@avadisabelle/ava-pi-ai 0.64.9 → 0.65.2
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.
- package/README.md +86 -0
- package/dist/cli.d.ts.map +1 -1
- package/dist/cli.js +1 -1
- package/dist/cli.js.map +1 -1
- package/dist/index.d.ts +3 -1
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +1 -0
- package/dist/index.js.map +1 -1
- package/dist/models.d.ts +1 -1
- package/dist/models.d.ts.map +1 -1
- package/dist/models.generated.d.ts +860 -422
- package/dist/models.generated.d.ts.map +1 -1
- package/dist/models.generated.js +966 -513
- package/dist/models.generated.js.map +1 -1
- package/dist/models.js +5 -2
- package/dist/models.js.map +1 -1
- package/dist/providers/amazon-bedrock.d.ts +23 -0
- package/dist/providers/amazon-bedrock.d.ts.map +1 -1
- package/dist/providers/amazon-bedrock.js +138 -33
- package/dist/providers/amazon-bedrock.js.map +1 -1
- package/dist/providers/anthropic.d.ts +16 -2
- package/dist/providers/anthropic.d.ts.map +1 -1
- package/dist/providers/anthropic.js +63 -26
- package/dist/providers/anthropic.js.map +1 -1
- package/dist/providers/azure-openai-responses.d.ts.map +1 -1
- package/dist/providers/azure-openai-responses.js +14 -15
- package/dist/providers/azure-openai-responses.js.map +1 -1
- package/dist/providers/faux.d.ts +56 -0
- package/dist/providers/faux.d.ts.map +1 -0
- package/dist/providers/faux.js +368 -0
- package/dist/providers/faux.js.map +1 -0
- package/dist/providers/google-gemini-cli.d.ts.map +1 -1
- package/dist/providers/google-gemini-cli.js +20 -1
- package/dist/providers/google-gemini-cli.js.map +1 -1
- package/dist/providers/google-shared.d.ts.map +1 -1
- package/dist/providers/google-shared.js +28 -1
- package/dist/providers/google-shared.js.map +1 -1
- package/dist/providers/google-vertex.d.ts.map +1 -1
- package/dist/providers/google-vertex.js +20 -2
- package/dist/providers/google-vertex.js.map +1 -1
- package/dist/providers/google.d.ts.map +1 -1
- package/dist/providers/google.js +45 -4
- package/dist/providers/google.js.map +1 -1
- package/dist/providers/mistral.d.ts +3 -0
- package/dist/providers/mistral.d.ts.map +1 -1
- package/dist/providers/mistral.js +37 -4
- package/dist/providers/mistral.js.map +1 -1
- package/dist/providers/openai-codex-responses.d.ts +2 -0
- package/dist/providers/openai-codex-responses.d.ts.map +1 -1
- package/dist/providers/openai-codex-responses.js +49 -12
- package/dist/providers/openai-codex-responses.js.map +1 -1
- package/dist/providers/openai-completions.d.ts.map +1 -1
- package/dist/providers/openai-completions.js +48 -14
- package/dist/providers/openai-completions.js.map +1 -1
- package/dist/providers/openai-responses-shared.d.ts +1 -0
- package/dist/providers/openai-responses-shared.d.ts.map +1 -1
- package/dist/providers/openai-responses-shared.js +39 -10
- package/dist/providers/openai-responses-shared.js.map +1 -1
- package/dist/providers/openai-responses.d.ts.map +1 -1
- package/dist/providers/openai-responses.js +23 -18
- package/dist/providers/openai-responses.js.map +1 -1
- package/dist/providers/simple-options.d.ts.map +1 -1
- package/dist/providers/simple-options.js +4 -1
- package/dist/providers/simple-options.js.map +1 -1
- package/dist/providers/transform-messages.d.ts.map +1 -1
- package/dist/providers/transform-messages.js +20 -32
- package/dist/providers/transform-messages.js.map +1 -1
- package/dist/types.d.ts +82 -4
- package/dist/types.d.ts.map +1 -1
- package/dist/types.js.map +1 -1
- package/dist/utils/headers.d.ts +2 -0
- package/dist/utils/headers.d.ts.map +1 -0
- package/dist/utils/headers.js +8 -0
- package/dist/utils/headers.js.map +1 -0
- package/dist/utils/overflow.d.ts +4 -3
- package/dist/utils/overflow.d.ts.map +1 -1
- package/dist/utils/overflow.js +28 -13
- package/dist/utils/overflow.js.map +1 -1
- package/package.json +4 -4
|
@@ -1 +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,wBAAwB,CAAC;AAWlE,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,CAmO9E,CAAC;AAEF,eAAO,MAAM,kBAAkB,EAAE,cAAc,CAAC,sBAAsB,EAAE,mBAAmB,CAmC1F,CAAC","sourcesContent":["import {\n\ttype GenerateContentConfig,\n\ttype GenerateContentParameters,\n\tGoogleGenAI,\n\ttype ThinkingConfig,\n} from \"@google/genai\";\nimport { getEnvApiKey } from \"../env-api-keys.js\";\nimport { calculateCost } from \"../models.js\";\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.js\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.js\";\nimport { sanitizeSurrogates } from \"../utils/sanitize-unicode.js\";\nimport type { GoogleThinkingLevel } from \"./google-gemini-cli.js\";\nimport {\n\tconvertMessages,\n\tconvertTools,\n\tisThinkingPart,\n\tmapStopReason,\n\tmapToolChoice,\n\tretainThoughtSignature,\n} from \"./google-shared.js\";\nimport { buildBaseOptions, clampReasoning } from \"./simple-options.js\";\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: chunk.usageMetadata.promptTokenCount || 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, options, apiKey);\n\tif (!options?.reasoning) {\n\t\treturn streamGoogle(model, context, { ...base, thinking: { enabled: false } } satisfies GoogleOptions);\n\t}\n\n\tconst effort = clampReasoning(options.reasoning)!;\n\tconst googleModel = model as Model<\"google-generative-ai\">;\n\n\tif (isGemini3ProModel(googleModel) || isGemini3FlashModel(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: getGemini3ThinkingLevel(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}\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 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 getGemini3ThinkingLevel(\n\teffort: ClampedThinkingLevel,\n\tmodel: Model<\"google-generative-ai\">,\n): 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\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\")) {\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"]}
|
|
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,wBAAwB,CAAC;AAWlE,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,CAmC1F,CAAC","sourcesContent":["import {\n\ttype GenerateContentConfig,\n\ttype GenerateContentParameters,\n\tGoogleGenAI,\n\ttype ThinkingConfig,\n} from \"@google/genai\";\nimport { getEnvApiKey } from \"../env-api-keys.js\";\nimport { calculateCost } from \"../models.js\";\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.js\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.js\";\nimport { sanitizeSurrogates } from \"../utils/sanitize-unicode.js\";\nimport type { GoogleThinkingLevel } from \"./google-gemini-cli.js\";\nimport {\n\tconvertMessages,\n\tconvertTools,\n\tisThinkingPart,\n\tmapStopReason,\n\tmapToolChoice,\n\tretainThoughtSignature,\n} from \"./google-shared.js\";\nimport { buildBaseOptions, clampReasoning } from \"./simple-options.js\";\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, options, apiKey);\n\tif (!options?.reasoning) {\n\t\treturn streamGoogle(model, context, { ...base, thinking: { enabled: false } } satisfies GoogleOptions);\n\t}\n\n\tconst effort = clampReasoning(options.reasoning)!;\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"]}
|
package/dist/providers/google.js
CHANGED
|
@@ -155,7 +155,7 @@ export const streamGoogle = (model, context, options) => {
|
|
|
155
155
|
}
|
|
156
156
|
if (chunk.usageMetadata) {
|
|
157
157
|
output.usage = {
|
|
158
|
-
input: chunk.usageMetadata.promptTokenCount || 0,
|
|
158
|
+
input: (chunk.usageMetadata.promptTokenCount || 0) - (chunk.usageMetadata.cachedContentTokenCount || 0),
|
|
159
159
|
output: (chunk.usageMetadata.candidatesTokenCount || 0) + (chunk.usageMetadata.thoughtsTokenCount || 0),
|
|
160
160
|
cacheRead: chunk.usageMetadata.cachedContentTokenCount || 0,
|
|
161
161
|
cacheWrite: 0,
|
|
@@ -224,12 +224,12 @@ export const streamSimpleGoogle = (model, context, options) => {
|
|
|
224
224
|
}
|
|
225
225
|
const effort = clampReasoning(options.reasoning);
|
|
226
226
|
const googleModel = model;
|
|
227
|
-
if (isGemini3ProModel(googleModel) || isGemini3FlashModel(googleModel)) {
|
|
227
|
+
if (isGemini3ProModel(googleModel) || isGemini3FlashModel(googleModel) || isGemma4Model(googleModel)) {
|
|
228
228
|
return streamGoogle(model, context, {
|
|
229
229
|
...base,
|
|
230
230
|
thinking: {
|
|
231
231
|
enabled: true,
|
|
232
|
-
level:
|
|
232
|
+
level: getThinkingLevel(effort, googleModel),
|
|
233
233
|
},
|
|
234
234
|
});
|
|
235
235
|
}
|
|
@@ -290,6 +290,9 @@ function buildParams(model, context, options = {}) {
|
|
|
290
290
|
}
|
|
291
291
|
config.thinkingConfig = thinkingConfig;
|
|
292
292
|
}
|
|
293
|
+
else if (model.reasoning && options.thinking && !options.thinking.enabled) {
|
|
294
|
+
config.thinkingConfig = getDisabledThinkingConfig(model);
|
|
295
|
+
}
|
|
293
296
|
if (options.signal) {
|
|
294
297
|
if (options.signal.aborted) {
|
|
295
298
|
throw new Error("Request aborted");
|
|
@@ -303,13 +306,32 @@ function buildParams(model, context, options = {}) {
|
|
|
303
306
|
};
|
|
304
307
|
return params;
|
|
305
308
|
}
|
|
309
|
+
function isGemma4Model(model) {
|
|
310
|
+
return /gemma-?4/.test(model.id.toLowerCase());
|
|
311
|
+
}
|
|
306
312
|
function isGemini3ProModel(model) {
|
|
307
313
|
return /gemini-3(?:\.\d+)?-pro/.test(model.id.toLowerCase());
|
|
308
314
|
}
|
|
309
315
|
function isGemini3FlashModel(model) {
|
|
310
316
|
return /gemini-3(?:\.\d+)?-flash/.test(model.id.toLowerCase());
|
|
311
317
|
}
|
|
312
|
-
function
|
|
318
|
+
function getDisabledThinkingConfig(model) {
|
|
319
|
+
// Google docs: Gemini 3.1 Pro cannot disable thinking, and Gemini 3 Flash / Flash-Lite
|
|
320
|
+
// do not support full thinking-off either. For Gemini 3 models, use the lowest supported
|
|
321
|
+
// thinkingLevel without includeThoughts so hidden thinking remains invisible to pi.
|
|
322
|
+
if (isGemini3ProModel(model)) {
|
|
323
|
+
return { thinkingLevel: "LOW" };
|
|
324
|
+
}
|
|
325
|
+
if (isGemini3FlashModel(model)) {
|
|
326
|
+
return { thinkingLevel: "MINIMAL" };
|
|
327
|
+
}
|
|
328
|
+
if (isGemma4Model(model)) {
|
|
329
|
+
return { thinkingLevel: "MINIMAL" };
|
|
330
|
+
}
|
|
331
|
+
// Gemini 2.x supports disabling via thinkingBudget = 0.
|
|
332
|
+
return { thinkingBudget: 0 };
|
|
333
|
+
}
|
|
334
|
+
function getThinkingLevel(effort, model) {
|
|
313
335
|
if (isGemini3ProModel(model)) {
|
|
314
336
|
switch (effort) {
|
|
315
337
|
case "minimal":
|
|
@@ -320,6 +342,16 @@ function getGemini3ThinkingLevel(effort, model) {
|
|
|
320
342
|
return "HIGH";
|
|
321
343
|
}
|
|
322
344
|
}
|
|
345
|
+
if (isGemma4Model(model)) {
|
|
346
|
+
switch (effort) {
|
|
347
|
+
case "minimal":
|
|
348
|
+
case "low":
|
|
349
|
+
return "MINIMAL";
|
|
350
|
+
case "medium":
|
|
351
|
+
case "high":
|
|
352
|
+
return "HIGH";
|
|
353
|
+
}
|
|
354
|
+
}
|
|
323
355
|
switch (effort) {
|
|
324
356
|
case "minimal":
|
|
325
357
|
return "MINIMAL";
|
|
@@ -344,6 +376,15 @@ function getGoogleBudget(model, effort, customBudgets) {
|
|
|
344
376
|
};
|
|
345
377
|
return budgets[effort];
|
|
346
378
|
}
|
|
379
|
+
if (model.id.includes("2.5-flash-lite")) {
|
|
380
|
+
const budgets = {
|
|
381
|
+
minimal: 512,
|
|
382
|
+
low: 2048,
|
|
383
|
+
medium: 8192,
|
|
384
|
+
high: 24576,
|
|
385
|
+
};
|
|
386
|
+
return budgets[effort];
|
|
387
|
+
}
|
|
347
388
|
if (model.id.includes("2.5-flash")) {
|
|
348
389
|
const budgets = {
|
|
349
390
|
minimal: 128,
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"google.js","sourceRoot":"","sources":["../../src/providers/google.ts"],"names":[],"mappings":"AAAA,OAAO,EAGN,WAAW,GAEX,MAAM,eAAe,CAAC;AACvB,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,EAAE,aAAa,EAAE,MAAM,cAAc,CAAC;AAe7C,OAAO,EAAE,2BAA2B,EAAE,MAAM,0BAA0B,CAAC;AACvE,OAAO,EAAE,kBAAkB,EAAE,MAAM,8BAA8B,CAAC;AAElE,OAAO,EACN,eAAe,EACf,YAAY,EACZ,cAAc,EACd,aAAa,EACb,aAAa,EACb,sBAAsB,GACtB,MAAM,oBAAoB,CAAC;AAC5B,OAAO,EAAE,gBAAgB,EAAE,cAAc,EAAE,MAAM,qBAAqB,CAAC;AAWvE,8CAA8C;AAC9C,IAAI,eAAe,GAAG,CAAC,CAAC;AAExB,MAAM,CAAC,MAAM,YAAY,GAA0D,CAClF,KAAoC,EACpC,OAAgB,EAChB,OAAuB,EACO,EAAE,CAAC;IACjC,MAAM,MAAM,GAAG,IAAI,2BAA2B,EAAE,CAAC;IAEjD,CAAC,KAAK,IAAI,EAAE,CAAC;QACZ,MAAM,MAAM,GAAqB;YAChC,IAAI,EAAE,WAAW;YACjB,OAAO,EAAE,EAAE;YACX,GAAG,EAAE,sBAA6B;YAClC,QAAQ,EAAE,KAAK,CAAC,QAAQ;YACxB,KAAK,EAAE,KAAK,CAAC,EAAE;YACf,KAAK,EAAE;gBACN,KAAK,EAAE,CAAC;gBACR,MAAM,EAAE,CAAC;gBACT,SAAS,EAAE,CAAC;gBACZ,UAAU,EAAE,CAAC;gBACb,WAAW,EAAE,CAAC;gBACd,IAAI,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE;aACpE;YACD,UAAU,EAAE,MAAM;YAClB,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;SACrB,CAAC;QAEF,IAAI,CAAC;YACJ,MAAM,MAAM,GAAG,OAAO,EAAE,MAAM,IAAI,YAAY,CAAC,KAAK,CAAC,QAAQ,CAAC,IAAI,EAAE,CAAC;YACrE,MAAM,MAAM,GAAG,YAAY,CAAC,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;YAC7D,IAAI,MAAM,GAAG,WAAW,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;YAClD,MAAM,UAAU,GAAG,MAAM,OAAO,EAAE,SAAS,EAAE,CAAC,MAAM,EAAE,KAAK,CAAC,CAAC;YAC7D,IAAI,UAAU,KAAK,SAAS,EAAE,CAAC;gBAC9B,MAAM,GAAG,UAAuC,CAAC;YAClD,CAAC;YACD,MAAM,YAAY,GAAG,MAAM,MAAM,CAAC,MAAM,CAAC,qBAAqB,CAAC,MAAM,CAAC,CAAC;YAEvE,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;YAChD,IAAI,YAAY,GAAyC,IAAI,CAAC;YAC9D,MAAM,MAAM,GAAG,MAAM,CAAC,OAAO,CAAC;YAC9B,MAAM,UAAU,GAAG,GAAG,EAAE,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC;YAC3C,IAAI,KAAK,EAAE,MAAM,KAAK,IAAI,YAAY,EAAE,CAAC;gBACxC,qFAAqF;gBACrF,gFAAgF;gBAChF,MAAM,CAAC,UAAU,KAAK,KAAK,CAAC,UAAU,CAAC;gBACvC,MAAM,SAAS,GAAG,KAAK,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,CAAC;gBACxC,IAAI,SAAS,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC;oBAC/B,KAAK,MAAM,IAAI,IAAI,SAAS,CAAC,OAAO,CAAC,KAAK,EAAE,CAAC;wBAC5C,IAAI,IAAI,CAAC,IAAI,KAAK,SAAS,EAAE,CAAC;4BAC7B,MAAM,UAAU,GAAG,cAAc,CAAC,IAAI,CAAC,CAAC;4BACxC,IACC,CAAC,YAAY;gCACb,CAAC,UAAU,IAAI,YAAY,CAAC,IAAI,KAAK,UAAU,CAAC;gCAChD,CAAC,CAAC,UAAU,IAAI,YAAY,CAAC,IAAI,KAAK,MAAM,CAAC,EAC5C,CAAC;gCACF,IAAI,YAAY,EAAE,CAAC;oCAClB,IAAI,YAAY,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;wCAClC,MAAM,CAAC,IAAI,CAAC;4CACX,IAAI,EAAE,UAAU;4CAChB,YAAY,EAAE,MAAM,CAAC,MAAM,GAAG,CAAC;4CAC/B,OAAO,EAAE,YAAY,CAAC,IAAI;4CAC1B,OAAO,EAAE,MAAM;yCACf,CAAC,CAAC;oCACJ,CAAC;yCAAM,CAAC;wCACP,MAAM,CAAC,IAAI,CAAC;4CACX,IAAI,EAAE,cAAc;4CACpB,YAAY,EAAE,UAAU,EAAE;4CAC1B,OAAO,EAAE,YAAY,CAAC,QAAQ;4CAC9B,OAAO,EAAE,MAAM;yCACf,CAAC,CAAC;oCACJ,CAAC;gCACF,CAAC;gCACD,IAAI,UAAU,EAAE,CAAC;oCAChB,YAAY,GAAG,EAAE,IAAI,EAAE,UAAU,EAAE,QAAQ,EAAE,EAAE,EAAE,iBAAiB,EAAE,SAAS,EAAE,CAAC;oCAChF,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;oCAClC,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,gBAAgB,EAAE,YAAY,EAAE,UAAU,EAAE,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;gCACtF,CAAC;qCAAM,CAAC;oCACP,YAAY,GAAG,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,EAAE,EAAE,CAAC;oCAC1C,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;oCAClC,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,YAAY,EAAE,YAAY,EAAE,UAAU,EAAE,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;gCAClF,CAAC;4BACF,CAAC;4BACD,IAAI,YAAY,CAAC,IAAI,KAAK,UAAU,EAAE,CAAC;gCACtC,YAAY,CAAC,QAAQ,IAAI,IAAI,CAAC,IAAI,CAAC;gCACnC,YAAY,CAAC,iBAAiB,GAAG,sBAAsB,CACtD,YAAY,CAAC,iBAAiB,EAC9B,IAAI,CAAC,gBAAgB,CACrB,CAAC;gCACF,MAAM,CAAC,IAAI,CAAC;oCACX,IAAI,EAAE,gBAAgB;oCACtB,YAAY,EAAE,UAAU,EAAE;oCAC1B,KAAK,EAAE,IAAI,CAAC,IAAI;oCAChB,OAAO,EAAE,MAAM;iCACf,CAAC,CAAC;4BACJ,CAAC;iCAAM,CAAC;gCACP,YAAY,CAAC,IAAI,IAAI,IAAI,CAAC,IAAI,CAAC;gCAC/B,YAAY,CAAC,aAAa,GAAG,sBAAsB,CAClD,YAAY,CAAC,aAAa,EAC1B,IAAI,CAAC,gBAAgB,CACrB,CAAC;gCACF,MAAM,CAAC,IAAI,CAAC;oCACX,IAAI,EAAE,YAAY;oCAClB,YAAY,EAAE,UAAU,EAAE;oCAC1B,KAAK,EAAE,IAAI,CAAC,IAAI;oCAChB,OAAO,EAAE,MAAM;iCACf,CAAC,CAAC;4BACJ,CAAC;wBACF,CAAC;wBAED,IAAI,IAAI,CAAC,YAAY,EAAE,CAAC;4BACvB,IAAI,YAAY,EAAE,CAAC;gCAClB,IAAI,YAAY,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;oCAClC,MAAM,CAAC,IAAI,CAAC;wCACX,IAAI,EAAE,UAAU;wCAChB,YAAY,EAAE,UAAU,EAAE;wCAC1B,OAAO,EAAE,YAAY,CAAC,IAAI;wCAC1B,OAAO,EAAE,MAAM;qCACf,CAAC,CAAC;gCACJ,CAAC;qCAAM,CAAC;oCACP,MAAM,CAAC,IAAI,CAAC;wCACX,IAAI,EAAE,cAAc;wCACpB,YAAY,EAAE,UAAU,EAAE;wCAC1B,OAAO,EAAE,YAAY,CAAC,QAAQ;wCAC9B,OAAO,EAAE,MAAM;qCACf,CAAC,CAAC;gCACJ,CAAC;gCACD,YAAY,GAAG,IAAI,CAAC;4BACrB,CAAC;4BAED,4DAA4D;4BAC5D,MAAM,UAAU,GAAG,IAAI,CAAC,YAAY,CAAC,EAAE,CAAC;4BACxC,MAAM,UAAU,GACf,CAAC,UAAU,IAAI,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,UAAU,IAAI,CAAC,CAAC,EAAE,KAAK,UAAU,CAAC,CAAC;4BACzF,MAAM,UAAU,GAAG,UAAU;gCAC5B,CAAC,CAAC,GAAG,IAAI,CAAC,YAAY,CAAC,IAAI,IAAI,IAAI,CAAC,GAAG,EAAE,IAAI,EAAE,eAAe,EAAE;gCAChE,CAAC,CAAC,UAAU,CAAC;4BAEd,MAAM,QAAQ,GAAa;gCAC1B,IAAI,EAAE,UAAU;gCAChB,EAAE,EAAE,UAAU;gCACd,IAAI,EAAE,IAAI,CAAC,YAAY,CAAC,IAAI,IAAI,EAAE;gCAClC,SAAS,EAAG,IAAI,CAAC,YAAY,CAAC,IAA4B,IAAI,EAAE;gCAChE,GAAG,CAAC,IAAI,CAAC,gBAAgB,IAAI,EAAE,gBAAgB,EAAE,IAAI,CAAC,gBAAgB,EAAE,CAAC;6BACzE,CAAC;4BAEF,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;4BAC9B,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,gBAAgB,EAAE,YAAY,EAAE,UAAU,EAAE,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;4BACrF,MAAM,CAAC,IAAI,CAAC;gCACX,IAAI,EAAE,gBAAgB;gCACtB,YAAY,EAAE,UAAU,EAAE;gCAC1B,KAAK,EAAE,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,SAAS,CAAC;gCACzC,OAAO,EAAE,MAAM;6BACf,CAAC,CAAC;4BACH,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,cAAc,EAAE,YAAY,EAAE,UAAU,EAAE,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;wBAC9F,CAAC;oBACF,CAAC;gBACF,CAAC;gBAED,IAAI,SAAS,EAAE,YAAY,EAAE,CAAC;oBAC7B,MAAM,CAAC,UAAU,GAAG,aAAa,CAAC,SAAS,CAAC,YAAY,CAAC,CAAC;oBAC1D,IAAI,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,UAAU,CAAC,EAAE,CAAC;wBACvD,MAAM,CAAC,UAAU,GAAG,SAAS,CAAC;oBAC/B,CAAC;gBACF,CAAC;gBAED,IAAI,KAAK,CAAC,aAAa,EAAE,CAAC;oBACzB,MAAM,CAAC,KAAK,GAAG;wBACd,KAAK,EAAE,KAAK,CAAC,aAAa,CAAC,gBAAgB,IAAI,CAAC;wBAChD,MAAM,EACL,CAAC,KAAK,CAAC,aAAa,CAAC,oBAAoB,IAAI,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,aAAa,CAAC,kBAAkB,IAAI,CAAC,CAAC;wBAChG,SAAS,EAAE,KAAK,CAAC,aAAa,CAAC,uBAAuB,IAAI,CAAC;wBAC3D,UAAU,EAAE,CAAC;wBACb,WAAW,EAAE,KAAK,CAAC,aAAa,CAAC,eAAe,IAAI,CAAC;wBACrD,IAAI,EAAE;4BACL,KAAK,EAAE,CAAC;4BACR,MAAM,EAAE,CAAC;4BACT,SAAS,EAAE,CAAC;4BACZ,UAAU,EAAE,CAAC;4BACb,KAAK,EAAE,CAAC;yBACR;qBACD,CAAC;oBACF,aAAa,CAAC,KAAK,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC;gBACpC,CAAC;YACF,CAAC;YAED,IAAI,YAAY,EAAE,CAAC;gBAClB,IAAI,YAAY,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;oBAClC,MAAM,CAAC,IAAI,CAAC;wBACX,IAAI,EAAE,UAAU;wBAChB,YAAY,EAAE,UAAU,EAAE;wBAC1B,OAAO,EAAE,YAAY,CAAC,IAAI;wBAC1B,OAAO,EAAE,MAAM;qBACf,CAAC,CAAC;gBACJ,CAAC;qBAAM,CAAC;oBACP,MAAM,CAAC,IAAI,CAAC;wBACX,IAAI,EAAE,cAAc;wBACpB,YAAY,EAAE,UAAU,EAAE;wBAC1B,OAAO,EAAE,YAAY,CAAC,QAAQ;wBAC9B,OAAO,EAAE,MAAM;qBACf,CAAC,CAAC;gBACJ,CAAC;YACF,CAAC;YAED,IAAI,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,CAAC;gBAC9B,MAAM,IAAI,KAAK,CAAC,qBAAqB,CAAC,CAAC;YACxC,CAAC;YAED,IAAI,MAAM,CAAC,UAAU,KAAK,SAAS,IAAI,MAAM,CAAC,UAAU,KAAK,OAAO,EAAE,CAAC;gBACtE,MAAM,IAAI,KAAK,CAAC,2BAA2B,CAAC,CAAC;YAC9C,CAAC;YAED,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,UAAU,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;YAC1E,MAAM,CAAC,GAAG,EAAE,CAAC;QACd,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,uDAAuD;YACvD,KAAK,MAAM,KAAK,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;gBACpC,IAAI,OAAO,IAAI,KAAK,EAAE,CAAC;oBACtB,OAAQ,KAA4B,CAAC,KAAK,CAAC;gBAC5C,CAAC;YACF,CAAC;YACD,MAAM,CAAC,UAAU,GAAG,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC;YACnE,MAAM,CAAC,YAAY,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;YACrF,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,CAAC,UAAU,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;YACzE,MAAM,CAAC,GAAG,EAAE,CAAC;QACd,CAAC;IAAA,CACD,CAAC,EAAE,CAAC;IAEL,OAAO,MAAM,CAAC;AAAA,CACd,CAAC;AAEF,MAAM,CAAC,MAAM,kBAAkB,GAAgE,CAC9F,KAAoC,EACpC,OAAgB,EAChB,OAA6B,EACC,EAAE,CAAC;IACjC,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,IAAI,GAAG,gBAAgB,CAAC,KAAK,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC;IACtD,IAAI,CAAC,OAAO,EAAE,SAAS,EAAE,CAAC;QACzB,OAAO,YAAY,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE,GAAG,IAAI,EAAE,QAAQ,EAAE,EAAE,OAAO,EAAE,KAAK,EAAE,EAA0B,CAAC,CAAC;IACxG,CAAC;IAED,MAAM,MAAM,GAAG,cAAc,CAAC,OAAO,CAAC,SAAS,CAAE,CAAC;IAClD,MAAM,WAAW,GAAG,KAAsC,CAAC;IAE3D,IAAI,iBAAiB,CAAC,WAAW,CAAC,IAAI,mBAAmB,CAAC,WAAW,CAAC,EAAE,CAAC;QACxE,OAAO,YAAY,CAAC,KAAK,EAAE,OAAO,EAAE;YACnC,GAAG,IAAI;YACP,QAAQ,EAAE;gBACT,OAAO,EAAE,IAAI;gBACb,KAAK,EAAE,uBAAuB,CAAC,MAAM,EAAE,WAAW,CAAC;aACnD;SACuB,CAAC,CAAC;IAC5B,CAAC;IAED,OAAO,YAAY,CAAC,KAAK,EAAE,OAAO,EAAE;QACnC,GAAG,IAAI;QACP,QAAQ,EAAE;YACT,OAAO,EAAE,IAAI;YACb,YAAY,EAAE,eAAe,CAAC,WAAW,EAAE,MAAM,EAAE,OAAO,CAAC,eAAe,CAAC;SAC3E;KACuB,CAAC,CAAC;AAAA,CAC3B,CAAC;AAEF,SAAS,YAAY,CACpB,KAAoC,EACpC,MAAe,EACf,cAAuC,EACzB;IACd,MAAM,WAAW,GAAgF,EAAE,CAAC;IACpG,IAAI,KAAK,CAAC,OAAO,EAAE,CAAC;QACnB,WAAW,CAAC,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC;QACpC,WAAW,CAAC,UAAU,GAAG,EAAE,CAAC,CAAC,sDAAsD;IACpF,CAAC;IACD,IAAI,KAAK,CAAC,OAAO,IAAI,cAAc,EAAE,CAAC;QACrC,WAAW,CAAC,OAAO,GAAG,EAAE,GAAG,KAAK,CAAC,OAAO,EAAE,GAAG,cAAc,EAAE,CAAC;IAC/D,CAAC;IAED,OAAO,IAAI,WAAW,CAAC;QACtB,MAAM;QACN,WAAW,EAAE,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,SAAS;KAC1E,CAAC,CAAC;AAAA,CACH;AAED,SAAS,WAAW,CACnB,KAAoC,EACpC,OAAgB,EAChB,OAAO,GAAkB,EAAE,EACC;IAC5B,MAAM,QAAQ,GAAG,eAAe,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC;IAEjD,MAAM,gBAAgB,GAA0B,EAAE,CAAC;IACnD,IAAI,OAAO,CAAC,WAAW,KAAK,SAAS,EAAE,CAAC;QACvC,gBAAgB,CAAC,WAAW,GAAG,OAAO,CAAC,WAAW,CAAC;IACpD,CAAC;IACD,IAAI,OAAO,CAAC,SAAS,KAAK,SAAS,EAAE,CAAC;QACrC,gBAAgB,CAAC,eAAe,GAAG,OAAO,CAAC,SAAS,CAAC;IACtD,CAAC;IAED,MAAM,MAAM,GAA0B;QACrC,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAC,MAAM,GAAG,CAAC,IAAI,gBAAgB,CAAC;QACjE,GAAG,CAAC,OAAO,CAAC,YAAY,IAAI,EAAE,iBAAiB,EAAE,kBAAkB,CAAC,OAAO,CAAC,YAAY,CAAC,EAAE,CAAC;QAC5F,GAAG,CAAC,OAAO,CAAC,KAAK,IAAI,OAAO,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,KAAK,EAAE,YAAY,CAAC,OAAO,CAAC,KAAK,CAAC,EAAE,CAAC;KACxF,CAAC;IAEF,IAAI,OAAO,CAAC,KAAK,IAAI,OAAO,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,IAAI,OAAO,CAAC,UAAU,EAAE,CAAC;QACrE,MAAM,CAAC,UAAU,GAAG;YACnB,qBAAqB,EAAE;gBACtB,IAAI,EAAE,aAAa,CAAC,OAAO,CAAC,UAAU,CAAC;aACvC;SACD,CAAC;IACH,CAAC;SAAM,CAAC;QACP,MAAM,CAAC,UAAU,GAAG,SAAS,CAAC;IAC/B,CAAC;IAED,IAAI,OAAO,CAAC,QAAQ,EAAE,OAAO,IAAI,KAAK,CAAC,SAAS,EAAE,CAAC;QAClD,MAAM,cAAc,GAAmB,EAAE,eAAe,EAAE,IAAI,EAAE,CAAC;QACjE,IAAI,OAAO,CAAC,QAAQ,CAAC,KAAK,KAAK,SAAS,EAAE,CAAC;YAC1C,uFAAuF;YACvF,cAAc,CAAC,aAAa,GAAG,OAAO,CAAC,QAAQ,CAAC,KAAY,CAAC;QAC9D,CAAC;aAAM,IAAI,OAAO,CAAC,QAAQ,CAAC,YAAY,KAAK,SAAS,EAAE,CAAC;YACxD,cAAc,CAAC,cAAc,GAAG,OAAO,CAAC,QAAQ,CAAC,YAAY,CAAC;QAC/D,CAAC;QACD,MAAM,CAAC,cAAc,GAAG,cAAc,CAAC;IACxC,CAAC;IAED,IAAI,OAAO,CAAC,MAAM,EAAE,CAAC;QACpB,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;YAC5B,MAAM,IAAI,KAAK,CAAC,iBAAiB,CAAC,CAAC;QACpC,CAAC;QACD,MAAM,CAAC,WAAW,GAAG,OAAO,CAAC,MAAM,CAAC;IACrC,CAAC;IAED,MAAM,MAAM,GAA8B;QACzC,KAAK,EAAE,KAAK,CAAC,EAAE;QACf,QAAQ;QACR,MAAM;KACN,CAAC;IAEF,OAAO,MAAM,CAAC;AAAA,CACd;AAID,SAAS,iBAAiB,CAAC,KAAoC,EAAW;IACzE,OAAO,wBAAwB,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,WAAW,EAAE,CAAC,CAAC;AAAA,CAC7D;AAED,SAAS,mBAAmB,CAAC,KAAoC,EAAW;IAC3E,OAAO,0BAA0B,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,WAAW,EAAE,CAAC,CAAC;AAAA,CAC/D;AAED,SAAS,uBAAuB,CAC/B,MAA4B,EAC5B,KAAoC,EACd;IACtB,IAAI,iBAAiB,CAAC,KAAK,CAAC,EAAE,CAAC;QAC9B,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,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,CACvB,KAAoC,EACpC,MAA4B,EAC5B,aAA+B,EACtB;IACT,IAAI,aAAa,EAAE,CAAC,MAAM,CAAC,KAAK,SAAS,EAAE,CAAC;QAC3C,OAAO,aAAa,CAAC,MAAM,CAAE,CAAC;IAC/B,CAAC;IAED,IAAI,KAAK,CAAC,EAAE,CAAC,QAAQ,CAAC,SAAS,CAAC,EAAE,CAAC;QAClC,MAAM,OAAO,GAAyC;YACrD,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,MAAM,OAAO,GAAyC;YACrD,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,OAAO,CAAC,CAAC,CAAC;AAAA,CACV","sourcesContent":["import {\n\ttype GenerateContentConfig,\n\ttype GenerateContentParameters,\n\tGoogleGenAI,\n\ttype ThinkingConfig,\n} from \"@google/genai\";\nimport { getEnvApiKey } from \"../env-api-keys.js\";\nimport { calculateCost } from \"../models.js\";\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.js\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.js\";\nimport { sanitizeSurrogates } from \"../utils/sanitize-unicode.js\";\nimport type { GoogleThinkingLevel } from \"./google-gemini-cli.js\";\nimport {\n\tconvertMessages,\n\tconvertTools,\n\tisThinkingPart,\n\tmapStopReason,\n\tmapToolChoice,\n\tretainThoughtSignature,\n} from \"./google-shared.js\";\nimport { buildBaseOptions, clampReasoning } from \"./simple-options.js\";\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: chunk.usageMetadata.promptTokenCount || 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, options, apiKey);\n\tif (!options?.reasoning) {\n\t\treturn streamGoogle(model, context, { ...base, thinking: { enabled: false } } satisfies GoogleOptions);\n\t}\n\n\tconst effort = clampReasoning(options.reasoning)!;\n\tconst googleModel = model as Model<\"google-generative-ai\">;\n\n\tif (isGemini3ProModel(googleModel) || isGemini3FlashModel(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: getGemini3ThinkingLevel(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}\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 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 getGemini3ThinkingLevel(\n\teffort: ClampedThinkingLevel,\n\tmodel: Model<\"google-generative-ai\">,\n): 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\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\")) {\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"]}
|
|
1
|
+
{"version":3,"file":"google.js","sourceRoot":"","sources":["../../src/providers/google.ts"],"names":[],"mappings":"AAAA,OAAO,EAGN,WAAW,GAEX,MAAM,eAAe,CAAC;AACvB,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,EAAE,aAAa,EAAE,MAAM,cAAc,CAAC;AAe7C,OAAO,EAAE,2BAA2B,EAAE,MAAM,0BAA0B,CAAC;AACvE,OAAO,EAAE,kBAAkB,EAAE,MAAM,8BAA8B,CAAC;AAElE,OAAO,EACN,eAAe,EACf,YAAY,EACZ,cAAc,EACd,aAAa,EACb,aAAa,EACb,sBAAsB,GACtB,MAAM,oBAAoB,CAAC;AAC5B,OAAO,EAAE,gBAAgB,EAAE,cAAc,EAAE,MAAM,qBAAqB,CAAC;AAWvE,8CAA8C;AAC9C,IAAI,eAAe,GAAG,CAAC,CAAC;AAExB,MAAM,CAAC,MAAM,YAAY,GAA0D,CAClF,KAAoC,EACpC,OAAgB,EAChB,OAAuB,EACO,EAAE,CAAC;IACjC,MAAM,MAAM,GAAG,IAAI,2BAA2B,EAAE,CAAC;IAEjD,CAAC,KAAK,IAAI,EAAE,CAAC;QACZ,MAAM,MAAM,GAAqB;YAChC,IAAI,EAAE,WAAW;YACjB,OAAO,EAAE,EAAE;YACX,GAAG,EAAE,sBAA6B;YAClC,QAAQ,EAAE,KAAK,CAAC,QAAQ;YACxB,KAAK,EAAE,KAAK,CAAC,EAAE;YACf,KAAK,EAAE;gBACN,KAAK,EAAE,CAAC;gBACR,MAAM,EAAE,CAAC;gBACT,SAAS,EAAE,CAAC;gBACZ,UAAU,EAAE,CAAC;gBACb,WAAW,EAAE,CAAC;gBACd,IAAI,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,UAAU,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE;aACpE;YACD,UAAU,EAAE,MAAM;YAClB,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;SACrB,CAAC;QAEF,IAAI,CAAC;YACJ,MAAM,MAAM,GAAG,OAAO,EAAE,MAAM,IAAI,YAAY,CAAC,KAAK,CAAC,QAAQ,CAAC,IAAI,EAAE,CAAC;YACrE,MAAM,MAAM,GAAG,YAAY,CAAC,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;YAC7D,IAAI,MAAM,GAAG,WAAW,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;YAClD,MAAM,UAAU,GAAG,MAAM,OAAO,EAAE,SAAS,EAAE,CAAC,MAAM,EAAE,KAAK,CAAC,CAAC;YAC7D,IAAI,UAAU,KAAK,SAAS,EAAE,CAAC;gBAC9B,MAAM,GAAG,UAAuC,CAAC;YAClD,CAAC;YACD,MAAM,YAAY,GAAG,MAAM,MAAM,CAAC,MAAM,CAAC,qBAAqB,CAAC,MAAM,CAAC,CAAC;YAEvE,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;YAChD,IAAI,YAAY,GAAyC,IAAI,CAAC;YAC9D,MAAM,MAAM,GAAG,MAAM,CAAC,OAAO,CAAC;YAC9B,MAAM,UAAU,GAAG,GAAG,EAAE,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC;YAC3C,IAAI,KAAK,EAAE,MAAM,KAAK,IAAI,YAAY,EAAE,CAAC;gBACxC,qFAAqF;gBACrF,gFAAgF;gBAChF,MAAM,CAAC,UAAU,KAAK,KAAK,CAAC,UAAU,CAAC;gBACvC,MAAM,SAAS,GAAG,KAAK,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,CAAC;gBACxC,IAAI,SAAS,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC;oBAC/B,KAAK,MAAM,IAAI,IAAI,SAAS,CAAC,OAAO,CAAC,KAAK,EAAE,CAAC;wBAC5C,IAAI,IAAI,CAAC,IAAI,KAAK,SAAS,EAAE,CAAC;4BAC7B,MAAM,UAAU,GAAG,cAAc,CAAC,IAAI,CAAC,CAAC;4BACxC,IACC,CAAC,YAAY;gCACb,CAAC,UAAU,IAAI,YAAY,CAAC,IAAI,KAAK,UAAU,CAAC;gCAChD,CAAC,CAAC,UAAU,IAAI,YAAY,CAAC,IAAI,KAAK,MAAM,CAAC,EAC5C,CAAC;gCACF,IAAI,YAAY,EAAE,CAAC;oCAClB,IAAI,YAAY,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;wCAClC,MAAM,CAAC,IAAI,CAAC;4CACX,IAAI,EAAE,UAAU;4CAChB,YAAY,EAAE,MAAM,CAAC,MAAM,GAAG,CAAC;4CAC/B,OAAO,EAAE,YAAY,CAAC,IAAI;4CAC1B,OAAO,EAAE,MAAM;yCACf,CAAC,CAAC;oCACJ,CAAC;yCAAM,CAAC;wCACP,MAAM,CAAC,IAAI,CAAC;4CACX,IAAI,EAAE,cAAc;4CACpB,YAAY,EAAE,UAAU,EAAE;4CAC1B,OAAO,EAAE,YAAY,CAAC,QAAQ;4CAC9B,OAAO,EAAE,MAAM;yCACf,CAAC,CAAC;oCACJ,CAAC;gCACF,CAAC;gCACD,IAAI,UAAU,EAAE,CAAC;oCAChB,YAAY,GAAG,EAAE,IAAI,EAAE,UAAU,EAAE,QAAQ,EAAE,EAAE,EAAE,iBAAiB,EAAE,SAAS,EAAE,CAAC;oCAChF,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;oCAClC,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,gBAAgB,EAAE,YAAY,EAAE,UAAU,EAAE,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;gCACtF,CAAC;qCAAM,CAAC;oCACP,YAAY,GAAG,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,EAAE,EAAE,CAAC;oCAC1C,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;oCAClC,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,YAAY,EAAE,YAAY,EAAE,UAAU,EAAE,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;gCAClF,CAAC;4BACF,CAAC;4BACD,IAAI,YAAY,CAAC,IAAI,KAAK,UAAU,EAAE,CAAC;gCACtC,YAAY,CAAC,QAAQ,IAAI,IAAI,CAAC,IAAI,CAAC;gCACnC,YAAY,CAAC,iBAAiB,GAAG,sBAAsB,CACtD,YAAY,CAAC,iBAAiB,EAC9B,IAAI,CAAC,gBAAgB,CACrB,CAAC;gCACF,MAAM,CAAC,IAAI,CAAC;oCACX,IAAI,EAAE,gBAAgB;oCACtB,YAAY,EAAE,UAAU,EAAE;oCAC1B,KAAK,EAAE,IAAI,CAAC,IAAI;oCAChB,OAAO,EAAE,MAAM;iCACf,CAAC,CAAC;4BACJ,CAAC;iCAAM,CAAC;gCACP,YAAY,CAAC,IAAI,IAAI,IAAI,CAAC,IAAI,CAAC;gCAC/B,YAAY,CAAC,aAAa,GAAG,sBAAsB,CAClD,YAAY,CAAC,aAAa,EAC1B,IAAI,CAAC,gBAAgB,CACrB,CAAC;gCACF,MAAM,CAAC,IAAI,CAAC;oCACX,IAAI,EAAE,YAAY;oCAClB,YAAY,EAAE,UAAU,EAAE;oCAC1B,KAAK,EAAE,IAAI,CAAC,IAAI;oCAChB,OAAO,EAAE,MAAM;iCACf,CAAC,CAAC;4BACJ,CAAC;wBACF,CAAC;wBAED,IAAI,IAAI,CAAC,YAAY,EAAE,CAAC;4BACvB,IAAI,YAAY,EAAE,CAAC;gCAClB,IAAI,YAAY,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;oCAClC,MAAM,CAAC,IAAI,CAAC;wCACX,IAAI,EAAE,UAAU;wCAChB,YAAY,EAAE,UAAU,EAAE;wCAC1B,OAAO,EAAE,YAAY,CAAC,IAAI;wCAC1B,OAAO,EAAE,MAAM;qCACf,CAAC,CAAC;gCACJ,CAAC;qCAAM,CAAC;oCACP,MAAM,CAAC,IAAI,CAAC;wCACX,IAAI,EAAE,cAAc;wCACpB,YAAY,EAAE,UAAU,EAAE;wCAC1B,OAAO,EAAE,YAAY,CAAC,QAAQ;wCAC9B,OAAO,EAAE,MAAM;qCACf,CAAC,CAAC;gCACJ,CAAC;gCACD,YAAY,GAAG,IAAI,CAAC;4BACrB,CAAC;4BAED,4DAA4D;4BAC5D,MAAM,UAAU,GAAG,IAAI,CAAC,YAAY,CAAC,EAAE,CAAC;4BACxC,MAAM,UAAU,GACf,CAAC,UAAU,IAAI,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,UAAU,IAAI,CAAC,CAAC,EAAE,KAAK,UAAU,CAAC,CAAC;4BACzF,MAAM,UAAU,GAAG,UAAU;gCAC5B,CAAC,CAAC,GAAG,IAAI,CAAC,YAAY,CAAC,IAAI,IAAI,IAAI,CAAC,GAAG,EAAE,IAAI,EAAE,eAAe,EAAE;gCAChE,CAAC,CAAC,UAAU,CAAC;4BAEd,MAAM,QAAQ,GAAa;gCAC1B,IAAI,EAAE,UAAU;gCAChB,EAAE,EAAE,UAAU;gCACd,IAAI,EAAE,IAAI,CAAC,YAAY,CAAC,IAAI,IAAI,EAAE;gCAClC,SAAS,EAAG,IAAI,CAAC,YAAY,CAAC,IAA4B,IAAI,EAAE;gCAChE,GAAG,CAAC,IAAI,CAAC,gBAAgB,IAAI,EAAE,gBAAgB,EAAE,IAAI,CAAC,gBAAgB,EAAE,CAAC;6BACzE,CAAC;4BAEF,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;4BAC9B,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,gBAAgB,EAAE,YAAY,EAAE,UAAU,EAAE,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;4BACrF,MAAM,CAAC,IAAI,CAAC;gCACX,IAAI,EAAE,gBAAgB;gCACtB,YAAY,EAAE,UAAU,EAAE;gCAC1B,KAAK,EAAE,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,SAAS,CAAC;gCACzC,OAAO,EAAE,MAAM;6BACf,CAAC,CAAC;4BACH,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,cAAc,EAAE,YAAY,EAAE,UAAU,EAAE,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;wBAC9F,CAAC;oBACF,CAAC;gBACF,CAAC;gBAED,IAAI,SAAS,EAAE,YAAY,EAAE,CAAC;oBAC7B,MAAM,CAAC,UAAU,GAAG,aAAa,CAAC,SAAS,CAAC,YAAY,CAAC,CAAC;oBAC1D,IAAI,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,UAAU,CAAC,EAAE,CAAC;wBACvD,MAAM,CAAC,UAAU,GAAG,SAAS,CAAC;oBAC/B,CAAC;gBACF,CAAC;gBAED,IAAI,KAAK,CAAC,aAAa,EAAE,CAAC;oBACzB,MAAM,CAAC,KAAK,GAAG;wBACd,KAAK,EACJ,CAAC,KAAK,CAAC,aAAa,CAAC,gBAAgB,IAAI,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,aAAa,CAAC,uBAAuB,IAAI,CAAC,CAAC;wBACjG,MAAM,EACL,CAAC,KAAK,CAAC,aAAa,CAAC,oBAAoB,IAAI,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,aAAa,CAAC,kBAAkB,IAAI,CAAC,CAAC;wBAChG,SAAS,EAAE,KAAK,CAAC,aAAa,CAAC,uBAAuB,IAAI,CAAC;wBAC3D,UAAU,EAAE,CAAC;wBACb,WAAW,EAAE,KAAK,CAAC,aAAa,CAAC,eAAe,IAAI,CAAC;wBACrD,IAAI,EAAE;4BACL,KAAK,EAAE,CAAC;4BACR,MAAM,EAAE,CAAC;4BACT,SAAS,EAAE,CAAC;4BACZ,UAAU,EAAE,CAAC;4BACb,KAAK,EAAE,CAAC;yBACR;qBACD,CAAC;oBACF,aAAa,CAAC,KAAK,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC;gBACpC,CAAC;YACF,CAAC;YAED,IAAI,YAAY,EAAE,CAAC;gBAClB,IAAI,YAAY,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;oBAClC,MAAM,CAAC,IAAI,CAAC;wBACX,IAAI,EAAE,UAAU;wBAChB,YAAY,EAAE,UAAU,EAAE;wBAC1B,OAAO,EAAE,YAAY,CAAC,IAAI;wBAC1B,OAAO,EAAE,MAAM;qBACf,CAAC,CAAC;gBACJ,CAAC;qBAAM,CAAC;oBACP,MAAM,CAAC,IAAI,CAAC;wBACX,IAAI,EAAE,cAAc;wBACpB,YAAY,EAAE,UAAU,EAAE;wBAC1B,OAAO,EAAE,YAAY,CAAC,QAAQ;wBAC9B,OAAO,EAAE,MAAM;qBACf,CAAC,CAAC;gBACJ,CAAC;YACF,CAAC;YAED,IAAI,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,CAAC;gBAC9B,MAAM,IAAI,KAAK,CAAC,qBAAqB,CAAC,CAAC;YACxC,CAAC;YAED,IAAI,MAAM,CAAC,UAAU,KAAK,SAAS,IAAI,MAAM,CAAC,UAAU,KAAK,OAAO,EAAE,CAAC;gBACtE,MAAM,IAAI,KAAK,CAAC,2BAA2B,CAAC,CAAC;YAC9C,CAAC;YAED,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,UAAU,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;YAC1E,MAAM,CAAC,GAAG,EAAE,CAAC;QACd,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,uDAAuD;YACvD,KAAK,MAAM,KAAK,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;gBACpC,IAAI,OAAO,IAAI,KAAK,EAAE,CAAC;oBACtB,OAAQ,KAA4B,CAAC,KAAK,CAAC;gBAC5C,CAAC;YACF,CAAC;YACD,MAAM,CAAC,UAAU,GAAG,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC;YACnE,MAAM,CAAC,YAAY,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;YACrF,MAAM,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,CAAC,UAAU,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;YACzE,MAAM,CAAC,GAAG,EAAE,CAAC;QACd,CAAC;IAAA,CACD,CAAC,EAAE,CAAC;IAEL,OAAO,MAAM,CAAC;AAAA,CACd,CAAC;AAEF,MAAM,CAAC,MAAM,kBAAkB,GAAgE,CAC9F,KAAoC,EACpC,OAAgB,EAChB,OAA6B,EACC,EAAE,CAAC;IACjC,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,IAAI,GAAG,gBAAgB,CAAC,KAAK,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC;IACtD,IAAI,CAAC,OAAO,EAAE,SAAS,EAAE,CAAC;QACzB,OAAO,YAAY,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE,GAAG,IAAI,EAAE,QAAQ,EAAE,EAAE,OAAO,EAAE,KAAK,EAAE,EAA0B,CAAC,CAAC;IACxG,CAAC;IAED,MAAM,MAAM,GAAG,cAAc,CAAC,OAAO,CAAC,SAAS,CAAE,CAAC;IAClD,MAAM,WAAW,GAAG,KAAsC,CAAC;IAE3D,IAAI,iBAAiB,CAAC,WAAW,CAAC,IAAI,mBAAmB,CAAC,WAAW,CAAC,IAAI,aAAa,CAAC,WAAW,CAAC,EAAE,CAAC;QACtG,OAAO,YAAY,CAAC,KAAK,EAAE,OAAO,EAAE;YACnC,GAAG,IAAI;YACP,QAAQ,EAAE;gBACT,OAAO,EAAE,IAAI;gBACb,KAAK,EAAE,gBAAgB,CAAC,MAAM,EAAE,WAAW,CAAC;aAC5C;SACuB,CAAC,CAAC;IAC5B,CAAC;IAED,OAAO,YAAY,CAAC,KAAK,EAAE,OAAO,EAAE;QACnC,GAAG,IAAI;QACP,QAAQ,EAAE;YACT,OAAO,EAAE,IAAI;YACb,YAAY,EAAE,eAAe,CAAC,WAAW,EAAE,MAAM,EAAE,OAAO,CAAC,eAAe,CAAC;SAC3E;KACuB,CAAC,CAAC;AAAA,CAC3B,CAAC;AAEF,SAAS,YAAY,CACpB,KAAoC,EACpC,MAAe,EACf,cAAuC,EACzB;IACd,MAAM,WAAW,GAAgF,EAAE,CAAC;IACpG,IAAI,KAAK,CAAC,OAAO,EAAE,CAAC;QACnB,WAAW,CAAC,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC;QACpC,WAAW,CAAC,UAAU,GAAG,EAAE,CAAC,CAAC,sDAAsD;IACpF,CAAC;IACD,IAAI,KAAK,CAAC,OAAO,IAAI,cAAc,EAAE,CAAC;QACrC,WAAW,CAAC,OAAO,GAAG,EAAE,GAAG,KAAK,CAAC,OAAO,EAAE,GAAG,cAAc,EAAE,CAAC;IAC/D,CAAC;IAED,OAAO,IAAI,WAAW,CAAC;QACtB,MAAM;QACN,WAAW,EAAE,MAAM,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,SAAS;KAC1E,CAAC,CAAC;AAAA,CACH;AAED,SAAS,WAAW,CACnB,KAAoC,EACpC,OAAgB,EAChB,OAAO,GAAkB,EAAE,EACC;IAC5B,MAAM,QAAQ,GAAG,eAAe,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC;IAEjD,MAAM,gBAAgB,GAA0B,EAAE,CAAC;IACnD,IAAI,OAAO,CAAC,WAAW,KAAK,SAAS,EAAE,CAAC;QACvC,gBAAgB,CAAC,WAAW,GAAG,OAAO,CAAC,WAAW,CAAC;IACpD,CAAC;IACD,IAAI,OAAO,CAAC,SAAS,KAAK,SAAS,EAAE,CAAC;QACrC,gBAAgB,CAAC,eAAe,GAAG,OAAO,CAAC,SAAS,CAAC;IACtD,CAAC;IAED,MAAM,MAAM,GAA0B;QACrC,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAC,MAAM,GAAG,CAAC,IAAI,gBAAgB,CAAC;QACjE,GAAG,CAAC,OAAO,CAAC,YAAY,IAAI,EAAE,iBAAiB,EAAE,kBAAkB,CAAC,OAAO,CAAC,YAAY,CAAC,EAAE,CAAC;QAC5F,GAAG,CAAC,OAAO,CAAC,KAAK,IAAI,OAAO,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,KAAK,EAAE,YAAY,CAAC,OAAO,CAAC,KAAK,CAAC,EAAE,CAAC;KACxF,CAAC;IAEF,IAAI,OAAO,CAAC,KAAK,IAAI,OAAO,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,IAAI,OAAO,CAAC,UAAU,EAAE,CAAC;QACrE,MAAM,CAAC,UAAU,GAAG;YACnB,qBAAqB,EAAE;gBACtB,IAAI,EAAE,aAAa,CAAC,OAAO,CAAC,UAAU,CAAC;aACvC;SACD,CAAC;IACH,CAAC;SAAM,CAAC;QACP,MAAM,CAAC,UAAU,GAAG,SAAS,CAAC;IAC/B,CAAC;IAED,IAAI,OAAO,CAAC,QAAQ,EAAE,OAAO,IAAI,KAAK,CAAC,SAAS,EAAE,CAAC;QAClD,MAAM,cAAc,GAAmB,EAAE,eAAe,EAAE,IAAI,EAAE,CAAC;QACjE,IAAI,OAAO,CAAC,QAAQ,CAAC,KAAK,KAAK,SAAS,EAAE,CAAC;YAC1C,uFAAuF;YACvF,cAAc,CAAC,aAAa,GAAG,OAAO,CAAC,QAAQ,CAAC,KAAY,CAAC;QAC9D,CAAC;aAAM,IAAI,OAAO,CAAC,QAAQ,CAAC,YAAY,KAAK,SAAS,EAAE,CAAC;YACxD,cAAc,CAAC,cAAc,GAAG,OAAO,CAAC,QAAQ,CAAC,YAAY,CAAC;QAC/D,CAAC;QACD,MAAM,CAAC,cAAc,GAAG,cAAc,CAAC;IACxC,CAAC;SAAM,IAAI,KAAK,CAAC,SAAS,IAAI,OAAO,CAAC,QAAQ,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,OAAO,EAAE,CAAC;QAC7E,MAAM,CAAC,cAAc,GAAG,yBAAyB,CAAC,KAAK,CAAC,CAAC;IAC1D,CAAC;IAED,IAAI,OAAO,CAAC,MAAM,EAAE,CAAC;QACpB,IAAI,OAAO,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;YAC5B,MAAM,IAAI,KAAK,CAAC,iBAAiB,CAAC,CAAC;QACpC,CAAC;QACD,MAAM,CAAC,WAAW,GAAG,OAAO,CAAC,MAAM,CAAC;IACrC,CAAC;IAED,MAAM,MAAM,GAA8B;QACzC,KAAK,EAAE,KAAK,CAAC,EAAE;QACf,QAAQ;QACR,MAAM;KACN,CAAC;IAEF,OAAO,MAAM,CAAC;AAAA,CACd;AAID,SAAS,aAAa,CAAC,KAAoC,EAAW;IACrE,OAAO,UAAU,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,WAAW,EAAE,CAAC,CAAC;AAAA,CAC/C;AAED,SAAS,iBAAiB,CAAC,KAAoC,EAAW;IACzE,OAAO,wBAAwB,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,WAAW,EAAE,CAAC,CAAC;AAAA,CAC7D;AAED,SAAS,mBAAmB,CAAC,KAAoC,EAAW;IAC3E,OAAO,0BAA0B,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,WAAW,EAAE,CAAC,CAAC;AAAA,CAC/D;AAED,SAAS,yBAAyB,CAAC,KAAoC,EAAkB;IACxF,uFAAuF;IACvF,yFAAyF;IACzF,oFAAoF;IACpF,IAAI,iBAAiB,CAAC,KAAK,CAAC,EAAE,CAAC;QAC9B,OAAO,EAAE,aAAa,EAAE,KAAY,EAAE,CAAC;IACxC,CAAC;IACD,IAAI,mBAAmB,CAAC,KAAK,CAAC,EAAE,CAAC;QAChC,OAAO,EAAE,aAAa,EAAE,SAAgB,EAAE,CAAC;IAC5C,CAAC;IACD,IAAI,aAAa,CAAC,KAAK,CAAC,EAAE,CAAC;QAC1B,OAAO,EAAE,aAAa,EAAE,SAAgB,EAAE,CAAC;IAC5C,CAAC;IAED,wDAAwD;IACxD,OAAO,EAAE,cAAc,EAAE,CAAC,EAAE,CAAC;AAAA,CAC7B;AAED,SAAS,gBAAgB,CAAC,MAA4B,EAAE,KAAoC,EAAuB;IAClH,IAAI,iBAAiB,CAAC,KAAK,CAAC,EAAE,CAAC;QAC9B,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,IAAI,aAAa,CAAC,KAAK,CAAC,EAAE,CAAC;QAC1B,QAAQ,MAAM,EAAE,CAAC;YAChB,KAAK,SAAS,CAAC;YACf,KAAK,KAAK;gBACT,OAAO,SAAS,CAAC;YAClB,KAAK,QAAQ,CAAC;YACd,KAAK,MAAM;gBACV,OAAO,MAAM,CAAC;QAChB,CAAC;IACF,CAAC;IACD,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,CACvB,KAAoC,EACpC,MAA4B,EAC5B,aAA+B,EACtB;IACT,IAAI,aAAa,EAAE,CAAC,MAAM,CAAC,KAAK,SAAS,EAAE,CAAC;QAC3C,OAAO,aAAa,CAAC,MAAM,CAAE,CAAC;IAC/B,CAAC;IAED,IAAI,KAAK,CAAC,EAAE,CAAC,QAAQ,CAAC,SAAS,CAAC,EAAE,CAAC;QAClC,MAAM,OAAO,GAAyC;YACrD,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,gBAAgB,CAAC,EAAE,CAAC;QACzC,MAAM,OAAO,GAAyC;YACrD,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,MAAM,OAAO,GAAyC;YACrD,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,OAAO,CAAC,CAAC,CAAC;AAAA,CACV","sourcesContent":["import {\n\ttype GenerateContentConfig,\n\ttype GenerateContentParameters,\n\tGoogleGenAI,\n\ttype ThinkingConfig,\n} from \"@google/genai\";\nimport { getEnvApiKey } from \"../env-api-keys.js\";\nimport { calculateCost } from \"../models.js\";\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.js\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.js\";\nimport { sanitizeSurrogates } from \"../utils/sanitize-unicode.js\";\nimport type { GoogleThinkingLevel } from \"./google-gemini-cli.js\";\nimport {\n\tconvertMessages,\n\tconvertTools,\n\tisThinkingPart,\n\tmapStopReason,\n\tmapToolChoice,\n\tretainThoughtSignature,\n} from \"./google-shared.js\";\nimport { buildBaseOptions, clampReasoning } from \"./simple-options.js\";\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, options, apiKey);\n\tif (!options?.reasoning) {\n\t\treturn streamGoogle(model, context, { ...base, thinking: { enabled: false } } satisfies GoogleOptions);\n\t}\n\n\tconst effort = clampReasoning(options.reasoning)!;\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"]}
|
|
@@ -2,6 +2,7 @@ import type { SimpleStreamOptions, StreamFunction, StreamOptions } from "../type
|
|
|
2
2
|
/**
|
|
3
3
|
* Provider-specific options for the Mistral API.
|
|
4
4
|
*/
|
|
5
|
+
type MistralReasoningEffort = "none" | "high";
|
|
5
6
|
export interface MistralOptions extends StreamOptions {
|
|
6
7
|
toolChoice?: "auto" | "none" | "any" | "required" | {
|
|
7
8
|
type: "function";
|
|
@@ -10,6 +11,7 @@ export interface MistralOptions extends StreamOptions {
|
|
|
10
11
|
};
|
|
11
12
|
};
|
|
12
13
|
promptMode?: "reasoning";
|
|
14
|
+
reasoningEffort?: MistralReasoningEffort;
|
|
13
15
|
}
|
|
14
16
|
/**
|
|
15
17
|
* Stream responses from Mistral using `chat.stream`.
|
|
@@ -19,4 +21,5 @@ export declare const streamMistral: StreamFunction<"mistral-conversations", Mist
|
|
|
19
21
|
* Maps provider-agnostic `SimpleStreamOptions` to Mistral options.
|
|
20
22
|
*/
|
|
21
23
|
export declare const streamSimpleMistral: StreamFunction<"mistral-conversations", SimpleStreamOptions>;
|
|
24
|
+
export {};
|
|
22
25
|
//# sourceMappingURL=mistral.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"mistral.d.ts","sourceRoot":"","sources":["../../src/providers/mistral.ts"],"names":[],"mappings":"AAWA,OAAO,KAAK,EAKX,mBAAmB,EAEnB,cAAc,EACd,aAAa,EAKb,MAAM,aAAa,CAAC;AAWrB;;GAEG;AACH,MAAM,WAAW,cAAe,SAAQ,aAAa;IACpD,UAAU,CAAC,EAAE,MAAM,GAAG,MAAM,GAAG,KAAK,GAAG,UAAU,GAAG;QAAE,IAAI,EAAE,UAAU,CAAC;QAAC,QAAQ,EAAE;YAAE,IAAI,EAAE,MAAM,CAAA;SAAE,CAAA;KAAE,CAAC;IACrG,UAAU,CAAC,EAAE,WAAW,CAAC;CACzB;AAED;;GAEG;AACH,eAAO,MAAM,aAAa,EAAE,cAAc,CAAC,uBAAuB,EAAE,cAAc,CAqDjF,CAAC;AAEF;;GAEG;AACH,eAAO,MAAM,mBAAmB,EAAE,cAAc,CAAC,uBAAuB,EAAE,mBAAmB,CAiB5F,CAAC","sourcesContent":["import { Mistral } from \"@mistralai/mistralai\";\nimport type { RequestOptions } from \"@mistralai/mistralai/lib/sdks.js\";\nimport type {\n\tChatCompletionStreamRequest,\n\tChatCompletionStreamRequestMessages,\n\tCompletionEvent,\n\tContentChunk,\n\tFunctionTool,\n} from \"@mistralai/mistralai/models/components/index.js\";\nimport { getEnvApiKey } from \"../env-api-keys.js\";\nimport { calculateCost } from \"../models.js\";\nimport type {\n\tAssistantMessage,\n\tContext,\n\tMessage,\n\tModel,\n\tSimpleStreamOptions,\n\tStopReason,\n\tStreamFunction,\n\tStreamOptions,\n\tTextContent,\n\tThinkingContent,\n\tTool,\n\tToolCall,\n} from \"../types.js\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.js\";\nimport { shortHash } from \"../utils/hash.js\";\nimport { parseStreamingJson } from \"../utils/json-parse.js\";\nimport { sanitizeSurrogates } from \"../utils/sanitize-unicode.js\";\nimport { buildBaseOptions, clampReasoning } from \"./simple-options.js\";\nimport { transformMessages } from \"./transform-messages.js\";\n\nconst MISTRAL_TOOL_CALL_ID_LENGTH = 9;\nconst MAX_MISTRAL_ERROR_BODY_CHARS = 4000;\n\n/**\n * Provider-specific options for the Mistral API.\n */\nexport interface MistralOptions extends StreamOptions {\n\ttoolChoice?: \"auto\" | \"none\" | \"any\" | \"required\" | { type: \"function\"; function: { name: string } };\n\tpromptMode?: \"reasoning\";\n}\n\n/**\n * Stream responses from Mistral using `chat.stream`.\n */\nexport const streamMistral: StreamFunction<\"mistral-conversations\", MistralOptions> = (\n\tmodel: Model<\"mistral-conversations\">,\n\tcontext: Context,\n\toptions?: MistralOptions,\n): AssistantMessageEventStream => {\n\tconst stream = new AssistantMessageEventStream();\n\n\t(async () => {\n\t\tconst output = createOutput(model);\n\n\t\ttry {\n\t\t\tconst apiKey = options?.apiKey || getEnvApiKey(model.provider);\n\t\t\tif (!apiKey) {\n\t\t\t\tthrow new Error(`No API key for provider: ${model.provider}`);\n\t\t\t}\n\n\t\t\t// Intentionally per-request: avoids shared SDK mutable state across concurrent consumers.\n\t\t\tconst mistral = new Mistral({\n\t\t\t\tapiKey,\n\t\t\t\tserverURL: model.baseUrl,\n\t\t\t});\n\n\t\t\tconst normalizeMistralToolCallId = createMistralToolCallIdNormalizer();\n\t\t\tconst transformedMessages = transformMessages(context.messages, model, (id) => normalizeMistralToolCallId(id));\n\n\t\t\tlet payload = buildChatPayload(model, context, transformedMessages, options);\n\t\t\tconst nextPayload = await options?.onPayload?.(payload, model);\n\t\t\tif (nextPayload !== undefined) {\n\t\t\t\tpayload = nextPayload as ChatCompletionStreamRequest;\n\t\t\t}\n\t\t\tconst mistralStream = await mistral.chat.stream(payload, buildRequestOptions(model, options));\n\t\t\tstream.push({ type: \"start\", partial: output });\n\t\t\tawait consumeChatStream(model, output, stream, mistralStream);\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\toutput.stopReason = options?.signal?.aborted ? \"aborted\" : \"error\";\n\t\t\toutput.errorMessage = formatMistralError(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\n/**\n * Maps provider-agnostic `SimpleStreamOptions` to Mistral options.\n */\nexport const streamSimpleMistral: StreamFunction<\"mistral-conversations\", SimpleStreamOptions> = (\n\tmodel: Model<\"mistral-conversations\">,\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: ${model.provider}`);\n\t}\n\n\tconst base = buildBaseOptions(model, options, apiKey);\n\tconst reasoning = clampReasoning(options?.reasoning);\n\n\treturn streamMistral(model, context, {\n\t\t...base,\n\t\tpromptMode: model.reasoning && reasoning ? \"reasoning\" : undefined,\n\t} satisfies MistralOptions);\n};\n\nfunction createOutput(model: Model<\"mistral-conversations\">): AssistantMessage {\n\treturn {\n\t\trole: \"assistant\",\n\t\tcontent: [],\n\t\tapi: model.api,\n\t\tprovider: model.provider,\n\t\tmodel: model.id,\n\t\tusage: {\n\t\t\tinput: 0,\n\t\t\toutput: 0,\n\t\t\tcacheRead: 0,\n\t\t\tcacheWrite: 0,\n\t\t\ttotalTokens: 0,\n\t\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t\t},\n\t\tstopReason: \"stop\",\n\t\ttimestamp: Date.now(),\n\t};\n}\n\nfunction createMistralToolCallIdNormalizer(): (id: string) => string {\n\tconst idMap = new Map<string, string>();\n\tconst reverseMap = new Map<string, string>();\n\n\treturn (id: string): string => {\n\t\tconst existing = idMap.get(id);\n\t\tif (existing) return existing;\n\n\t\tlet attempt = 0;\n\t\twhile (true) {\n\t\t\tconst candidate = deriveMistralToolCallId(id, attempt);\n\t\t\tconst owner = reverseMap.get(candidate);\n\t\t\tif (!owner || owner === id) {\n\t\t\t\tidMap.set(id, candidate);\n\t\t\t\treverseMap.set(candidate, id);\n\t\t\t\treturn candidate;\n\t\t\t}\n\t\t\tattempt++;\n\t\t}\n\t};\n}\n\nfunction deriveMistralToolCallId(id: string, attempt: number): string {\n\tconst normalized = id.replace(/[^a-zA-Z0-9]/g, \"\");\n\tif (attempt === 0 && normalized.length === MISTRAL_TOOL_CALL_ID_LENGTH) return normalized;\n\tconst seedBase = normalized || id;\n\tconst seed = attempt === 0 ? seedBase : `${seedBase}:${attempt}`;\n\treturn shortHash(seed)\n\t\t.replace(/[^a-zA-Z0-9]/g, \"\")\n\t\t.slice(0, MISTRAL_TOOL_CALL_ID_LENGTH);\n}\n\nfunction formatMistralError(error: unknown): string {\n\tif (error instanceof Error) {\n\t\tconst sdkError = error as Error & { statusCode?: unknown; body?: unknown };\n\t\tconst statusCode = typeof sdkError.statusCode === \"number\" ? sdkError.statusCode : undefined;\n\t\tconst bodyText = typeof sdkError.body === \"string\" ? sdkError.body.trim() : undefined;\n\t\tif (statusCode !== undefined && bodyText) {\n\t\t\treturn `Mistral API error (${statusCode}): ${truncateErrorText(bodyText, MAX_MISTRAL_ERROR_BODY_CHARS)}`;\n\t\t}\n\t\tif (statusCode !== undefined) return `Mistral API error (${statusCode}): ${error.message}`;\n\t\treturn error.message;\n\t}\n\treturn safeJsonStringify(error);\n}\n\nfunction truncateErrorText(text: string, maxChars: number): string {\n\tif (text.length <= maxChars) return text;\n\treturn `${text.slice(0, maxChars)}... [truncated ${text.length - maxChars} chars]`;\n}\n\nfunction safeJsonStringify(value: unknown): string {\n\ttry {\n\t\tconst serialized = JSON.stringify(value);\n\t\treturn serialized === undefined ? String(value) : serialized;\n\t} catch {\n\t\treturn String(value);\n\t}\n}\n\nfunction buildRequestOptions(model: Model<\"mistral-conversations\">, options?: MistralOptions): RequestOptions {\n\tconst requestOptions: RequestOptions = {};\n\tif (options?.signal) requestOptions.signal = options.signal;\n\trequestOptions.retries = { strategy: \"none\" };\n\n\tconst headers: Record<string, string> = {};\n\tif (model.headers) Object.assign(headers, model.headers);\n\tif (options?.headers) Object.assign(headers, options.headers);\n\n\t// Mistral infrastructure uses `x-affinity` for KV-cache reuse (prefix caching).\n\t// Respect explicit caller-provided header values.\n\tif (options?.sessionId && !headers[\"x-affinity\"]) {\n\t\theaders[\"x-affinity\"] = options.sessionId;\n\t}\n\n\tif (Object.keys(headers).length > 0) {\n\t\trequestOptions.headers = headers;\n\t}\n\n\treturn requestOptions;\n}\n\nfunction buildChatPayload(\n\tmodel: Model<\"mistral-conversations\">,\n\tcontext: Context,\n\tmessages: Message[],\n\toptions?: MistralOptions,\n): ChatCompletionStreamRequest {\n\tconst payload: ChatCompletionStreamRequest = {\n\t\tmodel: model.id,\n\t\tstream: true,\n\t\tmessages: toChatMessages(messages, model.input.includes(\"image\")),\n\t};\n\n\tif (context.tools?.length) payload.tools = toFunctionTools(context.tools);\n\tif (options?.temperature !== undefined) payload.temperature = options.temperature;\n\tif (options?.maxTokens !== undefined) payload.maxTokens = options.maxTokens;\n\tif (options?.toolChoice) payload.toolChoice = mapToolChoice(options.toolChoice);\n\tif (options?.promptMode) payload.promptMode = options.promptMode as any;\n\n\tif (context.systemPrompt) {\n\t\tpayload.messages.unshift({\n\t\t\trole: \"system\",\n\t\t\tcontent: sanitizeSurrogates(context.systemPrompt),\n\t\t});\n\t}\n\n\treturn payload;\n}\n\nasync function consumeChatStream(\n\tmodel: Model<\"mistral-conversations\">,\n\toutput: AssistantMessage,\n\tstream: AssistantMessageEventStream,\n\tmistralStream: AsyncIterable<CompletionEvent>,\n): Promise<void> {\n\tlet currentBlock: TextContent | ThinkingContent | null = null;\n\tconst blocks = output.content;\n\tconst blockIndex = () => blocks.length - 1;\n\tconst toolBlocksByKey = new Map<string, number>();\n\n\tconst finishCurrentBlock = (block?: typeof currentBlock) => {\n\t\tif (!block) return;\n\t\tif (block.type === \"text\") {\n\t\t\tstream.push({\n\t\t\t\ttype: \"text_end\",\n\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\tcontent: block.text,\n\t\t\t\tpartial: output,\n\t\t\t});\n\t\t\treturn;\n\t\t}\n\t\tif (block.type === \"thinking\") {\n\t\t\tstream.push({\n\t\t\t\ttype: \"thinking_end\",\n\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\tcontent: block.thinking,\n\t\t\t\tpartial: output,\n\t\t\t});\n\t\t}\n\t};\n\n\tfor await (const event of mistralStream) {\n\t\tconst chunk = event.data;\n\t\t// Mistral's streamed CompletionChunk carries an id field. Keep the first non-empty one,\n\t\t// mirroring how OpenAI-style streaming exposes a stable response identifier per stream.\n\t\toutput.responseId ||= chunk.id;\n\n\t\tif (chunk.usage) {\n\t\t\toutput.usage.input = chunk.usage.promptTokens || 0;\n\t\t\toutput.usage.output = chunk.usage.completionTokens || 0;\n\t\t\toutput.usage.cacheRead = 0;\n\t\t\toutput.usage.cacheWrite = 0;\n\t\t\toutput.usage.totalTokens = chunk.usage.totalTokens || output.usage.input + output.usage.output;\n\t\t\tcalculateCost(model, output.usage);\n\t\t}\n\n\t\tconst choice = chunk.choices[0];\n\t\tif (!choice) continue;\n\n\t\tif (choice.finishReason) {\n\t\t\toutput.stopReason = mapChatStopReason(choice.finishReason);\n\t\t}\n\n\t\tconst delta = choice.delta;\n\t\tif (delta.content !== null && delta.content !== undefined) {\n\t\t\tconst contentItems = typeof delta.content === \"string\" ? [delta.content] : delta.content;\n\t\t\tfor (const item of contentItems) {\n\t\t\t\tif (typeof item === \"string\") {\n\t\t\t\t\tconst textDelta = sanitizeSurrogates(item);\n\t\t\t\t\tif (!currentBlock || currentBlock.type !== \"text\") {\n\t\t\t\t\t\tfinishCurrentBlock(currentBlock);\n\t\t\t\t\t\tcurrentBlock = { type: \"text\", text: \"\" };\n\t\t\t\t\t\toutput.content.push(currentBlock);\n\t\t\t\t\t\tstream.push({ type: \"text_start\", contentIndex: blockIndex(), partial: output });\n\t\t\t\t\t}\n\t\t\t\t\tcurrentBlock.text += textDelta;\n\t\t\t\t\tstream.push({\n\t\t\t\t\t\ttype: \"text_delta\",\n\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\tdelta: textDelta,\n\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t});\n\t\t\t\t\tcontinue;\n\t\t\t\t}\n\n\t\t\t\tif (item.type === \"thinking\") {\n\t\t\t\t\tconst deltaText = item.thinking\n\t\t\t\t\t\t.map((part) => (\"text\" in part ? part.text : \"\"))\n\t\t\t\t\t\t.filter((text) => text.length > 0)\n\t\t\t\t\t\t.join(\"\");\n\t\t\t\t\tconst thinkingDelta = sanitizeSurrogates(deltaText);\n\t\t\t\t\tif (!thinkingDelta) continue;\n\t\t\t\t\tif (!currentBlock || currentBlock.type !== \"thinking\") {\n\t\t\t\t\t\tfinishCurrentBlock(currentBlock);\n\t\t\t\t\t\tcurrentBlock = { type: \"thinking\", thinking: \"\" };\n\t\t\t\t\t\toutput.content.push(currentBlock);\n\t\t\t\t\t\tstream.push({ type: \"thinking_start\", contentIndex: blockIndex(), partial: output });\n\t\t\t\t\t}\n\t\t\t\t\tcurrentBlock.thinking += thinkingDelta;\n\t\t\t\t\tstream.push({\n\t\t\t\t\t\ttype: \"thinking_delta\",\n\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\tdelta: thinkingDelta,\n\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t});\n\t\t\t\t\tcontinue;\n\t\t\t\t}\n\n\t\t\t\tif (item.type === \"text\") {\n\t\t\t\t\tconst textDelta = sanitizeSurrogates(item.text);\n\t\t\t\t\tif (!currentBlock || currentBlock.type !== \"text\") {\n\t\t\t\t\t\tfinishCurrentBlock(currentBlock);\n\t\t\t\t\t\tcurrentBlock = { type: \"text\", text: \"\" };\n\t\t\t\t\t\toutput.content.push(currentBlock);\n\t\t\t\t\t\tstream.push({ type: \"text_start\", contentIndex: blockIndex(), partial: output });\n\t\t\t\t\t}\n\t\t\t\t\tcurrentBlock.text += textDelta;\n\t\t\t\t\tstream.push({\n\t\t\t\t\t\ttype: \"text_delta\",\n\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\tdelta: textDelta,\n\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t});\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\tconst toolCalls = delta.toolCalls || [];\n\t\tfor (const toolCall of toolCalls) {\n\t\t\tif (currentBlock) {\n\t\t\t\tfinishCurrentBlock(currentBlock);\n\t\t\t\tcurrentBlock = null;\n\t\t\t}\n\t\t\tconst callId =\n\t\t\t\ttoolCall.id && toolCall.id !== \"null\"\n\t\t\t\t\t? toolCall.id\n\t\t\t\t\t: deriveMistralToolCallId(`toolcall:${toolCall.index ?? 0}`, 0);\n\t\t\tconst key = `${callId}:${toolCall.index || 0}`;\n\t\t\tconst existingIndex = toolBlocksByKey.get(key);\n\t\t\tlet block: (ToolCall & { partialArgs?: string }) | undefined;\n\n\t\t\tif (existingIndex !== undefined) {\n\t\t\t\tconst existing = output.content[existingIndex];\n\t\t\t\tif (existing?.type === \"toolCall\") {\n\t\t\t\t\tblock = existing as ToolCall & { partialArgs?: string };\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tif (!block) {\n\t\t\t\tblock = {\n\t\t\t\t\ttype: \"toolCall\",\n\t\t\t\t\tid: callId,\n\t\t\t\t\tname: toolCall.function.name,\n\t\t\t\t\targuments: {},\n\t\t\t\t\tpartialArgs: \"\",\n\t\t\t\t};\n\t\t\t\toutput.content.push(block);\n\t\t\t\ttoolBlocksByKey.set(key, output.content.length - 1);\n\t\t\t\tstream.push({ type: \"toolcall_start\", contentIndex: output.content.length - 1, partial: output });\n\t\t\t}\n\n\t\t\tconst argsDelta =\n\t\t\t\ttypeof toolCall.function.arguments === \"string\"\n\t\t\t\t\t? toolCall.function.arguments\n\t\t\t\t\t: JSON.stringify(toolCall.function.arguments || {});\n\t\t\tblock.partialArgs = (block.partialArgs || \"\") + argsDelta;\n\t\t\tblock.arguments = parseStreamingJson<Record<string, unknown>>(block.partialArgs);\n\t\t\tstream.push({\n\t\t\t\ttype: \"toolcall_delta\",\n\t\t\t\tcontentIndex: toolBlocksByKey.get(key)!,\n\t\t\t\tdelta: argsDelta,\n\t\t\t\tpartial: output,\n\t\t\t});\n\t\t}\n\t}\n\n\tfinishCurrentBlock(currentBlock);\n\tfor (const index of toolBlocksByKey.values()) {\n\t\tconst block = output.content[index];\n\t\tif (block.type !== \"toolCall\") continue;\n\t\tconst toolBlock = block as ToolCall & { partialArgs?: string };\n\t\ttoolBlock.arguments = parseStreamingJson<Record<string, unknown>>(toolBlock.partialArgs);\n\t\tdelete toolBlock.partialArgs;\n\t\tstream.push({\n\t\t\ttype: \"toolcall_end\",\n\t\t\tcontentIndex: index,\n\t\t\ttoolCall: toolBlock,\n\t\t\tpartial: output,\n\t\t});\n\t}\n}\n\nfunction toFunctionTools(tools: Tool[]): Array<FunctionTool & { type: \"function\" }> {\n\treturn tools.map((tool) => ({\n\t\ttype: \"function\",\n\t\tfunction: {\n\t\t\tname: tool.name,\n\t\t\tdescription: tool.description,\n\t\t\tparameters: tool.parameters as unknown as Record<string, unknown>,\n\t\t\tstrict: false,\n\t\t},\n\t}));\n}\n\nfunction toChatMessages(messages: Message[], supportsImages: boolean): ChatCompletionStreamRequestMessages[] {\n\tconst result: ChatCompletionStreamRequestMessages[] = [];\n\n\tfor (const msg of messages) {\n\t\tif (msg.role === \"user\") {\n\t\t\tif (typeof msg.content === \"string\") {\n\t\t\t\tresult.push({ role: \"user\", content: sanitizeSurrogates(msg.content) });\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tconst hadImages = msg.content.some((item) => item.type === \"image\");\n\t\t\tconst content: ContentChunk[] = msg.content\n\t\t\t\t.filter((item) => item.type === \"text\" || supportsImages)\n\t\t\t\t.map((item) => {\n\t\t\t\t\tif (item.type === \"text\") return { type: \"text\", text: sanitizeSurrogates(item.text) };\n\t\t\t\t\treturn { type: \"image_url\", imageUrl: `data:${item.mimeType};base64,${item.data}` };\n\t\t\t\t});\n\t\t\tif (content.length > 0) {\n\t\t\t\tresult.push({ role: \"user\", content });\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tif (hadImages && !supportsImages) {\n\t\t\t\tresult.push({ role: \"user\", content: \"(image omitted: model does not support images)\" });\n\t\t\t}\n\t\t\tcontinue;\n\t\t}\n\n\t\tif (msg.role === \"assistant\") {\n\t\t\tconst contentParts: ContentChunk[] = [];\n\t\t\tconst toolCalls: Array<{ id: string; type: \"function\"; function: { name: string; arguments: string } }> = [];\n\n\t\t\tfor (const block of msg.content) {\n\t\t\t\tif (block.type === \"text\") {\n\t\t\t\t\tif (block.text.trim().length > 0) {\n\t\t\t\t\t\tcontentParts.push({ type: \"text\", text: sanitizeSurrogates(block.text) });\n\t\t\t\t\t}\n\t\t\t\t\tcontinue;\n\t\t\t\t}\n\t\t\t\tif (block.type === \"thinking\") {\n\t\t\t\t\tif (block.thinking.trim().length > 0) {\n\t\t\t\t\t\tcontentParts.push({\n\t\t\t\t\t\t\ttype: \"thinking\",\n\t\t\t\t\t\t\tthinking: [{ type: \"text\", text: sanitizeSurrogates(block.thinking) }],\n\t\t\t\t\t\t});\n\t\t\t\t\t}\n\t\t\t\t\tcontinue;\n\t\t\t\t}\n\t\t\t\ttoolCalls.push({\n\t\t\t\t\tid: block.id,\n\t\t\t\t\ttype: \"function\",\n\t\t\t\t\tfunction: { name: block.name, arguments: JSON.stringify(block.arguments || {}) },\n\t\t\t\t});\n\t\t\t}\n\n\t\t\tconst assistantMessage: ChatCompletionStreamRequestMessages = { role: \"assistant\" };\n\t\t\tif (contentParts.length > 0) assistantMessage.content = contentParts;\n\t\t\tif (toolCalls.length > 0) assistantMessage.toolCalls = toolCalls;\n\t\t\tif (contentParts.length > 0 || toolCalls.length > 0) result.push(assistantMessage);\n\t\t\tcontinue;\n\t\t}\n\n\t\tconst toolContent: ContentChunk[] = [];\n\t\tconst textResult = msg.content\n\t\t\t.filter((part) => part.type === \"text\")\n\t\t\t.map((part) => (part.type === \"text\" ? sanitizeSurrogates(part.text) : \"\"))\n\t\t\t.join(\"\\n\");\n\t\tconst hasImages = msg.content.some((part) => part.type === \"image\");\n\t\tconst toolText = buildToolResultText(textResult, hasImages, supportsImages, msg.isError);\n\t\ttoolContent.push({ type: \"text\", text: toolText });\n\t\tfor (const part of msg.content) {\n\t\t\tif (!supportsImages) continue;\n\t\t\tif (part.type !== \"image\") continue;\n\t\t\ttoolContent.push({\n\t\t\t\ttype: \"image_url\",\n\t\t\t\timageUrl: `data:${part.mimeType};base64,${part.data}`,\n\t\t\t});\n\t\t}\n\t\tresult.push({\n\t\t\trole: \"tool\",\n\t\t\ttoolCallId: msg.toolCallId,\n\t\t\tname: msg.toolName,\n\t\t\tcontent: toolContent,\n\t\t});\n\t}\n\n\treturn result;\n}\n\nfunction buildToolResultText(text: string, hasImages: boolean, supportsImages: boolean, isError: boolean): string {\n\tconst trimmed = text.trim();\n\tconst errorPrefix = isError ? \"[tool error] \" : \"\";\n\n\tif (trimmed.length > 0) {\n\t\tconst imageSuffix = hasImages && !supportsImages ? \"\\n[tool image omitted: model does not support images]\" : \"\";\n\t\treturn `${errorPrefix}${trimmed}${imageSuffix}`;\n\t}\n\n\tif (hasImages) {\n\t\tif (supportsImages) {\n\t\t\treturn isError ? \"[tool error] (see attached image)\" : \"(see attached image)\";\n\t\t}\n\t\treturn isError\n\t\t\t? \"[tool error] (image omitted: model does not support images)\"\n\t\t\t: \"(image omitted: model does not support images)\";\n\t}\n\n\treturn isError ? \"[tool error] (no tool output)\" : \"(no tool output)\";\n}\n\nfunction mapToolChoice(\n\tchoice: MistralOptions[\"toolChoice\"],\n): \"auto\" | \"none\" | \"any\" | \"required\" | { type: \"function\"; function: { name: string } } | undefined {\n\tif (!choice) return undefined;\n\tif (choice === \"auto\" || choice === \"none\" || choice === \"any\" || choice === \"required\") {\n\t\treturn choice as any;\n\t}\n\treturn {\n\t\ttype: \"function\",\n\t\tfunction: { name: choice.function.name },\n\t};\n}\n\nfunction mapChatStopReason(reason: string | null): StopReason {\n\tif (reason === null) return \"stop\";\n\tswitch (reason) {\n\t\tcase \"stop\":\n\t\t\treturn \"stop\";\n\t\tcase \"length\":\n\t\tcase \"model_length\":\n\t\t\treturn \"length\";\n\t\tcase \"tool_calls\":\n\t\t\treturn \"toolUse\";\n\t\tcase \"error\":\n\t\t\treturn \"error\";\n\t\tdefault:\n\t\t\treturn \"stop\";\n\t}\n}\n"]}
|
|
1
|
+
{"version":3,"file":"mistral.d.ts","sourceRoot":"","sources":["../../src/providers/mistral.ts"],"names":[],"mappings":"AAUA,OAAO,KAAK,EAKX,mBAAmB,EAEnB,cAAc,EACd,aAAa,EAKb,MAAM,aAAa,CAAC;AAWrB;;GAEG;AACH,KAAK,sBAAsB,GAAG,MAAM,GAAG,MAAM,CAAC;AAE9C,MAAM,WAAW,cAAe,SAAQ,aAAa;IACpD,UAAU,CAAC,EAAE,MAAM,GAAG,MAAM,GAAG,KAAK,GAAG,UAAU,GAAG;QAAE,IAAI,EAAE,UAAU,CAAC;QAAC,QAAQ,EAAE;YAAE,IAAI,EAAE,MAAM,CAAA;SAAE,CAAA;KAAE,CAAC;IACrG,UAAU,CAAC,EAAE,WAAW,CAAC;IACzB,eAAe,CAAC,EAAE,sBAAsB,CAAC;CACzC;AAED;;GAEG;AACH,eAAO,MAAM,aAAa,EAAE,cAAc,CAAC,uBAAuB,EAAE,cAAc,CAyDjF,CAAC;AAEF;;GAEG;AACH,eAAO,MAAM,mBAAmB,EAAE,cAAc,CAAC,uBAAuB,EAAE,mBAAmB,CAmB5F,CAAC","sourcesContent":["import { Mistral } from \"@mistralai/mistralai\";\nimport type {\n\tChatCompletionStreamRequest,\n\tChatCompletionStreamRequestMessage,\n\tCompletionEvent,\n\tContentChunk,\n\tFunctionTool,\n} from \"@mistralai/mistralai/models/components\";\nimport { getEnvApiKey } from \"../env-api-keys.js\";\nimport { calculateCost } from \"../models.js\";\nimport type {\n\tAssistantMessage,\n\tContext,\n\tMessage,\n\tModel,\n\tSimpleStreamOptions,\n\tStopReason,\n\tStreamFunction,\n\tStreamOptions,\n\tTextContent,\n\tThinkingContent,\n\tTool,\n\tToolCall,\n} from \"../types.js\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.js\";\nimport { shortHash } from \"../utils/hash.js\";\nimport { parseStreamingJson } from \"../utils/json-parse.js\";\nimport { sanitizeSurrogates } from \"../utils/sanitize-unicode.js\";\nimport { buildBaseOptions, clampReasoning } from \"./simple-options.js\";\nimport { transformMessages } from \"./transform-messages.js\";\n\nconst MISTRAL_TOOL_CALL_ID_LENGTH = 9;\nconst MAX_MISTRAL_ERROR_BODY_CHARS = 4000;\n\n/**\n * Provider-specific options for the Mistral API.\n */\ntype MistralReasoningEffort = \"none\" | \"high\";\n\nexport interface MistralOptions extends StreamOptions {\n\ttoolChoice?: \"auto\" | \"none\" | \"any\" | \"required\" | { type: \"function\"; function: { name: string } };\n\tpromptMode?: \"reasoning\";\n\treasoningEffort?: MistralReasoningEffort;\n}\n\n/**\n * Stream responses from Mistral using `chat.stream`.\n */\nexport const streamMistral: StreamFunction<\"mistral-conversations\", MistralOptions> = (\n\tmodel: Model<\"mistral-conversations\">,\n\tcontext: Context,\n\toptions?: MistralOptions,\n): AssistantMessageEventStream => {\n\tconst stream = new AssistantMessageEventStream();\n\n\t(async () => {\n\t\tconst output = createOutput(model);\n\n\t\ttry {\n\t\t\tconst apiKey = options?.apiKey || getEnvApiKey(model.provider);\n\t\t\tif (!apiKey) {\n\t\t\t\tthrow new Error(`No API key for provider: ${model.provider}`);\n\t\t\t}\n\n\t\t\t// Intentionally per-request: avoids shared SDK mutable state across concurrent consumers.\n\t\t\tconst mistral = new Mistral({\n\t\t\t\tapiKey,\n\t\t\t\tserverURL: model.baseUrl,\n\t\t\t});\n\n\t\t\tconst normalizeMistralToolCallId = createMistralToolCallIdNormalizer();\n\t\t\tconst transformedMessages = transformMessages(context.messages, model, (id) => normalizeMistralToolCallId(id));\n\n\t\t\tlet payload = buildChatPayload(model, context, transformedMessages, options);\n\t\t\tconst nextPayload = await options?.onPayload?.(payload, model);\n\t\t\tif (nextPayload !== undefined) {\n\t\t\t\tpayload = nextPayload as ChatCompletionStreamRequest;\n\t\t\t}\n\t\t\tconst mistralStream = await mistral.chat.stream(payload, buildRequestOptions(model, options));\n\t\t\tstream.push({ type: \"start\", partial: output });\n\t\t\tawait consumeChatStream(model, output, stream, mistralStream);\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\tfor (const block of output.content) {\n\t\t\t\t// partialArgs is only a streaming scratch buffer; never persist it.\n\t\t\t\tdelete (block as { partialArgs?: string }).partialArgs;\n\t\t\t}\n\t\t\toutput.stopReason = options?.signal?.aborted ? \"aborted\" : \"error\";\n\t\t\toutput.errorMessage = formatMistralError(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\n/**\n * Maps provider-agnostic `SimpleStreamOptions` to Mistral options.\n */\nexport const streamSimpleMistral: StreamFunction<\"mistral-conversations\", SimpleStreamOptions> = (\n\tmodel: Model<\"mistral-conversations\">,\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: ${model.provider}`);\n\t}\n\n\tconst base = buildBaseOptions(model, options, apiKey);\n\tconst reasoning = clampReasoning(options?.reasoning);\n\tconst shouldUseReasoning = model.reasoning && reasoning !== undefined;\n\n\treturn streamMistral(model, context, {\n\t\t...base,\n\t\tpromptMode: shouldUseReasoning && usesPromptModeReasoning(model) ? \"reasoning\" : undefined,\n\t\treasoningEffort: shouldUseReasoning && usesReasoningEffort(model) ? mapReasoningEffort(reasoning) : undefined,\n\t} satisfies MistralOptions);\n};\n\nfunction createOutput(model: Model<\"mistral-conversations\">): AssistantMessage {\n\treturn {\n\t\trole: \"assistant\",\n\t\tcontent: [],\n\t\tapi: model.api,\n\t\tprovider: model.provider,\n\t\tmodel: model.id,\n\t\tusage: {\n\t\t\tinput: 0,\n\t\t\toutput: 0,\n\t\t\tcacheRead: 0,\n\t\t\tcacheWrite: 0,\n\t\t\ttotalTokens: 0,\n\t\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t\t},\n\t\tstopReason: \"stop\",\n\t\ttimestamp: Date.now(),\n\t};\n}\n\nfunction createMistralToolCallIdNormalizer(): (id: string) => string {\n\tconst idMap = new Map<string, string>();\n\tconst reverseMap = new Map<string, string>();\n\n\treturn (id: string): string => {\n\t\tconst existing = idMap.get(id);\n\t\tif (existing) return existing;\n\n\t\tlet attempt = 0;\n\t\twhile (true) {\n\t\t\tconst candidate = deriveMistralToolCallId(id, attempt);\n\t\t\tconst owner = reverseMap.get(candidate);\n\t\t\tif (!owner || owner === id) {\n\t\t\t\tidMap.set(id, candidate);\n\t\t\t\treverseMap.set(candidate, id);\n\t\t\t\treturn candidate;\n\t\t\t}\n\t\t\tattempt++;\n\t\t}\n\t};\n}\n\nfunction deriveMistralToolCallId(id: string, attempt: number): string {\n\tconst normalized = id.replace(/[^a-zA-Z0-9]/g, \"\");\n\tif (attempt === 0 && normalized.length === MISTRAL_TOOL_CALL_ID_LENGTH) return normalized;\n\tconst seedBase = normalized || id;\n\tconst seed = attempt === 0 ? seedBase : `${seedBase}:${attempt}`;\n\treturn shortHash(seed)\n\t\t.replace(/[^a-zA-Z0-9]/g, \"\")\n\t\t.slice(0, MISTRAL_TOOL_CALL_ID_LENGTH);\n}\n\nfunction formatMistralError(error: unknown): string {\n\tif (error instanceof Error) {\n\t\tconst sdkError = error as Error & { statusCode?: unknown; body?: unknown };\n\t\tconst statusCode = typeof sdkError.statusCode === \"number\" ? sdkError.statusCode : undefined;\n\t\tconst bodyText = typeof sdkError.body === \"string\" ? sdkError.body.trim() : undefined;\n\t\tif (statusCode !== undefined && bodyText) {\n\t\t\treturn `Mistral API error (${statusCode}): ${truncateErrorText(bodyText, MAX_MISTRAL_ERROR_BODY_CHARS)}`;\n\t\t}\n\t\tif (statusCode !== undefined) return `Mistral API error (${statusCode}): ${error.message}`;\n\t\treturn error.message;\n\t}\n\treturn safeJsonStringify(error);\n}\n\nfunction truncateErrorText(text: string, maxChars: number): string {\n\tif (text.length <= maxChars) return text;\n\treturn `${text.slice(0, maxChars)}... [truncated ${text.length - maxChars} chars]`;\n}\n\nfunction safeJsonStringify(value: unknown): string {\n\ttry {\n\t\tconst serialized = JSON.stringify(value);\n\t\treturn serialized === undefined ? String(value) : serialized;\n\t} catch {\n\t\treturn String(value);\n\t}\n}\n\nfunction buildRequestOptions(model: Model<\"mistral-conversations\">, options?: MistralOptions) {\n\tconst requestOptions: {\n\t\tsignal?: AbortSignal;\n\t\tretries: { strategy: \"none\" };\n\t\theaders?: Record<string, string>;\n\t} = {\n\t\tretries: { strategy: \"none\" },\n\t};\n\tif (options?.signal) requestOptions.signal = options.signal;\n\n\tconst headers: Record<string, string> = {};\n\tif (model.headers) Object.assign(headers, model.headers);\n\tif (options?.headers) Object.assign(headers, options.headers);\n\n\t// Mistral infrastructure uses `x-affinity` for KV-cache reuse (prefix caching).\n\t// Respect explicit caller-provided header values.\n\tif (options?.sessionId && !headers[\"x-affinity\"]) {\n\t\theaders[\"x-affinity\"] = options.sessionId;\n\t}\n\n\tif (Object.keys(headers).length > 0) {\n\t\trequestOptions.headers = headers;\n\t}\n\n\treturn requestOptions;\n}\n\nfunction buildChatPayload(\n\tmodel: Model<\"mistral-conversations\">,\n\tcontext: Context,\n\tmessages: Message[],\n\toptions?: MistralOptions,\n): ChatCompletionStreamRequest {\n\tconst payload: ChatCompletionStreamRequest = {\n\t\tmodel: model.id,\n\t\tstream: true,\n\t\tmessages: toChatMessages(messages, model.input.includes(\"image\")),\n\t};\n\n\tif (context.tools?.length) payload.tools = toFunctionTools(context.tools);\n\tif (options?.temperature !== undefined) payload.temperature = options.temperature;\n\tif (options?.maxTokens !== undefined) payload.maxTokens = options.maxTokens;\n\tif (options?.toolChoice) payload.toolChoice = mapToolChoice(options.toolChoice);\n\tif (options?.promptMode) payload.promptMode = options.promptMode;\n\tif (options?.reasoningEffort) payload.reasoningEffort = options.reasoningEffort;\n\n\tif (context.systemPrompt) {\n\t\tpayload.messages.unshift({\n\t\t\trole: \"system\",\n\t\t\tcontent: sanitizeSurrogates(context.systemPrompt),\n\t\t});\n\t}\n\n\treturn payload;\n}\n\nasync function consumeChatStream(\n\tmodel: Model<\"mistral-conversations\">,\n\toutput: AssistantMessage,\n\tstream: AssistantMessageEventStream,\n\tmistralStream: AsyncIterable<CompletionEvent>,\n): Promise<void> {\n\tlet currentBlock: TextContent | ThinkingContent | null = null;\n\tconst blocks = output.content;\n\tconst blockIndex = () => blocks.length - 1;\n\tconst toolBlocksByKey = new Map<string, number>();\n\n\tconst finishCurrentBlock = (block?: typeof currentBlock) => {\n\t\tif (!block) return;\n\t\tif (block.type === \"text\") {\n\t\t\tstream.push({\n\t\t\t\ttype: \"text_end\",\n\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\tcontent: block.text,\n\t\t\t\tpartial: output,\n\t\t\t});\n\t\t\treturn;\n\t\t}\n\t\tif (block.type === \"thinking\") {\n\t\t\tstream.push({\n\t\t\t\ttype: \"thinking_end\",\n\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\tcontent: block.thinking,\n\t\t\t\tpartial: output,\n\t\t\t});\n\t\t}\n\t};\n\n\tfor await (const event of mistralStream) {\n\t\tconst chunk = event.data;\n\t\t// Mistral's streamed CompletionChunk carries an id field. Keep the first non-empty one,\n\t\t// mirroring how OpenAI-style streaming exposes a stable response identifier per stream.\n\t\toutput.responseId ||= chunk.id;\n\n\t\tif (chunk.usage) {\n\t\t\toutput.usage.input = chunk.usage.promptTokens || 0;\n\t\t\toutput.usage.output = chunk.usage.completionTokens || 0;\n\t\t\toutput.usage.cacheRead = 0;\n\t\t\toutput.usage.cacheWrite = 0;\n\t\t\toutput.usage.totalTokens = chunk.usage.totalTokens || output.usage.input + output.usage.output;\n\t\t\tcalculateCost(model, output.usage);\n\t\t}\n\n\t\tconst choice = chunk.choices[0];\n\t\tif (!choice) continue;\n\n\t\tif (choice.finishReason) {\n\t\t\toutput.stopReason = mapChatStopReason(choice.finishReason);\n\t\t}\n\n\t\tconst delta = choice.delta;\n\t\tif (delta.content !== null && delta.content !== undefined) {\n\t\t\tconst contentItems = typeof delta.content === \"string\" ? [delta.content] : delta.content;\n\t\t\tfor (const item of contentItems) {\n\t\t\t\tif (typeof item === \"string\") {\n\t\t\t\t\tconst textDelta = sanitizeSurrogates(item);\n\t\t\t\t\tif (!currentBlock || currentBlock.type !== \"text\") {\n\t\t\t\t\t\tfinishCurrentBlock(currentBlock);\n\t\t\t\t\t\tcurrentBlock = { type: \"text\", text: \"\" };\n\t\t\t\t\t\toutput.content.push(currentBlock);\n\t\t\t\t\t\tstream.push({ type: \"text_start\", contentIndex: blockIndex(), partial: output });\n\t\t\t\t\t}\n\t\t\t\t\tcurrentBlock.text += textDelta;\n\t\t\t\t\tstream.push({\n\t\t\t\t\t\ttype: \"text_delta\",\n\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\tdelta: textDelta,\n\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t});\n\t\t\t\t\tcontinue;\n\t\t\t\t}\n\n\t\t\t\tif (item.type === \"thinking\") {\n\t\t\t\t\tconst deltaText = item.thinking\n\t\t\t\t\t\t.map((part) => (\"text\" in part ? part.text : \"\"))\n\t\t\t\t\t\t.filter((text) => text.length > 0)\n\t\t\t\t\t\t.join(\"\");\n\t\t\t\t\tconst thinkingDelta = sanitizeSurrogates(deltaText);\n\t\t\t\t\tif (!thinkingDelta) continue;\n\t\t\t\t\tif (!currentBlock || currentBlock.type !== \"thinking\") {\n\t\t\t\t\t\tfinishCurrentBlock(currentBlock);\n\t\t\t\t\t\tcurrentBlock = { type: \"thinking\", thinking: \"\" };\n\t\t\t\t\t\toutput.content.push(currentBlock);\n\t\t\t\t\t\tstream.push({ type: \"thinking_start\", contentIndex: blockIndex(), partial: output });\n\t\t\t\t\t}\n\t\t\t\t\tcurrentBlock.thinking += thinkingDelta;\n\t\t\t\t\tstream.push({\n\t\t\t\t\t\ttype: \"thinking_delta\",\n\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\tdelta: thinkingDelta,\n\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t});\n\t\t\t\t\tcontinue;\n\t\t\t\t}\n\n\t\t\t\tif (item.type === \"text\") {\n\t\t\t\t\tconst textDelta = sanitizeSurrogates(item.text);\n\t\t\t\t\tif (!currentBlock || currentBlock.type !== \"text\") {\n\t\t\t\t\t\tfinishCurrentBlock(currentBlock);\n\t\t\t\t\t\tcurrentBlock = { type: \"text\", text: \"\" };\n\t\t\t\t\t\toutput.content.push(currentBlock);\n\t\t\t\t\t\tstream.push({ type: \"text_start\", contentIndex: blockIndex(), partial: output });\n\t\t\t\t\t}\n\t\t\t\t\tcurrentBlock.text += textDelta;\n\t\t\t\t\tstream.push({\n\t\t\t\t\t\ttype: \"text_delta\",\n\t\t\t\t\t\tcontentIndex: blockIndex(),\n\t\t\t\t\t\tdelta: textDelta,\n\t\t\t\t\t\tpartial: output,\n\t\t\t\t\t});\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\tconst toolCalls = delta.toolCalls || [];\n\t\tfor (const toolCall of toolCalls) {\n\t\t\tif (currentBlock) {\n\t\t\t\tfinishCurrentBlock(currentBlock);\n\t\t\t\tcurrentBlock = null;\n\t\t\t}\n\t\t\tconst callId =\n\t\t\t\ttoolCall.id && toolCall.id !== \"null\"\n\t\t\t\t\t? toolCall.id\n\t\t\t\t\t: deriveMistralToolCallId(`toolcall:${toolCall.index ?? 0}`, 0);\n\t\t\tconst key = `${callId}:${toolCall.index || 0}`;\n\t\t\tconst existingIndex = toolBlocksByKey.get(key);\n\t\t\tlet block: (ToolCall & { partialArgs?: string }) | undefined;\n\n\t\t\tif (existingIndex !== undefined) {\n\t\t\t\tconst existing = output.content[existingIndex];\n\t\t\t\tif (existing?.type === \"toolCall\") {\n\t\t\t\t\tblock = existing as ToolCall & { partialArgs?: string };\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tif (!block) {\n\t\t\t\tblock = {\n\t\t\t\t\ttype: \"toolCall\",\n\t\t\t\t\tid: callId,\n\t\t\t\t\tname: toolCall.function.name,\n\t\t\t\t\targuments: {},\n\t\t\t\t\tpartialArgs: \"\",\n\t\t\t\t};\n\t\t\t\toutput.content.push(block);\n\t\t\t\ttoolBlocksByKey.set(key, output.content.length - 1);\n\t\t\t\tstream.push({ type: \"toolcall_start\", contentIndex: output.content.length - 1, partial: output });\n\t\t\t}\n\n\t\t\tconst argsDelta =\n\t\t\t\ttypeof toolCall.function.arguments === \"string\"\n\t\t\t\t\t? toolCall.function.arguments\n\t\t\t\t\t: JSON.stringify(toolCall.function.arguments || {});\n\t\t\tblock.partialArgs = (block.partialArgs || \"\") + argsDelta;\n\t\t\tblock.arguments = parseStreamingJson<Record<string, unknown>>(block.partialArgs);\n\t\t\tstream.push({\n\t\t\t\ttype: \"toolcall_delta\",\n\t\t\t\tcontentIndex: toolBlocksByKey.get(key)!,\n\t\t\t\tdelta: argsDelta,\n\t\t\t\tpartial: output,\n\t\t\t});\n\t\t}\n\t}\n\n\tfinishCurrentBlock(currentBlock);\n\tfor (const index of toolBlocksByKey.values()) {\n\t\tconst block = output.content[index];\n\t\tif (block.type !== \"toolCall\") continue;\n\t\tconst toolBlock = block as ToolCall & { partialArgs?: string };\n\t\ttoolBlock.arguments = parseStreamingJson<Record<string, unknown>>(toolBlock.partialArgs);\n\t\t// Finalize in-place and strip the scratch buffer so replay only\n\t\t// carries parsed arguments.\n\t\tdelete toolBlock.partialArgs;\n\t\tstream.push({\n\t\t\ttype: \"toolcall_end\",\n\t\t\tcontentIndex: index,\n\t\t\ttoolCall: toolBlock,\n\t\t\tpartial: output,\n\t\t});\n\t}\n}\n\nfunction toFunctionTools(tools: Tool[]): Array<FunctionTool & { type: \"function\" }> {\n\treturn tools.map((tool) => ({\n\t\ttype: \"function\",\n\t\tfunction: {\n\t\t\tname: tool.name,\n\t\t\tdescription: tool.description,\n\t\t\tparameters: stripSymbolKeys(tool.parameters) as Record<string, unknown>,\n\t\t\tstrict: false,\n\t\t},\n\t}));\n}\n\nfunction stripSymbolKeys(value: unknown): unknown {\n\tif (Array.isArray(value)) {\n\t\treturn value.map((item) => stripSymbolKeys(item));\n\t}\n\n\tif (value && typeof value === \"object\") {\n\t\tconst result: Record<string, unknown> = {};\n\t\tfor (const [key, entry] of Object.entries(value)) {\n\t\t\tresult[key] = stripSymbolKeys(entry);\n\t\t}\n\t\treturn result;\n\t}\n\n\treturn value;\n}\n\nfunction toChatMessages(messages: Message[], supportsImages: boolean): ChatCompletionStreamRequestMessage[] {\n\tconst result: ChatCompletionStreamRequestMessage[] = [];\n\n\tfor (const msg of messages) {\n\t\tif (msg.role === \"user\") {\n\t\t\tif (typeof msg.content === \"string\") {\n\t\t\t\tresult.push({ role: \"user\", content: sanitizeSurrogates(msg.content) });\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tconst hadImages = msg.content.some((item) => item.type === \"image\");\n\t\t\tconst content: ContentChunk[] = msg.content\n\t\t\t\t.filter((item) => item.type === \"text\" || supportsImages)\n\t\t\t\t.map((item) => {\n\t\t\t\t\tif (item.type === \"text\") return { type: \"text\", text: sanitizeSurrogates(item.text) };\n\t\t\t\t\treturn { type: \"image_url\", imageUrl: `data:${item.mimeType};base64,${item.data}` };\n\t\t\t\t});\n\t\t\tif (content.length > 0) {\n\t\t\t\tresult.push({ role: \"user\", content });\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tif (hadImages && !supportsImages) {\n\t\t\t\tresult.push({ role: \"user\", content: \"(image omitted: model does not support images)\" });\n\t\t\t}\n\t\t\tcontinue;\n\t\t}\n\n\t\tif (msg.role === \"assistant\") {\n\t\t\tconst contentParts: ContentChunk[] = [];\n\t\t\tconst toolCalls: Array<{ id: string; type: \"function\"; function: { name: string; arguments: string } }> = [];\n\n\t\t\tfor (const block of msg.content) {\n\t\t\t\tif (block.type === \"text\") {\n\t\t\t\t\tif (block.text.trim().length > 0) {\n\t\t\t\t\t\tcontentParts.push({ type: \"text\", text: sanitizeSurrogates(block.text) });\n\t\t\t\t\t}\n\t\t\t\t\tcontinue;\n\t\t\t\t}\n\t\t\t\tif (block.type === \"thinking\") {\n\t\t\t\t\tif (block.thinking.trim().length > 0) {\n\t\t\t\t\t\tcontentParts.push({\n\t\t\t\t\t\t\ttype: \"thinking\",\n\t\t\t\t\t\t\tthinking: [{ type: \"text\", text: sanitizeSurrogates(block.thinking) }],\n\t\t\t\t\t\t});\n\t\t\t\t\t}\n\t\t\t\t\tcontinue;\n\t\t\t\t}\n\t\t\t\ttoolCalls.push({\n\t\t\t\t\tid: block.id,\n\t\t\t\t\ttype: \"function\",\n\t\t\t\t\tfunction: { name: block.name, arguments: JSON.stringify(block.arguments || {}) },\n\t\t\t\t});\n\t\t\t}\n\n\t\t\tconst assistantMessage: ChatCompletionStreamRequestMessage = { role: \"assistant\" };\n\t\t\tif (contentParts.length > 0) assistantMessage.content = contentParts;\n\t\t\tif (toolCalls.length > 0) assistantMessage.toolCalls = toolCalls;\n\t\t\tif (contentParts.length > 0 || toolCalls.length > 0) result.push(assistantMessage);\n\t\t\tcontinue;\n\t\t}\n\n\t\tconst toolContent: ContentChunk[] = [];\n\t\tconst textResult = msg.content\n\t\t\t.filter((part) => part.type === \"text\")\n\t\t\t.map((part) => (part.type === \"text\" ? sanitizeSurrogates(part.text) : \"\"))\n\t\t\t.join(\"\\n\");\n\t\tconst hasImages = msg.content.some((part) => part.type === \"image\");\n\t\tconst toolText = buildToolResultText(textResult, hasImages, supportsImages, msg.isError);\n\t\ttoolContent.push({ type: \"text\", text: toolText });\n\t\tfor (const part of msg.content) {\n\t\t\tif (!supportsImages) continue;\n\t\t\tif (part.type !== \"image\") continue;\n\t\t\ttoolContent.push({\n\t\t\t\ttype: \"image_url\",\n\t\t\t\timageUrl: `data:${part.mimeType};base64,${part.data}`,\n\t\t\t});\n\t\t}\n\t\tresult.push({\n\t\t\trole: \"tool\",\n\t\t\ttoolCallId: msg.toolCallId,\n\t\t\tname: msg.toolName,\n\t\t\tcontent: toolContent,\n\t\t});\n\t}\n\n\treturn result;\n}\n\nfunction buildToolResultText(text: string, hasImages: boolean, supportsImages: boolean, isError: boolean): string {\n\tconst trimmed = text.trim();\n\tconst errorPrefix = isError ? \"[tool error] \" : \"\";\n\n\tif (trimmed.length > 0) {\n\t\tconst imageSuffix = hasImages && !supportsImages ? \"\\n[tool image omitted: model does not support images]\" : \"\";\n\t\treturn `${errorPrefix}${trimmed}${imageSuffix}`;\n\t}\n\n\tif (hasImages) {\n\t\tif (supportsImages) {\n\t\t\treturn isError ? \"[tool error] (see attached image)\" : \"(see attached image)\";\n\t\t}\n\t\treturn isError\n\t\t\t? \"[tool error] (image omitted: model does not support images)\"\n\t\t\t: \"(image omitted: model does not support images)\";\n\t}\n\n\treturn isError ? \"[tool error] (no tool output)\" : \"(no tool output)\";\n}\n\nfunction usesReasoningEffort(model: Model<\"mistral-conversations\">): boolean {\n\treturn model.id === \"mistral-small-2603\" || model.id === \"mistral-small-latest\";\n}\n\nfunction usesPromptModeReasoning(model: Model<\"mistral-conversations\">): boolean {\n\treturn model.reasoning && !usesReasoningEffort(model);\n}\n\nfunction mapReasoningEffort(_level: Exclude<SimpleStreamOptions[\"reasoning\"], undefined>): MistralReasoningEffort {\n\treturn \"high\";\n}\n\nfunction mapToolChoice(\n\tchoice: MistralOptions[\"toolChoice\"],\n): \"auto\" | \"none\" | \"any\" | \"required\" | { type: \"function\"; function: { name: string } } | undefined {\n\tif (!choice) return undefined;\n\tif (choice === \"auto\" || choice === \"none\" || choice === \"any\" || choice === \"required\") {\n\t\treturn choice as any;\n\t}\n\treturn {\n\t\ttype: \"function\",\n\t\tfunction: { name: choice.function.name },\n\t};\n}\n\nfunction mapChatStopReason(reason: string | null): StopReason {\n\tif (reason === null) return \"stop\";\n\tswitch (reason) {\n\t\tcase \"stop\":\n\t\t\treturn \"stop\";\n\t\tcase \"length\":\n\t\tcase \"model_length\":\n\t\t\treturn \"length\";\n\t\tcase \"tool_calls\":\n\t\t\treturn \"toolUse\";\n\t\tcase \"error\":\n\t\t\treturn \"error\";\n\t\tdefault:\n\t\t\treturn \"stop\";\n\t}\n}\n"]}
|
|
@@ -46,6 +46,10 @@ export const streamMistral = (model, context, options) => {
|
|
|
46
46
|
stream.end();
|
|
47
47
|
}
|
|
48
48
|
catch (error) {
|
|
49
|
+
for (const block of output.content) {
|
|
50
|
+
// partialArgs is only a streaming scratch buffer; never persist it.
|
|
51
|
+
delete block.partialArgs;
|
|
52
|
+
}
|
|
49
53
|
output.stopReason = options?.signal?.aborted ? "aborted" : "error";
|
|
50
54
|
output.errorMessage = formatMistralError(error);
|
|
51
55
|
stream.push({ type: "error", reason: output.stopReason, error: output });
|
|
@@ -64,9 +68,11 @@ export const streamSimpleMistral = (model, context, options) => {
|
|
|
64
68
|
}
|
|
65
69
|
const base = buildBaseOptions(model, options, apiKey);
|
|
66
70
|
const reasoning = clampReasoning(options?.reasoning);
|
|
71
|
+
const shouldUseReasoning = model.reasoning && reasoning !== undefined;
|
|
67
72
|
return streamMistral(model, context, {
|
|
68
73
|
...base,
|
|
69
|
-
promptMode:
|
|
74
|
+
promptMode: shouldUseReasoning && usesPromptModeReasoning(model) ? "reasoning" : undefined,
|
|
75
|
+
reasoningEffort: shouldUseReasoning && usesReasoningEffort(model) ? mapReasoningEffort(reasoning) : undefined,
|
|
70
76
|
});
|
|
71
77
|
};
|
|
72
78
|
function createOutput(model) {
|
|
@@ -147,10 +153,11 @@ function safeJsonStringify(value) {
|
|
|
147
153
|
}
|
|
148
154
|
}
|
|
149
155
|
function buildRequestOptions(model, options) {
|
|
150
|
-
const requestOptions = {
|
|
156
|
+
const requestOptions = {
|
|
157
|
+
retries: { strategy: "none" },
|
|
158
|
+
};
|
|
151
159
|
if (options?.signal)
|
|
152
160
|
requestOptions.signal = options.signal;
|
|
153
|
-
requestOptions.retries = { strategy: "none" };
|
|
154
161
|
const headers = {};
|
|
155
162
|
if (model.headers)
|
|
156
163
|
Object.assign(headers, model.headers);
|
|
@@ -182,6 +189,8 @@ function buildChatPayload(model, context, messages, options) {
|
|
|
182
189
|
payload.toolChoice = mapToolChoice(options.toolChoice);
|
|
183
190
|
if (options?.promptMode)
|
|
184
191
|
payload.promptMode = options.promptMode;
|
|
192
|
+
if (options?.reasoningEffort)
|
|
193
|
+
payload.reasoningEffort = options.reasoningEffort;
|
|
185
194
|
if (context.systemPrompt) {
|
|
186
195
|
payload.messages.unshift({
|
|
187
196
|
role: "system",
|
|
@@ -347,6 +356,8 @@ async function consumeChatStream(model, output, stream, mistralStream) {
|
|
|
347
356
|
continue;
|
|
348
357
|
const toolBlock = block;
|
|
349
358
|
toolBlock.arguments = parseStreamingJson(toolBlock.partialArgs);
|
|
359
|
+
// Finalize in-place and strip the scratch buffer so replay only
|
|
360
|
+
// carries parsed arguments.
|
|
350
361
|
delete toolBlock.partialArgs;
|
|
351
362
|
stream.push({
|
|
352
363
|
type: "toolcall_end",
|
|
@@ -362,11 +373,24 @@ function toFunctionTools(tools) {
|
|
|
362
373
|
function: {
|
|
363
374
|
name: tool.name,
|
|
364
375
|
description: tool.description,
|
|
365
|
-
parameters: tool.parameters,
|
|
376
|
+
parameters: stripSymbolKeys(tool.parameters),
|
|
366
377
|
strict: false,
|
|
367
378
|
},
|
|
368
379
|
}));
|
|
369
380
|
}
|
|
381
|
+
function stripSymbolKeys(value) {
|
|
382
|
+
if (Array.isArray(value)) {
|
|
383
|
+
return value.map((item) => stripSymbolKeys(item));
|
|
384
|
+
}
|
|
385
|
+
if (value && typeof value === "object") {
|
|
386
|
+
const result = {};
|
|
387
|
+
for (const [key, entry] of Object.entries(value)) {
|
|
388
|
+
result[key] = stripSymbolKeys(entry);
|
|
389
|
+
}
|
|
390
|
+
return result;
|
|
391
|
+
}
|
|
392
|
+
return value;
|
|
393
|
+
}
|
|
370
394
|
function toChatMessages(messages, supportsImages) {
|
|
371
395
|
const result = [];
|
|
372
396
|
for (const msg of messages) {
|
|
@@ -470,6 +494,15 @@ function buildToolResultText(text, hasImages, supportsImages, isError) {
|
|
|
470
494
|
}
|
|
471
495
|
return isError ? "[tool error] (no tool output)" : "(no tool output)";
|
|
472
496
|
}
|
|
497
|
+
function usesReasoningEffort(model) {
|
|
498
|
+
return model.id === "mistral-small-2603" || model.id === "mistral-small-latest";
|
|
499
|
+
}
|
|
500
|
+
function usesPromptModeReasoning(model) {
|
|
501
|
+
return model.reasoning && !usesReasoningEffort(model);
|
|
502
|
+
}
|
|
503
|
+
function mapReasoningEffort(_level) {
|
|
504
|
+
return "high";
|
|
505
|
+
}
|
|
473
506
|
function mapToolChoice(choice) {
|
|
474
507
|
if (!choice)
|
|
475
508
|
return undefined;
|