@fleetagent/pi-ai 0.1.8 → 0.2.0

package/dist/providers/google.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"google.d.ts","sourceRoot":"","sources":["../../src/providers/google.ts"],"names":[],"mappings":"AAQA,OAAO,KAAK,EAKX,mBAAmB,EACnB,cAAc,EACd,aAAa,EAMb,MAAM,aAAa,CAAC;AAGrB,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,oBAAoB,CAAC;AAW9D,MAAM,WAAW,aAAc,SAAQ,aAAa;IACnD,UAAU,CAAC,EAAE,MAAM,GAAG,MAAM,GAAG,KAAK,CAAC;IACrC,QAAQ,CAAC,EAAE;QACV,OAAO,EAAE,OAAO,CAAC;QACjB,YAAY,CAAC,EAAE,MAAM,CAAC;QACtB,KAAK,CAAC,EAAE,mBAAmB,CAAC;KAC5B,CAAC;CACF;AAKD,eAAO,MAAM,YAAY,EAAE,cAAc,CAAC,sBAAsB,EAAE,aAAa,CAoO9E,CAAC;AAEF,eAAO,MAAM,kBAAkB,EAAE,cAAc,CAAC,sBAAsB,EAAE,mBAAmB,CAoC1F,CAAC","sourcesContent":["import {\n\ttype GenerateContentConfig,\n\ttype GenerateContentParameters,\n\tGoogleGenAI,\n\ttype ThinkingConfig,\n} from \"@google/genai\";\nimport { getEnvApiKey } from \"../env-api-keys.ts\";\nimport { calculateCost, clampThinkingLevel } from \"../models.ts\";\nimport type {\n\tApi,\n\tAssistantMessage,\n\tContext,\n\tModel,\n\tSimpleStreamOptions,\n\tStreamFunction,\n\tStreamOptions,\n\tTextContent,\n\tThinkingBudgets,\n\tThinkingContent,\n\tThinkingLevel,\n\tToolCall,\n} from \"../types.ts\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.ts\";\nimport { sanitizeSurrogates } from \"../utils/sanitize-unicode.ts\";\nimport type { GoogleThinkingLevel } from \"./google-shared.ts\";\nimport {\n\tconvertMessages,\n\tconvertTools,\n\tisThinkingPart,\n\tmapStopReason,\n\tmapToolChoice,\n\tretainThoughtSignature,\n} from \"./google-shared.ts\";\nimport { buildBaseOptions } from \"./simple-options.ts\";\n\nexport interface GoogleOptions extends StreamOptions {\n\ttoolChoice?: \"auto\" | \"none\" | \"any\";\n\tthinking?: {\n\t\tenabled: boolean;\n\t\tbudgetTokens?: number; // -1 for dynamic, 0 to disable\n\t\tlevel?: GoogleThinkingLevel;\n\t};\n}\n\n// Counter for generating unique tool call IDs\nlet toolCallCounter = 0;\n\nexport const streamGoogle: StreamFunction<\"google-generative-ai\", GoogleOptions> = (\n\tmodel: Model<\"google-generative-ai\">,\n\tcontext: Context,\n\toptions?: GoogleOptions,\n): AssistantMessageEventStream => {\n\tconst stream = new AssistantMessageEventStream();\n\n\t(async () => {\n\t\tconst output: AssistantMessage = {\n\t\t\trole: \"assistant\",\n\t\t\tcontent: [],\n\t\t\tapi: \"google-generative-ai\" as Api,\n\t\t\tprovider: model.provider,\n\t\t\tmodel: model.id,\n\t\t\tusage: {\n\t\t\t\tinput: 0,\n\t\t\t\toutput: 0,\n\t\t\t\tcacheRead: 0,\n\t\t\t\tcacheWrite: 0,\n\t\t\t\ttotalTokens: 0,\n\t\t\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t\t\t},\n\t\t\tstopReason: \"stop\",\n\t\t\ttimestamp: Date.now(),\n\t\t};\n\n\t\ttry {\n\t\t\tconst apiKey = options?.apiKey || getEnvApiKey(model.provider) || \"\";\n\t\t\tconst client = createClient(model, apiKey, options?.headers);\n\t\t\tlet params = buildParams(model, context, options);\n\t\t\tconst nextParams = await options?.onPayload?.(params, model);\n\t\t\tif (nextParams !== undefined) {\n\t\t\t\tparams = nextParams as GenerateContentParameters;\n\t\t\t}\n\t\t\tconst googleStream = await client.models.generateContentStream(params);\n\n\t\t\tstream.push({ type: \"start\", partial: output });\n\t\t\tlet currentBlock: TextContent | ThinkingContent | null = null;\n\t\t\tconst blocks = output.content;\n\t\t\tconst blockIndex = () => blocks.length - 1;\n\t\t\tfor await (const chunk of googleStream) {\n\t\t\t\t// @google/genai documents GenerateContentResponse.responseId as an output-only field\n\t\t\t\t// used to identify each response. Keep the first non-empty one from the stream.\n\t\t\t\toutput.responseId ||= chunk.responseId;\n\t\t\t\tconst candidate = chunk.candidates?.[0];\n\t\t\t\tif (candidate?.content?.parts) {\n\t\t\t\t\tfor (const part of candidate.content.parts) {\n\t\t\t\t\t\tif (part.text !== undefined) {\n\t\t\t\t\t\t\tconst isThinking = isThinkingPart(part);\n\t\t\t\t\t\t\tif (\n\t\t\t\t\t\t\t\t!currentBlock ||\n\t\t\t\t\t\t\t\t(isThinking && currentBlock.type !== \"thinking\") ||\n\t\t\t\t\t\t\t\t(!isThinking && currentBlock.type !== \"text\")\n\t\t\t\t\t\t\t) {\n\t\t\t\t\t\t\t\tif (currentBlock) {\n\t\t\t\t\t\t\t\t\tif (currentBlock.type === \"text\") {\n\t\t\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\t\t\ttype: \"text_end\",\n\t\t\t\t\t\t\t\t\t\t\tcontentIndex: blocks.length - 1,\n\t\t\t\t\t\t\t\t\t\t\tcontent: currentBlock.text,\n\t\t\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\t\t\ttype: \"thinking_end\",\n\t\t\t\t\t\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\t\t\t\t\t\tcontent: currentBlock.thinking,\n\t\t\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\tif (isThinking) {\n\t\t\t\t\t\t\t\t\tcurrentBlock = { type: \"thinking\", thinking: \"\", thinkingSignature: undefined };\n\t\t\t\t\t\t\t\t\toutput.content.push(currentBlock);\n\t\t\t\t\t\t\t\t\tstream.push({ type: \"thinking_start\", contentIndex: blockIndex(), partial: output });\n\t\t\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\t\t\tcurrentBlock = { type: \"text\", text: \"\" };\n\t\t\t\t\t\t\t\t\toutput.content.push(currentBlock);\n\t\t\t\t\t\t\t\t\tstream.push({ type: \"text_start\", contentIndex: blockIndex(), partial: output });\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tif (currentBlock.type === \"thinking\") {\n\t\t\t\t\t\t\t\tcurrentBlock.thinking += part.text;\n\t\t\t\t\t\t\t\tcurrentBlock.thinkingSignature = retainThoughtSignature(\n\t\t\t\t\t\t\t\t\tcurrentBlock.thinkingSignature,\n\t\t\t\t\t\t\t\t\tpart.thoughtSignature,\n\t\t\t\t\t\t\t\t);\n\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\ttype: \"thinking_delta\",\n\t\t\t\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\t\t\t\tdelta: part.text,\n\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\t\tcurrentBlock.text += part.text;\n\t\t\t\t\t\t\t\tcurrentBlock.textSignature = retainThoughtSignature(\n\t\t\t\t\t\t\t\t\tcurrentBlock.textSignature,\n\t\t\t\t\t\t\t\t\tpart.thoughtSignature,\n\t\t\t\t\t\t\t\t);\n\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\ttype: \"text_delta\",\n\t\t\t\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\t\t\t\tdelta: part.text,\n\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\tif (part.functionCall) {\n\t\t\t\t\t\t\tif (currentBlock) {\n\t\t\t\t\t\t\t\tif (currentBlock.type === \"text\") {\n\t\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\t\ttype: \"text_end\",\n\t\t\t\t\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\t\t\t\t\tcontent: currentBlock.text,\n\t\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\t\t\ttype: \"thinking_end\",\n\t\t\t\t\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\t\t\t\t\tcontent: currentBlock.thinking,\n\t\t\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\tcurrentBlock = null;\n\t\t\t\t\t\t\t}\n\n\t\t\t\t\t\t\t// Generate unique ID if not provided or if it's a duplicate\n\t\t\t\t\t\t\tconst providedId = part.functionCall.id;\n\t\t\t\t\t\t\tconst needsNewId =\n\t\t\t\t\t\t\t\t!providedId || output.content.some((b) => b.type === \"toolCall\" && b.id === providedId);\n\t\t\t\t\t\t\tconst toolCallId = needsNewId\n\t\t\t\t\t\t\t\t? `${part.functionCall.name}_${Date.now()}_${++toolCallCounter}`\n\t\t\t\t\t\t\t\t: providedId;\n\n\t\t\t\t\t\t\tconst toolCall: ToolCall = {\n\t\t\t\t\t\t\t\ttype: \"toolCall\",\n\t\t\t\t\t\t\t\tid: toolCallId,\n\t\t\t\t\t\t\t\tname: part.functionCall.name || \"\",\n\t\t\t\t\t\t\t\targuments: (part.functionCall.args as Record<string, any>) ?? {},\n\t\t\t\t\t\t\t\t...(part.thoughtSignature && { thoughtSignature: part.thoughtSignature }),\n\t\t\t\t\t\t\t};\n\n\t\t\t\t\t\t\toutput.content.push(toolCall);\n\t\t\t\t\t\t\tstream.push({ type: \"toolcall_start\", contentIndex: blockIndex(), partial: output });\n\t\t\t\t\t\t\tstream.push({\n\t\t\t\t\t\t\t\ttype: \"toolcall_delta\",\n\t\t\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\t\t\tdelta: JSON.stringify(toolCall.arguments),\n\t\t\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\tstream.push({ type: \"toolcall_end\", contentIndex: blockIndex(), toolCall, partial: output });\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\tif (candidate?.finishReason) {\n\t\t\t\t\toutput.stopReason = mapStopReason(candidate.finishReason);\n\t\t\t\t\tif (output.content.some((b) => b.type === \"toolCall\")) {\n\t\t\t\t\t\toutput.stopReason = \"toolUse\";\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\tif (chunk.usageMetadata) {\n\t\t\t\t\toutput.usage = {\n\t\t\t\t\t\tinput:\n\t\t\t\t\t\t\t(chunk.usageMetadata.promptTokenCount || 0) - (chunk.usageMetadata.cachedContentTokenCount || 0),\n\t\t\t\t\t\toutput:\n\t\t\t\t\t\t\t(chunk.usageMetadata.candidatesTokenCount || 0) + (chunk.usageMetadata.thoughtsTokenCount || 0),\n\t\t\t\t\t\tcacheRead: chunk.usageMetadata.cachedContentTokenCount || 0,\n\t\t\t\t\t\tcacheWrite: 0,\n\t\t\t\t\t\ttotalTokens: chunk.usageMetadata.totalTokenCount || 0,\n\t\t\t\t\t\tcost: {\n\t\t\t\t\t\t\tinput: 0,\n\t\t\t\t\t\t\toutput: 0,\n\t\t\t\t\t\t\tcacheRead: 0,\n\t\t\t\t\t\t\tcacheWrite: 0,\n\t\t\t\t\t\t\ttotal: 0,\n\t\t\t\t\t\t},\n\t\t\t\t\t};\n\t\t\t\t\tcalculateCost(model, output.usage);\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tif (currentBlock) {\n\t\t\t\tif (currentBlock.type === \"text\") {\n\t\t\t\t\tstream.push({\n\t\t\t\t\t\ttype: \"text_end\",\n\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\tcontent: currentBlock.text,\n\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t});\n\t\t\t\t} else {\n\t\t\t\t\tstream.push({\n\t\t\t\t\t\ttype: \"thinking_end\",\n\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\tcontent: currentBlock.thinking,\n\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t});\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tif (options?.signal?.aborted) {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\n\t\t\tif (output.stopReason === \"aborted\" || output.stopReason === \"error\") {\n\t\t\t\tthrow new Error(\"An unknown error occurred\");\n\t\t\t}\n\n\t\t\tstream.push({ type: \"done\", reason: output.stopReason, message: output });\n\t\t\tstream.end();\n\t\t} catch (error) {\n\t\t\t// Remove internal index property used during streaming\n\t\t\tfor (const block of output.content) {\n\t\t\t\tif (\"index\" in block) {\n\t\t\t\t\tdelete (block as { index?: number }).index;\n\t\t\t\t}\n\t\t\t}\n\t\t\toutput.stopReason = options?.signal?.aborted ? \"aborted\" : \"error\";\n\t\t\toutput.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);\n\t\t\tstream.push({ type: \"error\", reason: output.stopReason, error: output });\n\t\t\tstream.end();\n\t\t}\n\t})();\n\n\treturn stream;\n};\n\nexport const streamSimpleGoogle: StreamFunction<\"google-generative-ai\", SimpleStreamOptions> = (\n\tmodel: Model<\"google-generative-ai\">,\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 base = buildBaseOptions(model, context, options, apiKey);\n\tif (!options?.reasoning) {\n\t\treturn streamGoogle(model, context, { ...base, thinking: { enabled: false } } satisfies GoogleOptions);\n\t}\n\n\tconst clampedReasoning = clampThinkingLevel(model, options.reasoning);\n\tconst effort = (clampedReasoning === \"off\" ? \"high\" : clampedReasoning) as ClampedThinkingLevel;\n\tconst googleModel = model as Model<\"google-generative-ai\">;\n\n\tif (isGemini3ProModel(googleModel) || isGemini3FlashModel(googleModel) || isGemma4Model(googleModel)) {\n\t\treturn streamGoogle(model, context, {\n\t\t\t...base,\n\t\t\tthinking: {\n\t\t\t\tenabled: true,\n\t\t\t\tlevel: getThinkingLevel(effort, googleModel),\n\t\t\t},\n\t\t} satisfies GoogleOptions);\n\t}\n\n\treturn streamGoogle(model, context, {\n\t\t...base,\n\t\tthinking: {\n\t\t\tenabled: true,\n\t\t\tbudgetTokens: getGoogleBudget(googleModel, effort, options.thinkingBudgets),\n\t\t},\n\t} satisfies GoogleOptions);\n};\n\nfunction createClient(\n\tmodel: Model<\"google-generative-ai\">,\n\tapiKey?: string,\n\toptionsHeaders?: Record<string, string>,\n): GoogleGenAI {\n\tconst httpOptions: { baseUrl?: string; apiVersion?: string; headers?: Record<string, string> } = {};\n\tif (model.baseUrl) {\n\t\thttpOptions.baseUrl = model.baseUrl;\n\t\thttpOptions.apiVersion = \"\"; // baseUrl already includes version path, don't append\n\t}\n\tif (model.headers || optionsHeaders) {\n\t\thttpOptions.headers = { ...model.headers, ...optionsHeaders };\n\t}\n\n\treturn new GoogleGenAI({\n\t\tapiKey,\n\t\thttpOptions: Object.keys(httpOptions).length > 0 ? httpOptions : undefined,\n\t});\n}\n\nfunction buildParams(\n\tmodel: Model<\"google-generative-ai\">,\n\tcontext: Context,\n\toptions: GoogleOptions = {},\n): GenerateContentParameters {\n\tconst contents = convertMessages(model, context);\n\n\tconst generationConfig: GenerateContentConfig = {};\n\tif (options.temperature !== undefined) {\n\t\tgenerationConfig.temperature = options.temperature;\n\t}\n\tif (options.maxTokens !== undefined) {\n\t\tgenerationConfig.maxOutputTokens = options.maxTokens;\n\t}\n\n\tconst config: GenerateContentConfig = {\n\t\t...(Object.keys(generationConfig).length > 0 && generationConfig),\n\t\t...(context.systemPrompt && { systemInstruction: sanitizeSurrogates(context.systemPrompt) }),\n\t\t...(context.tools && context.tools.length > 0 && { tools: convertTools(context.tools) }),\n\t};\n\n\tif (context.tools && context.tools.length > 0 && options.toolChoice) {\n\t\tconfig.toolConfig = {\n\t\t\tfunctionCallingConfig: {\n\t\t\t\tmode: mapToolChoice(options.toolChoice),\n\t\t\t},\n\t\t};\n\t} else {\n\t\tconfig.toolConfig = undefined;\n\t}\n\n\tif (options.thinking?.enabled && model.reasoning) {\n\t\tconst thinkingConfig: ThinkingConfig = { includeThoughts: true };\n\t\tif (options.thinking.level !== undefined) {\n\t\t\t// Cast to any since our GoogleThinkingLevel mirrors Google's ThinkingLevel enum values\n\t\t\tthinkingConfig.thinkingLevel = options.thinking.level as any;\n\t\t} else if (options.thinking.budgetTokens !== undefined) {\n\t\t\tthinkingConfig.thinkingBudget = options.thinking.budgetTokens;\n\t\t}\n\t\tconfig.thinkingConfig = thinkingConfig;\n\t} else if (model.reasoning && options.thinking && !options.thinking.enabled) {\n\t\tconfig.thinkingConfig = getDisabledThinkingConfig(model);\n\t}\n\n\tif (options.signal) {\n\t\tif (options.signal.aborted) {\n\t\t\tthrow new Error(\"Request aborted\");\n\t\t}\n\t\tconfig.abortSignal = options.signal;\n\t}\n\n\tconst params: GenerateContentParameters = {\n\t\tmodel: model.id,\n\t\tcontents,\n\t\tconfig,\n\t};\n\n\treturn params;\n}\n\ntype ClampedThinkingLevel = Exclude<ThinkingLevel, \"xhigh\">;\n\nfunction isGemma4Model(model: Model<\"google-generative-ai\">): boolean {\n\treturn /gemma-?4/.test(model.id.toLowerCase());\n}\n\nfunction isGemini3ProModel(model: Model<\"google-generative-ai\">): boolean {\n\treturn /gemini-3(?:\\.\\d+)?-pro/.test(model.id.toLowerCase());\n}\n\nfunction isGemini3FlashModel(model: Model<\"google-generative-ai\">): boolean {\n\treturn /gemini-3(?:\\.\\d+)?-flash/.test(model.id.toLowerCase());\n}\n\nfunction getDisabledThinkingConfig(model: Model<\"google-generative-ai\">): ThinkingConfig {\n\t// Google docs: Gemini 3.1 Pro cannot disable thinking, and Gemini 3 Flash / Flash-Lite\n\t// do not support full thinking-off either. For Gemini 3 models, use the lowest supported\n\t// thinkingLevel without includeThoughts so hidden thinking remains invisible to pi.\n\tif (isGemini3ProModel(model)) {\n\t\treturn { thinkingLevel: \"LOW\" as any };\n\t}\n\tif (isGemini3FlashModel(model)) {\n\t\treturn { thinkingLevel: \"MINIMAL\" as any };\n\t}\n\tif (isGemma4Model(model)) {\n\t\treturn { thinkingLevel: \"MINIMAL\" as any };\n\t}\n\n\t// Gemini 2.x supports disabling via thinkingBudget = 0.\n\treturn { thinkingBudget: 0 };\n}\n\nfunction getThinkingLevel(effort: ClampedThinkingLevel, model: Model<\"google-generative-ai\">): GoogleThinkingLevel {\n\tif (isGemini3ProModel(model)) {\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\tif (isGemma4Model(model)) {\n\t\tswitch (effort) {\n\t\t\tcase \"minimal\":\n\t\t\tcase \"low\":\n\t\t\t\treturn \"MINIMAL\";\n\t\t\tcase \"medium\":\n\t\t\tcase \"high\":\n\t\t\t\treturn \"HIGH\";\n\t\t}\n\t}\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\tif (customBudgets?.[effort] !== undefined) {\n\t\treturn customBudgets[effort]!;\n\t}\n\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-lite\")) {\n\t\tconst budgets: Record<ClampedThinkingLevel, number> = {\n\t\t\tminimal: 512,\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\tif (model.id.includes(\"2.5-flash\")) {\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\treturn -1;\n}\n"]}
+{"version":3,"file":"google.d.ts","sourceRoot":"","sources":["../../src/providers/google.ts"],"names":[],"mappings":"AAWA,OAAO,KAAK,EAMX,mBAAmB,EACnB,cAAc,EACd,aAAa,EAMb,MAAM,aAAa,CAAC;AAIrB,OAAO,KAAK,EAGX,qBAAqB,EACrB,gBAAgB,EAChB,MAAM,oBAAoB,CAAC;AAY5B,YAAY,EAAE,qBAAqB,EAAE,MAAM,oBAAoB,CAAC;AAEhE,MAAM,WAAW,aAAc,SAAQ,aAAa;IACnD,UAAU,CAAC,EAAE,gBAAgB,CAAC;IAC9B,QAAQ,CAAC,EAAE,qBAAqB,CAAC;CACjC;AA6MD,eAAO,MAAM,YAAY,EAAE,cAAc,CAAC,sBAAsB,EAAE,aAAa,CAa9E,CAAC;AAEF,eAAO,MAAM,kBAAkB,EAAE,cAAc,CAAC,sBAAsB,EAAE,mBAAmB,CAoC1F,CAAC","sourcesContent":["import {\n\ttype Candidate,\n\ttype GenerateContentConfig,\n\ttype GenerateContentParameters,\n\ttype GenerateContentResponse,\n\tGoogleGenAI,\n\ttype Part,\n\ttype ThinkingConfig,\n} from \"@google/genai\";\nimport { getEnvApiKey } from \"../env-api-keys.ts\";\nimport { calculateCost, clampThinkingLevel } from \"../models.ts\";\nimport type {\n\tApi,\n\tAssistantMessage,\n\tContext,\n\tModel,\n\tProviderHeaders,\n\tSimpleStreamOptions,\n\tStreamFunction,\n\tStreamOptions,\n\tTextContent,\n\tThinkingBudgets,\n\tThinkingContent,\n\tThinkingLevel,\n\tToolCall,\n} from \"../types.ts\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.ts\";\nimport { providerHeadersToRecord } from \"../utils/headers.ts\";\nimport { sanitizeSurrogates } from \"../utils/sanitize-unicode.ts\";\nimport type {\n\tGoogleStreamState,\n\tGoogleThinkingLevel,\n\tGoogleThinkingOptions,\n\tGoogleToolChoice,\n} from \"./google-shared.ts\";\nimport {\n\tconvertMessages,\n\tconvertTools,\n\tisThinkingPart,\n\tmapStopReason,\n\tmapToolChoice,\n\tretainThoughtSignature,\n\tretryGoogleRequest,\n} from \"./google-shared.ts\";\nimport { buildBaseOptions } from \"./simple-options.ts\";\n\nexport type { GoogleThinkingOptions } from \"./google-shared.ts\";\n\nexport interface GoogleOptions extends StreamOptions {\n\ttoolChoice?: GoogleToolChoice;\n\tthinking?: GoogleThinkingOptions;\n}\n\n// Counter for generating unique tool call IDs\nlet toolCallCounter = 0;\n\ntype GoogleGenerativeStreamState = GoogleStreamState;\n\nfunction createGoogleOutput(model: Model<\"google-generative-ai\">): AssistantMessage {\n\treturn {\n\t\trole: \"assistant\",\n\t\tcontent: [],\n\t\tapi: \"google-generative-ai\" as Api,\n\t\tprovider: model.provider,\n\t\tmodel: model.id,\n\t\tusage: {\n\t\t\tinput: 0,\n\t\t\toutput: 0,\n\t\t\tcacheRead: 0,\n\t\t\tcacheWrite: 0,\n\t\t\ttotalTokens: 0,\n\t\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t\t},\n\t\tstopReason: \"stop\",\n\t\ttimestamp: Date.now(),\n\t};\n}\n\nfunction googleBlockIndex(state: GoogleGenerativeStreamState): number {\n\treturn state.output.content.length - 1;\n}\n\nfunction endGoogleBlock(state: GoogleGenerativeStreamState): void {\n\tconst block = state.currentBlock;\n\tif (!block) return;\n\tif (block.type === \"text\") {\n\t\tstate.stream.push({\n\t\t\ttype: \"text_end\",\n\t\t\tcontentIndex: googleBlockIndex(state),\n\t\t\tcontent: block.text,\n\t\t\tpartial: state.output,\n\t\t});\n\t} else {\n\t\tstate.stream.push({\n\t\t\ttype: \"thinking_end\",\n\t\t\tcontentIndex: googleBlockIndex(state),\n\t\t\tcontent: block.thinking,\n\t\t\tpartial: state.output,\n\t\t});\n\t}\n\tstate.currentBlock = null;\n}\n\nfunction ensureGoogleTextBlock(state: GoogleGenerativeStreamState, isThinking: boolean): TextContent | ThinkingContent {\n\tconst current = state.currentBlock;\n\tif (current && ((isThinking && current.type === \"thinking\") || (!isThinking && current.type === \"text\"))) {\n\t\treturn current;\n\t}\n\tendGoogleBlock(state);\n\tif (isThinking) {\n\t\tconst block: ThinkingContent = { type: \"thinking\", thinking: \"\", thinkingSignature: undefined };\n\t\tstate.currentBlock = block;\n\t\tstate.output.content.push(block);\n\t\tstate.stream.push({ type: \"thinking_start\", contentIndex: googleBlockIndex(state), partial: state.output });\n\t\treturn block;\n\t}\n\tconst block: TextContent = { type: \"text\", text: \"\" };\n\tstate.currentBlock = block;\n\tstate.output.content.push(block);\n\tstate.stream.push({ type: \"text_start\", contentIndex: googleBlockIndex(state), partial: state.output });\n\treturn block;\n}\n\nfunction appendGoogleTextPart(state: GoogleGenerativeStreamState, part: Part): void {\n\tif (part.text === undefined) return;\n\tconst block = ensureGoogleTextBlock(state, isThinkingPart(part));\n\tif (block.type === \"thinking\") {\n\t\tblock.thinking += part.text;\n\t\tblock.thinkingSignature = retainThoughtSignature(block.thinkingSignature, part.thoughtSignature);\n\t\tstate.stream.push({\n\t\t\ttype: \"thinking_delta\",\n\t\t\tcontentIndex: googleBlockIndex(state),\n\t\t\tdelta: part.text,\n\t\t\tpartial: state.output,\n\t\t});\n\t\treturn;\n\t}\n\tblock.text += part.text;\n\tblock.textSignature = retainThoughtSignature(block.textSignature, part.thoughtSignature);\n\tstate.stream.push({\n\t\ttype: \"text_delta\",\n\t\tcontentIndex: googleBlockIndex(state),\n\t\tdelta: part.text,\n\t\tpartial: state.output,\n\t});\n}\n\nfunction appendGoogleToolCall(state: GoogleGenerativeStreamState, part: Part): void {\n\tif (!part.functionCall) return;\n\tendGoogleBlock(state);\n\tconst providedId = part.functionCall.id;\n\tconst needsNewId =\n\t\t!providedId || state.output.content.some((block) => block.type === \"toolCall\" && block.id === providedId);\n\tconst toolCall: ToolCall = {\n\t\ttype: \"toolCall\",\n\t\tid: needsNewId ? `${part.functionCall.name}_${Date.now()}_${++toolCallCounter}` : providedId,\n\t\tname: part.functionCall.name || \"\",\n\t\targuments: (part.functionCall.args as Record<string, any>) ?? {},\n\t\t...(part.thoughtSignature && { thoughtSignature: part.thoughtSignature }),\n\t};\n\tstate.output.content.push(toolCall);\n\tstate.stream.push({ type: \"toolcall_start\", contentIndex: googleBlockIndex(state), partial: state.output });\n\tstate.stream.push({\n\t\ttype: \"toolcall_delta\",\n\t\tcontentIndex: googleBlockIndex(state),\n\t\tdelta: JSON.stringify(toolCall.arguments),\n\t\tpartial: state.output,\n\t});\n\tstate.stream.push({\n\t\ttype: \"toolcall_end\",\n\t\tcontentIndex: googleBlockIndex(state),\n\t\ttoolCall,\n\t\tpartial: state.output,\n\t});\n}\n\nfunction applyGoogleFinishReason(output: AssistantMessage, candidate: Candidate | undefined): void {\n\tif (!candidate?.finishReason) return;\n\toutput.stopReason = mapStopReason(candidate.finishReason);\n\tif (output.content.some((block) => block.type === \"toolCall\")) output.stopReason = \"toolUse\";\n}\n\nfunction applyGoogleUsage(\n\toutput: AssistantMessage,\n\tchunk: GenerateContentResponse,\n\tmodel: Model<\"google-generative-ai\">,\n): void {\n\tif (!chunk.usageMetadata) return;\n\toutput.usage = {\n\t\tinput: (chunk.usageMetadata.promptTokenCount || 0) - (chunk.usageMetadata.cachedContentTokenCount || 0),\n\t\toutput: (chunk.usageMetadata.candidatesTokenCount || 0) + (chunk.usageMetadata.thoughtsTokenCount || 0),\n\t\tcacheRead: chunk.usageMetadata.cachedContentTokenCount || 0,\n\t\tcacheWrite: 0,\n\t\ttotalTokens: chunk.usageMetadata.totalTokenCount || 0,\n\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t};\n\tcalculateCost(model, output.usage);\n}\n\nfunction processGoogleChunk(\n\tstate: GoogleGenerativeStreamState,\n\tchunk: GenerateContentResponse,\n\tmodel: Model<\"google-generative-ai\">,\n): void {\n\tstate.output.responseId ||= chunk.responseId;\n\tconst candidate = chunk.candidates?.[0];\n\tfor (const part of candidate?.content?.parts ?? []) {\n\t\tappendGoogleTextPart(state, part);\n\t\tappendGoogleToolCall(state, part);\n\t}\n\tapplyGoogleFinishReason(state.output, candidate);\n\tapplyGoogleUsage(state.output, chunk, model);\n}\n\nfunction completeGoogleStream(state: GoogleGenerativeStreamState, signal: AbortSignal | undefined): void {\n\tendGoogleBlock(state);\n\tif (signal?.aborted) throw new Error(\"Request was aborted\");\n\tif (state.output.stopReason === \"pending\") throw new Error(\"Google stream ended without a finish reason\");\n\tif (state.output.stopReason === \"aborted\" || state.output.stopReason === \"error\") {\n\t\tthrow new Error(\"An unknown error occurred\");\n\t}\n\tstate.stream.push({ type: \"done\", reason: state.output.stopReason, message: state.output });\n\tstate.stream.end();\n}\n\nfunction failGoogleStream(state: GoogleGenerativeStreamState, error: unknown, signal: AbortSignal | undefined): void {\n\tfor (const block of state.output.content) {\n\t\tif (\"index\" in block) delete (block as { index?: number }).index;\n\t}\n\tstate.output.stopReason = signal?.aborted ? \"aborted\" : \"error\";\n\tstate.output.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);\n\tstate.stream.push({ type: \"error\", reason: state.output.stopReason, error: state.output });\n\tstate.stream.end();\n}\n\nasync function runGoogleStream(\n\tmodel: Model<\"google-generative-ai\">,\n\tcontext: Context,\n\toptions: GoogleOptions | undefined,\n\tstate: GoogleGenerativeStreamState,\n): Promise<void> {\n\ttry {\n\t\tconst apiKey = options?.apiKey || getEnvApiKey(model.provider) || \"\";\n\t\tconst client = createClient(model, apiKey, options?.headers);\n\t\tlet params = buildParams(model, context, options);\n\t\tconst nextParams = await options?.onPayload?.(params, model);\n\t\tif (nextParams !== undefined) params = nextParams as GenerateContentParameters;\n\t\tconst googleStream = await retryGoogleRequest(() => client.models.generateContentStream(params), options);\n\t\tstate.stream.push({ type: \"start\", partial: state.output });\n\t\tfor await (const chunk of googleStream) processGoogleChunk(state, chunk, model);\n\t\tcompleteGoogleStream(state, options?.signal);\n\t} catch (error) {\n\t\tfailGoogleStream(state, error, options?.signal);\n\t}\n}\n\nexport const streamGoogle: StreamFunction<\"google-generative-ai\", GoogleOptions> = (\n\tmodel: Model<\"google-generative-ai\">,\n\tcontext: Context,\n\toptions?: GoogleOptions,\n): AssistantMessageEventStream => {\n\tconst stream = new AssistantMessageEventStream();\n\tconst state: GoogleGenerativeStreamState = {\n\t\toutput: createGoogleOutput(model),\n\t\tstream,\n\t\tcurrentBlock: null,\n\t};\n\tvoid runGoogleStream(model, context, options, state);\n\treturn stream;\n};\n\nexport const streamSimpleGoogle: StreamFunction<\"google-generative-ai\", SimpleStreamOptions> = (\n\tmodel: Model<\"google-generative-ai\">,\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 base = buildBaseOptions(model, context, options, apiKey);\n\tif (!options?.reasoning) {\n\t\treturn streamGoogle(model, context, { ...base, thinking: { enabled: false } } satisfies GoogleOptions);\n\t}\n\n\tconst clampedReasoning = clampThinkingLevel(model, options.reasoning);\n\tconst effort = (clampedReasoning === \"off\" ? \"high\" : clampedReasoning) as ClampedThinkingLevel;\n\tconst googleModel = model as Model<\"google-generative-ai\">;\n\n\tif (isGemini3ProModel(googleModel) || isGemini3FlashModel(googleModel) || isGemma4Model(googleModel)) {\n\t\treturn streamGoogle(model, context, {\n\t\t\t...base,\n\t\t\tthinking: {\n\t\t\t\tenabled: true,\n\t\t\t\tlevel: getThinkingLevel(effort, googleModel),\n\t\t\t},\n\t\t} satisfies GoogleOptions);\n\t}\n\n\treturn streamGoogle(model, context, {\n\t\t...base,\n\t\tthinking: {\n\t\t\tenabled: true,\n\t\t\tbudgetTokens: getGoogleBudget(googleModel, effort, options.thinkingBudgets),\n\t\t},\n\t} satisfies GoogleOptions);\n};\n\nfunction createClient(\n\tmodel: Model<\"google-generative-ai\">,\n\tapiKey?: string,\n\toptionsHeaders?: ProviderHeaders,\n): GoogleGenAI {\n\tconst httpOptions: { baseUrl?: string; apiVersion?: string; headers?: Record<string, string> } = {};\n\tif (model.baseUrl) {\n\t\thttpOptions.baseUrl = model.baseUrl;\n\t\thttpOptions.apiVersion = \"\"; // baseUrl already includes version path, don't append\n\t}\n\tconst headers = providerHeadersToRecord({ ...model.headers, ...optionsHeaders });\n\tif (headers) {\n\t\thttpOptions.headers = headers;\n\t}\n\n\treturn new GoogleGenAI({\n\t\tapiKey,\n\t\thttpOptions: Object.keys(httpOptions).length > 0 ? httpOptions : undefined,\n\t});\n}\n\nfunction resolveGoogleThinkingConfig(\n\tmodel: Model<\"google-generative-ai\">,\n\tthinking: GoogleThinkingOptions | undefined,\n): ThinkingConfig | undefined {\n\tif (thinking?.enabled && model.reasoning) {\n\t\tconst config: ThinkingConfig = { includeThoughts: true };\n\t\tif (thinking.level !== undefined) {\n\t\t\t// Cast to any since our GoogleThinkingLevel mirrors Google's ThinkingLevel enum values\n\t\t\tconfig.thinkingLevel = thinking.level as any;\n\t\t} else if (thinking.budgetTokens !== undefined) {\n\t\t\tconfig.thinkingBudget = thinking.budgetTokens;\n\t\t}\n\t\treturn config;\n\t}\n\tif (model.reasoning && thinking && !thinking.enabled) return getDisabledThinkingConfig(model);\n\treturn undefined;\n}\n\nfunction buildParams(\n\tmodel: Model<\"google-generative-ai\">,\n\tcontext: Context,\n\toptions: GoogleOptions = {},\n): GenerateContentParameters {\n\tconst contents = convertMessages(model, context);\n\n\tconst generationConfig: GenerateContentConfig = {};\n\tif (options.temperature !== undefined) {\n\t\tgenerationConfig.temperature = options.temperature;\n\t}\n\tif (options.maxTokens !== undefined) {\n\t\tgenerationConfig.maxOutputTokens = options.maxTokens;\n\t}\n\n\tconst config: GenerateContentConfig = {\n\t\t...(Object.keys(generationConfig).length > 0 && generationConfig),\n\t\t...(context.systemPrompt && { systemInstruction: sanitizeSurrogates(context.systemPrompt) }),\n\t\t...(context.tools && context.tools.length > 0 && { tools: convertTools(context.tools) }),\n\t};\n\n\tif (context.tools && context.tools.length > 0 && options.toolChoice) {\n\t\tconfig.toolConfig = {\n\t\t\tfunctionCallingConfig: {\n\t\t\t\tmode: mapToolChoice(options.toolChoice),\n\t\t\t},\n\t\t};\n\t} else {\n\t\tconfig.toolConfig = undefined;\n\t}\n\n\tconst thinkingConfig = resolveGoogleThinkingConfig(model, options.thinking);\n\tif (thinkingConfig) config.thinkingConfig = thinkingConfig;\n\n\tif (options.signal) {\n\t\tif (options.signal.aborted) {\n\t\t\tthrow new Error(\"Request aborted\");\n\t\t}\n\t\tconfig.abortSignal = options.signal;\n\t}\n\n\tconst params: GenerateContentParameters = {\n\t\tmodel: model.id,\n\t\tcontents,\n\t\tconfig,\n\t};\n\n\treturn params;\n}\n\ntype ClampedThinkingLevel = Exclude<ThinkingLevel, \"xhigh\">;\n\nfunction isGemma4Model(model: Model<\"google-generative-ai\">): boolean {\n\treturn /gemma-?4/.test(model.id.toLowerCase());\n}\n\nfunction isGemini3ProModel(model: Model<\"google-generative-ai\">): boolean {\n\treturn /gemini-3(?:\\.\\d+)?-pro/.test(model.id.toLowerCase());\n}\n\nfunction isGemini3FlashModel(model: Model<\"google-generative-ai\">): boolean {\n\treturn /gemini-3(?:\\.\\d+)?-flash/.test(model.id.toLowerCase());\n}\n\nfunction getDisabledThinkingConfig(model: Model<\"google-generative-ai\">): ThinkingConfig {\n\t// Google docs: Gemini 3.1 Pro cannot disable thinking, and Gemini 3 Flash / Flash-Lite\n\t// do not support full thinking-off either. For Gemini 3 models, use the lowest supported\n\t// thinkingLevel without includeThoughts so hidden thinking remains invisible to pi.\n\tif (isGemini3ProModel(model)) {\n\t\treturn { thinkingLevel: \"LOW\" as any };\n\t}\n\tif (isGemini3FlashModel(model)) {\n\t\treturn { thinkingLevel: \"MINIMAL\" as any };\n\t}\n\tif (isGemma4Model(model)) {\n\t\treturn { thinkingLevel: \"MINIMAL\" as any };\n\t}\n\n\t// Gemini 2.x supports disabling via thinkingBudget = 0.\n\treturn { thinkingBudget: 0 };\n}\n\nfunction getThinkingLevel(effort: ClampedThinkingLevel, model: Model<\"google-generative-ai\">): GoogleThinkingLevel {\n\tif (isGemini3ProModel(model)) {\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\tif (isGemma4Model(model)) {\n\t\tswitch (effort) {\n\t\t\tcase \"minimal\":\n\t\t\tcase \"low\":\n\t\t\t\treturn \"MINIMAL\";\n\t\t\tcase \"medium\":\n\t\t\tcase \"high\":\n\t\t\t\treturn \"HIGH\";\n\t\t}\n\t}\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\tif (customBudgets?.[effort] !== undefined) {\n\t\treturn customBudgets[effort]!;\n\t}\n\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-lite\")) {\n\t\tconst budgets: Record<ClampedThinkingLevel, number> = {\n\t\t\tminimal: 512,\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\tif (model.id.includes(\"2.5-flash\")) {\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\treturn -1;\n}\n"]}

package/dist/providers/google.js

@@ -2,215 +2,205 @@ import { GoogleGenAI, } from "@google/genai";
 import { getEnvApiKey } from "../env-api-keys.js";
 import { calculateCost, clampThinkingLevel } from "../models.js";
 import { AssistantMessageEventStream } from "../utils/event-stream.js";
+import { providerHeadersToRecord } from "../utils/headers.js";
 import { sanitizeSurrogates } from "../utils/sanitize-unicode.js";
-import { convertMessages, convertTools, isThinkingPart, mapStopReason, mapToolChoice, retainThoughtSignature, } from "./google-shared.js";
+import { convertMessages, convertTools, isThinkingPart, mapStopReason, mapToolChoice, retainThoughtSignature, retryGoogleRequest, } from "./google-shared.js";
 import { buildBaseOptions } from "./simple-options.js";
 // Counter for generating unique tool call IDs
 let toolCallCounter = 0;
+function createGoogleOutput(model) {
+    return {
+        role: "assistant",
+        content: [],
+        api: "google-generative-ai",
+        provider: model.provider,
+        model: model.id,
+        usage: {
+            input: 0,
+            output: 0,
+            cacheRead: 0,
+            cacheWrite: 0,
+            totalTokens: 0,
+            cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
+        },
+        stopReason: "stop",
+        timestamp: Date.now(),
+    };
+}
+function googleBlockIndex(state) {
+    return state.output.content.length - 1;
+}
+function endGoogleBlock(state) {
+    const block = state.currentBlock;
+    if (!block)
+        return;
+    if (block.type === "text") {
+        state.stream.push({
+            type: "text_end",
+            contentIndex: googleBlockIndex(state),
+            content: block.text,
+            partial: state.output,
+        });
+    }
+    else {
+        state.stream.push({
+            type: "thinking_end",
+            contentIndex: googleBlockIndex(state),
+            content: block.thinking,
+            partial: state.output,
+        });
+    }
+    state.currentBlock = null;
+}
+function ensureGoogleTextBlock(state, isThinking) {
+    const current = state.currentBlock;
+    if (current && ((isThinking && current.type === "thinking") || (!isThinking && current.type === "text"))) {
+        return current;
+    }
+    endGoogleBlock(state);
+    if (isThinking) {
+        const block = { type: "thinking", thinking: "", thinkingSignature: undefined };
+        state.currentBlock = block;
+        state.output.content.push(block);
+        state.stream.push({ type: "thinking_start", contentIndex: googleBlockIndex(state), partial: state.output });
+        return block;
+    }
+    const block = { type: "text", text: "" };
+    state.currentBlock = block;
+    state.output.content.push(block);
+    state.stream.push({ type: "text_start", contentIndex: googleBlockIndex(state), partial: state.output });
+    return block;
+}
+function appendGoogleTextPart(state, part) {
+    if (part.text === undefined)
+        return;
+    const block = ensureGoogleTextBlock(state, isThinkingPart(part));
+    if (block.type === "thinking") {
+        block.thinking += part.text;
+        block.thinkingSignature = retainThoughtSignature(block.thinkingSignature, part.thoughtSignature);
+        state.stream.push({
+            type: "thinking_delta",
+            contentIndex: googleBlockIndex(state),
+            delta: part.text,
+            partial: state.output,
+        });
+        return;
+    }
+    block.text += part.text;
+    block.textSignature = retainThoughtSignature(block.textSignature, part.thoughtSignature);
+    state.stream.push({
+        type: "text_delta",
+        contentIndex: googleBlockIndex(state),
+        delta: part.text,
+        partial: state.output,
+    });
+}
+function appendGoogleToolCall(state, part) {
+    if (!part.functionCall)
+        return;
+    endGoogleBlock(state);
+    const providedId = part.functionCall.id;
+    const needsNewId = !providedId || state.output.content.some((block) => block.type === "toolCall" && block.id === providedId);
+    const toolCall = {
+        type: "toolCall",
+        id: needsNewId ? `${part.functionCall.name}_${Date.now()}_${++toolCallCounter}` : providedId,
+        name: part.functionCall.name || "",
+        arguments: part.functionCall.args ?? {},
+        ...(part.thoughtSignature && { thoughtSignature: part.thoughtSignature }),
+    };
+    state.output.content.push(toolCall);
+    state.stream.push({ type: "toolcall_start", contentIndex: googleBlockIndex(state), partial: state.output });
+    state.stream.push({
+        type: "toolcall_delta",
+        contentIndex: googleBlockIndex(state),
+        delta: JSON.stringify(toolCall.arguments),
+        partial: state.output,
+    });
+    state.stream.push({
+        type: "toolcall_end",
+        contentIndex: googleBlockIndex(state),
+        toolCall,
+        partial: state.output,
+    });
+}
+function applyGoogleFinishReason(output, candidate) {
+    if (!candidate?.finishReason)
+        return;
+    output.stopReason = mapStopReason(candidate.finishReason);
+    if (output.content.some((block) => block.type === "toolCall"))
+        output.stopReason = "toolUse";
+}
+function applyGoogleUsage(output, chunk, model) {
+    if (!chunk.usageMetadata)
+        return;
+    output.usage = {
+        input: (chunk.usageMetadata.promptTokenCount || 0) - (chunk.usageMetadata.cachedContentTokenCount || 0),
+        output: (chunk.usageMetadata.candidatesTokenCount || 0) + (chunk.usageMetadata.thoughtsTokenCount || 0),
+        cacheRead: chunk.usageMetadata.cachedContentTokenCount || 0,
+        cacheWrite: 0,
+        totalTokens: chunk.usageMetadata.totalTokenCount || 0,
+        cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
+    };
+    calculateCost(model, output.usage);
+}
+function processGoogleChunk(state, chunk, model) {
+    state.output.responseId ||= chunk.responseId;
+    const candidate = chunk.candidates?.[0];
+    for (const part of candidate?.content?.parts ?? []) {
+        appendGoogleTextPart(state, part);
+        appendGoogleToolCall(state, part);
+    }
+    applyGoogleFinishReason(state.output, candidate);
+    applyGoogleUsage(state.output, chunk, model);
+}
+function completeGoogleStream(state, signal) {
+    endGoogleBlock(state);
+    if (signal?.aborted)
+        throw new Error("Request was aborted");
+    if (state.output.stopReason === "pending")
+        throw new Error("Google stream ended without a finish reason");
+    if (state.output.stopReason === "aborted" || state.output.stopReason === "error") {
+        throw new Error("An unknown error occurred");
+    }
+    state.stream.push({ type: "done", reason: state.output.stopReason, message: state.output });
+    state.stream.end();
+}
+function failGoogleStream(state, error, signal) {
+    for (const block of state.output.content) {
+        if ("index" in block)
+            delete block.index;
+    }
+    state.output.stopReason = signal?.aborted ? "aborted" : "error";
+    state.output.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);
+    state.stream.push({ type: "error", reason: state.output.stopReason, error: state.output });
+    state.stream.end();
+}
+async function runGoogleStream(model, context, options, state) {
+    try {
+        const apiKey = options?.apiKey || getEnvApiKey(model.provider) || "";
+        const client = createClient(model, apiKey, options?.headers);
+        let params = buildParams(model, context, options);
+        const nextParams = await options?.onPayload?.(params, model);
+        if (nextParams !== undefined)
+            params = nextParams;
+        const googleStream = await retryGoogleRequest(() => client.models.generateContentStream(params), options);
+        state.stream.push({ type: "start", partial: state.output });
+        for await (const chunk of googleStream)
+            processGoogleChunk(state, chunk, model);
+        completeGoogleStream(state, options?.signal);
+    }
+    catch (error) {
+        failGoogleStream(state, error, options?.signal);
+    }
+}
 export const streamGoogle = (model, context, options) => {
     const stream = new AssistantMessageEventStream();
-    (async () => {
-        const output = {
-            role: "assistant",
-            content: [],
-            api: "google-generative-ai",
-            provider: model.provider,
-            model: model.id,
-            usage: {
-                input: 0,
-                output: 0,
-                cacheRead: 0,
-                cacheWrite: 0,
-                totalTokens: 0,
-                cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
-            },
-            stopReason: "stop",
-            timestamp: Date.now(),
-        };
-        try {
-            const apiKey = options?.apiKey || getEnvApiKey(model.provider) || "";
-            const client = createClient(model, apiKey, options?.headers);
-            let params = buildParams(model, context, options);
-            const nextParams = await options?.onPayload?.(params, model);
-            if (nextParams !== undefined) {
-                params = nextParams;
-            }
-            const googleStream = await client.models.generateContentStream(params);
-            stream.push({ type: "start", partial: output });
-            let currentBlock = null;
-            const blocks = output.content;
-            const blockIndex = () => blocks.length - 1;
-            for await (const chunk of googleStream) {
-                // @google/genai documents GenerateContentResponse.responseId as an output-only field
-                // used to identify each response. Keep the first non-empty one from the stream.
-                output.responseId ||= chunk.responseId;
-                const candidate = chunk.candidates?.[0];
-                if (candidate?.content?.parts) {
-                    for (const part of candidate.content.parts) {
-                        if (part.text !== undefined) {
-                            const isThinking = isThinkingPart(part);
-                            if (!currentBlock ||
-                                (isThinking && currentBlock.type !== "thinking") ||
-                                (!isThinking && currentBlock.type !== "text")) {
-                                if (currentBlock) {
-                                    if (currentBlock.type === "text") {
-                                        stream.push({
-                                            type: "text_end",
-                                            contentIndex: blocks.length - 1,
-                                            content: currentBlock.text,
-                                            partial: output,
-                                        });
-                                    }
-                                    else {
-                                        stream.push({
-                                            type: "thinking_end",
-                                            contentIndex: blockIndex(),
-                                            content: currentBlock.thinking,
-                                            partial: output,
-                                        });
-                                    }
-                                }
-                                if (isThinking) {
-                                    currentBlock = { type: "thinking", thinking: "", thinkingSignature: undefined };
-                                    output.content.push(currentBlock);
-                                    stream.push({ type: "thinking_start", contentIndex: blockIndex(), partial: output });
-                                }
-                                else {
-                                    currentBlock = { type: "text", text: "" };
-                                    output.content.push(currentBlock);
-                                    stream.push({ type: "text_start", contentIndex: blockIndex(), partial: output });
-                                }
-                            }
-                            if (currentBlock.type === "thinking") {
-                                currentBlock.thinking += part.text;
-                                currentBlock.thinkingSignature = retainThoughtSignature(currentBlock.thinkingSignature, part.thoughtSignature);
-                                stream.push({
-                                    type: "thinking_delta",
-                                    contentIndex: blockIndex(),
-                                    delta: part.text,
-                                    partial: output,
-                                });
-                            }
-                            else {
-                                currentBlock.text += part.text;
-                                currentBlock.textSignature = retainThoughtSignature(currentBlock.textSignature, part.thoughtSignature);
-                                stream.push({
-                                    type: "text_delta",
-                                    contentIndex: blockIndex(),
-                                    delta: part.text,
-                                    partial: output,
-                                });
-                            }
-                        }
-                        if (part.functionCall) {
-                            if (currentBlock) {
-                                if (currentBlock.type === "text") {
-                                    stream.push({
-                                        type: "text_end",
-                                        contentIndex: blockIndex(),
-                                        content: currentBlock.text,
-                                        partial: output,
-                                    });
-                                }
-                                else {
-                                    stream.push({
-                                        type: "thinking_end",
-                                        contentIndex: blockIndex(),
-                                        content: currentBlock.thinking,
-                                        partial: output,
-                                    });
-                                }
-                                currentBlock = null;
-                            }
-                            // Generate unique ID if not provided or if it's a duplicate
-                            const providedId = part.functionCall.id;
-                            const needsNewId = !providedId || output.content.some((b) => b.type === "toolCall" && b.id === providedId);
-                            const toolCallId = needsNewId
-                                ? `${part.functionCall.name}_${Date.now()}_${++toolCallCounter}`
-                                : providedId;
-                            const toolCall = {
-                                type: "toolCall",
-                                id: toolCallId,
-                                name: part.functionCall.name || "",
-                                arguments: part.functionCall.args ?? {},
-                                ...(part.thoughtSignature && { thoughtSignature: part.thoughtSignature }),
-                            };
-                            output.content.push(toolCall);
-                            stream.push({ type: "toolcall_start", contentIndex: blockIndex(), partial: output });
-                            stream.push({
-                                type: "toolcall_delta",
-                                contentIndex: blockIndex(),
-                                delta: JSON.stringify(toolCall.arguments),
-                                partial: output,
-                            });
-                            stream.push({ type: "toolcall_end", contentIndex: blockIndex(), toolCall, partial: output });
-                        }
-                    }
-                }
-                if (candidate?.finishReason) {
-                    output.stopReason = mapStopReason(candidate.finishReason);
-                    if (output.content.some((b) => b.type === "toolCall")) {
-                        output.stopReason = "toolUse";
-                    }
-                }
-                if (chunk.usageMetadata) {
-                    output.usage = {
-                        input: (chunk.usageMetadata.promptTokenCount || 0) - (chunk.usageMetadata.cachedContentTokenCount || 0),
-                        output: (chunk.usageMetadata.candidatesTokenCount || 0) + (chunk.usageMetadata.thoughtsTokenCount || 0),
-                        cacheRead: chunk.usageMetadata.cachedContentTokenCount || 0,
-                        cacheWrite: 0,
-                        totalTokens: chunk.usageMetadata.totalTokenCount || 0,
-                        cost: {
-                            input: 0,
-                            output: 0,
-                            cacheRead: 0,
-                            cacheWrite: 0,
-                            total: 0,
-                        },
-                    };
-                    calculateCost(model, output.usage);
-                }
-            }
-            if (currentBlock) {
-                if (currentBlock.type === "text") {
-                    stream.push({
-                        type: "text_end",
-                        contentIndex: blockIndex(),
-                        content: currentBlock.text,
-                        partial: output,
-                    });
-                }
-                else {
-                    stream.push({
-                        type: "thinking_end",
-                        contentIndex: blockIndex(),
-                        content: currentBlock.thinking,
-                        partial: output,
-                    });
-                }
-            }
-            if (options?.signal?.aborted) {
-                throw new Error("Request was aborted");
-            }
-            if (output.stopReason === "aborted" || output.stopReason === "error") {
-                throw new Error("An unknown error occurred");
-            }
-            stream.push({ type: "done", reason: output.stopReason, message: output });
-            stream.end();
-        }
-        catch (error) {
-            // Remove internal index property used during streaming
-            for (const block of output.content) {
-                if ("index" in block) {
-                    delete block.index;
-                }
-            }
-            output.stopReason = options?.signal?.aborted ? "aborted" : "error";
-            output.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);
-            stream.push({ type: "error", reason: output.stopReason, error: output });
-            stream.end();
-        }
-    })();
+    const state = {
+        output: createGoogleOutput(model),
+        stream,
+        currentBlock: null,
+    };
+    void runGoogleStream(model, context, options, state);
     return stream;
 };
 export const streamSimpleGoogle = (model, context, options) => {
@@ -248,14 +238,31 @@ function createClient(model, apiKey, optionsHeaders) {
         httpOptions.baseUrl = model.baseUrl;
         httpOptions.apiVersion = ""; // baseUrl already includes version path, don't append
     }
-    if (model.headers || optionsHeaders) {
-        httpOptions.headers = { ...model.headers, ...optionsHeaders };
+    const headers = providerHeadersToRecord({ ...model.headers, ...optionsHeaders });
+    if (headers) {
+        httpOptions.headers = headers;
     }
     return new GoogleGenAI({
         apiKey,
         httpOptions: Object.keys(httpOptions).length > 0 ? httpOptions : undefined,
     });
 }
+function resolveGoogleThinkingConfig(model, thinking) {
+    if (thinking?.enabled && model.reasoning) {
+        const config = { includeThoughts: true };
+        if (thinking.level !== undefined) {
+            // Cast to any since our GoogleThinkingLevel mirrors Google's ThinkingLevel enum values
+            config.thinkingLevel = thinking.level;
+        }
+        else if (thinking.budgetTokens !== undefined) {
+            config.thinkingBudget = thinking.budgetTokens;
+        }
+        return config;
+    }
+    if (model.reasoning && thinking && !thinking.enabled)
+        return getDisabledThinkingConfig(model);
+    return undefined;
+}
 function buildParams(model, context, options = {}) {
     const contents = convertMessages(model, context);
     const generationConfig = {};
@@ -280,20 +287,9 @@ function buildParams(model, context, options = {}) {
     else {
         config.toolConfig = undefined;
     }
-    if (options.thinking?.enabled && model.reasoning) {
-        const thinkingConfig = { includeThoughts: true };
-        if (options.thinking.level !== undefined) {
-            // Cast to any since our GoogleThinkingLevel mirrors Google's ThinkingLevel enum values
-            thinkingConfig.thinkingLevel = options.thinking.level;
-        }
-        else if (options.thinking.budgetTokens !== undefined) {
-            thinkingConfig.thinkingBudget = options.thinking.budgetTokens;
-        }
+    const thinkingConfig = resolveGoogleThinkingConfig(model, options.thinking);
+    if (thinkingConfig)
         config.thinkingConfig = thinkingConfig;
-    }
-    else if (model.reasoning && options.thinking && !options.thinking.enabled) {
-        config.thinkingConfig = getDisabledThinkingConfig(model);
-    }
     if (options.signal) {
         if (options.signal.aborted) {
             throw new Error("Request aborted");