAED,MAAM,WAAW,GAAG,KAA+B,CAAC;YACpD,MAAM,MAAM,GAAG,cAAc,CAAC,OAAO,CAAC,SAAS,CAAE,CAAC;YAClD,MAAM,WAAW,GAAG,WAAuD,CAAC;YAE5E,IAAI,iBAAiB,CAAC,WAAW,CAAC,IAAI,mBAAmB,CAAC,WAAW,CAAC,EAAE,CAAC;gBACxE,OAAO;oBACN,GAAG,IAAI;oBACP,QAAQ,EAAE;wBACT,OAAO,EAAE,IAAI;wBACb,KAAK,EAAE,uBAAuB,CAAC,MAAM,EAAE,WAAW,CAAC;qBACnD;iBAC6B,CAAC;YACjC,CAAC;YAED,OAAO;gBACN,GAAG,IAAI;gBACP,QAAQ,EAAE;oBACT,OAAO,EAAE,IAAI;oBACb,YAAY,EAAE,eAAe,CAAC,WAAW,EAAE,MAAM,CAAC;iBAClD;aAC6B,CAAC;QACjC,CAAC;QAED,SAAS,CAAC;YACT,uBAAuB;YACvB,MAAM,WAAW,GAAU,KAAK,CAAC,GAAG,CAAC;YACrC,MAAM,IAAI,KAAK,CAAC,sCAAsC,WAAW,EAAE,CAAC,CAAC;QACtE,CAAC;IACF,CAAC;AAAA,CACD;AAID,SAAS,iBAAiB,CAAC,KAAoC,EAAW;IACzE,2FAA2F;IAC3F,OAAO,KAAK,CAAC,EAAE,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;AAAA,CAClC;AAED,SAAS,mBAAmB,CAAC,KAAoC,EAAW;IAC3E,+FAA+F;IAC/F,OAAO,KAAK,CAAC,EAAE,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC;AAAA,CACpC;AAED,SAAS,uBAAuB,CAC/B,MAA8B,EAC9B,KAAoC,EACd;IACtB,IAAI,iBAAiB,CAAC,KAAK,CAAC,EAAE,CAAC;QAC9B,gDAAgD;QAChD,QAAQ,MAAM,EAAE,CAAC;YAChB,KAAK,SAAS,CAAC;YACf,KAAK,KAAK;gBACT,OAAO,KAAK,CAAC;YACd,KAAK,QAAQ,CAAC;YACd,KAAK,MAAM;gBACV,OAAO,MAAM,CAAC;QAChB,CAAC;IACF,CAAC;IACD,0CAA0C;IAC1C,QAAQ,MAAM,EAAE,CAAC;QAChB,KAAK,SAAS;YACb,OAAO,SAAS,CAAC;QAClB,KAAK,KAAK;YACT,OAAO,KAAK,CAAC;QACd,KAAK,QAAQ;YACZ,OAAO,QAAQ,CAAC;QACjB,KAAK,MAAM;YACV,OAAO,MAAM,CAAC;IAChB,CAAC;AAAA,CACD;AAED,SAAS,yBAAyB,CAAC,MAA8B,EAAE,OAAe,EAAuB;IACxG,IAAI,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,CAAC;QAC/B,gDAAgD;QAChD,QAAQ,MAAM,EAAE,CAAC;YAChB,KAAK,SAAS,CAAC;YACf,KAAK,KAAK;gBACT,OAAO,KAAK,CAAC;YACd,KAAK,QAAQ,CAAC;YACd,KAAK,MAAM;gBACV,OAAO,MAAM,CAAC;QAChB,CAAC;IACF,CAAC;IACD,0CAA0C;IAC1C,QAAQ,MAAM,EAAE,CAAC;QAChB,KAAK,SAAS;YACb,OAAO,SAAS,CAAC;QAClB,KAAK,KAAK;YACT,OAAO,KAAK,CAAC;QACd,KAAK,QAAQ;YACZ,OAAO,QAAQ,CAAC;QACjB,KAAK,MAAM;YACV,OAAO,MAAM,CAAC;IAChB,CAAC;AAAA,CACD;AAED,SAAS,eAAe,CAAC,KAAoC,EAAE,MAA8B,EAAU;IACtG,gEAAgE;IAChE,IAAI,KAAK,CAAC,EAAE,CAAC,QAAQ,CAAC,SAAS,CAAC,EAAE,CAAC;QAClC,MAAM,OAAO,GAA2C;YACvD,OAAO,EAAE,GAAG;YACZ,GAAG,EAAE,IAAI;YACT,MAAM,EAAE,IAAI;YACZ,IAAI,EAAE,KAAK;SACX,CAAC;QACF,OAAO,OAAO,CAAC,MAAM,CAAC,CAAC;IACxB,CAAC;IAED,IAAI,KAAK,CAAC,EAAE,CAAC,QAAQ,CAAC,WAAW,CAAC,EAAE,CAAC;QACpC,gCAAgC;QAChC,MAAM,OAAO,GAA2C;YACvD,OAAO,EAAE,GAAG;YACZ,GAAG,EAAE,IAAI;YACT,MAAM,EAAE,IAAI;YACZ,IAAI,EAAE,KAAK;SACX,CAAC;QACF,OAAO,OAAO,CAAC,MAAM,CAAC,CAAC;IACxB,CAAC;IAED,8BAA8B;IAC9B,OAAO,CAAC,CAAC,CAAC;AAAA,CACV","sourcesContent":["import { supportsXhigh } from \"./models.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 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\tReasoningEffort,\n\tSimpleStreamOptions,\n} from \"./types.js\";\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\t// Don't return a dummy value - require explicit auth.json configuration.\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\tzai: \"ZAI_API_KEY\",\n\t\tmistral: \"MISTRAL_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}\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 \"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}\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};\n\n\t// Helper to clamp xhigh to high for providers that don't support it\n\tconst clampReasoning = (effort: ReasoningEffort \| undefined) => (effort === \"xhigh\" ? \"high\" : effort);\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 anthropicBudgets = {\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\n\t\t\tconst minOutputTokens = 1024;\n\t\t\tlet thinkingBudget = anthropicBudgets[clampReasoning(options.reasoning)!];\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\tthinkingEnabled: true,\n\t\t\t\tthinkingBudgetTokens: thinkingBudget,\n\t\t\t} satisfies AnthropicOptions;\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 \"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),\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 budgets: Record<ClampedReasoningEffort, number> = {\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\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),\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 ClampedReasoningEffort = Exclude<ReasoningEffort, \"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: ClampedReasoningEffort,\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: ClampedReasoningEffort, 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(model: Model<\"google-generative-ai\">, effort: ClampedReasoningEffort): number {\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<ClampedReasoningEffort, 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<ClampedReasoningEffort, 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"]}

package/package.json CHANGED Viewed

@@ -1,6 +1,6 @@
 {
 	"name": "@mariozechner/pi-ai",
-	"version": "0.32.1",
+	"version": "0.32.3",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"type": "module",
 	"main": "./dist/index.js",