{"version":3,"file":"google-vertex.d.ts","sourceRoot":"","sources":["../../src/providers/google-vertex.ts"],"names":[],"mappings":"AAUA,OAAO,KAAK,EAMX,mBAAmB,EACnB,cAAc,EACd,aAAa,EAKb,MAAM,aAAa,CAAC;AAGrB,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,oBAAoB,CAAC;AAW9D,MAAM,WAAW,mBAAoB,SAAQ,aAAa;IACzD,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;IACF,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,QAAQ,CAAC,EAAE,MAAM,CAAC;CAClB;AAgBD,eAAO,MAAM,kBAAkB,EAAE,cAAc,CAAC,eAAe,EAAE,mBAAmB,CAsOnF,CAAC;AAEF,eAAO,MAAM,wBAAwB,EAAE,cAAc,CAAC,eAAe,EAAE,mBAAmB,CAkCzF,CAAC","sourcesContent":["import {\n\ttype GenerateContentConfig,\n\ttype GenerateContentParameters,\n\tGoogleGenAI,\n\ttype HttpOptions,\n\tResourceScope,\n\ttype ThinkingConfig,\n\tThinkingLevel,\n} from \"@google/genai\";\nimport { calculateCost, clampThinkingLevel } from \"../models.js\";\nimport type {\n\tApi,\n\tAssistantMessage,\n\tContext,\n\tModel,\n\tThinkingLevel as PiThinkingLevel,\n\tSimpleStreamOptions,\n\tStreamFunction,\n\tStreamOptions,\n\tTextContent,\n\tThinkingBudgets,\n\tThinkingContent,\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-shared.js\";\nimport {\n\tconvertMessages,\n\tconvertTools,\n\tisThinkingPart,\n\tmapStopReason,\n\tmapToolChoice,\n\tretainThoughtSignature,\n} from \"./google-shared.js\";\nimport { buildBaseOptions } from \"./simple-options.js\";\n\nexport interface GoogleVertexOptions 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\tproject?: string;\n\tlocation?: string;\n}\n\nconst API_VERSION = \"v1\";\nconst GCP_VERTEX_CREDENTIALS_MARKER = \"gcp-vertex-credentials\";\n\nconst THINKING_LEVEL_MAP: Record<GoogleThinkingLevel, ThinkingLevel> = {\n\tTHINKING_LEVEL_UNSPECIFIED: ThinkingLevel.THINKING_LEVEL_UNSPECIFIED,\n\tMINIMAL: ThinkingLevel.MINIMAL,\n\tLOW: ThinkingLevel.LOW,\n\tMEDIUM: ThinkingLevel.MEDIUM,\n\tHIGH: ThinkingLevel.HIGH,\n};\n\n// Counter for generating unique tool call IDs\nlet toolCallCounter = 0;\n\nexport const streamGoogleVertex: StreamFunction<\"google-vertex\", GoogleVertexOptions> = (\n\tmodel: Model<\"google-vertex\">,\n\tcontext: Context,\n\toptions?: GoogleVertexOptions,\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-vertex\" 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 = resolveApiKey(options);\n\t\t\t// Create the client using either a Vertex API key, if provided, or ADC with project and location\n\t\t\tconst client = apiKey\n\t\t\t\t? createClientWithApiKey(model, apiKey, options?.headers)\n\t\t\t\t: createClient(model, resolveProject(options), resolveLocation(options), 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// Vertex uses the same @google/genai GenerateContentResponse type as Gemini.\n\t\t\t\t// responseId is documented there as an output-only identifier for each response.\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\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 streamSimpleGoogleVertex: StreamFunction<\"google-vertex\", SimpleStreamOptions> = (\n\tmodel: Model<\"google-vertex\">,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream => {\n\tconst base = buildBaseOptions(model, options, undefined);\n\tif (!options?.reasoning) {\n\t\treturn streamGoogleVertex(model, context, {\n\t\t\t...base,\n\t\t\tthinking: { enabled: false },\n\t\t} satisfies GoogleVertexOptions);\n\t}\n\n\tconst clampedReasoning = clampThinkingLevel(model, options.reasoning);\n\tconst effort = (clampedReasoning === \"off\" ? \"high\" : clampedReasoning) as ClampedThinkingLevel;\n\tconst geminiModel = model as unknown as Model<\"google-generative-ai\">;\n\n\tif (isGemini3ProModel(geminiModel) || isGemini3FlashModel(geminiModel)) {\n\t\treturn streamGoogleVertex(model, context, {\n\t\t\t...base,\n\t\t\tthinking: {\n\t\t\t\tenabled: true,\n\t\t\t\tlevel: getGemini3ThinkingLevel(effort, geminiModel),\n\t\t\t},\n\t\t} satisfies GoogleVertexOptions);\n\t}\n\n\treturn streamGoogleVertex(model, context, {\n\t\t...base,\n\t\tthinking: {\n\t\t\tenabled: true,\n\t\t\tbudgetTokens: getGoogleBudget(geminiModel, effort, options.thinkingBudgets),\n\t\t},\n\t} satisfies GoogleVertexOptions);\n};\n\nfunction createClient(\n\tmodel: Model<\"google-vertex\">,\n\tproject: string,\n\tlocation: string,\n\toptionsHeaders?: Record<string, string>,\n): GoogleGenAI {\n\treturn new GoogleGenAI({\n\t\tvertexai: true,\n\t\tproject,\n\t\tlocation,\n\t\tapiVersion: API_VERSION,\n\t\thttpOptions: buildHttpOptions(model, optionsHeaders),\n\t});\n}\n\nfunction createClientWithApiKey(\n\tmodel: Model<\"google-vertex\">,\n\tapiKey: string,\n\toptionsHeaders?: Record<string, string>,\n): GoogleGenAI {\n\treturn new GoogleGenAI({\n\t\tvertexai: true,\n\t\tapiKey,\n\t\tapiVersion: API_VERSION,\n\t\thttpOptions: buildHttpOptions(model, optionsHeaders),\n\t});\n}\n\nfunction buildHttpOptions(\n\tmodel: Model<\"google-vertex\">,\n\toptionsHeaders?: Record<string, string>,\n): HttpOptions | undefined {\n\tconst httpOptions: HttpOptions = {};\n\tconst baseUrl = resolveCustomBaseUrl(model.baseUrl);\n\tif (baseUrl) {\n\t\thttpOptions.baseUrl = baseUrl;\n\t\thttpOptions.baseUrlResourceScope = ResourceScope.COLLECTION;\n\t\tif (baseUrlIncludesApiVersion(baseUrl)) {\n\t\t\thttpOptions.apiVersion = \"\";\n\t\t}\n\t}\n\n\tif (model.headers || optionsHeaders) {\n\t\thttpOptions.headers = { ...model.headers, ...optionsHeaders };\n\t}\n\n\treturn Object.keys(httpOptions).length > 0 ? httpOptions : undefined;\n}\n\nfunction resolveCustomBaseUrl(baseUrl: string): string | undefined {\n\tconst trimmed = baseUrl.trim();\n\tif (!trimmed || trimmed.includes(\"{location}\")) {\n\t\treturn undefined;\n\t}\n\treturn trimmed;\n}\n\nfunction baseUrlIncludesApiVersion(baseUrl: string): boolean {\n\ttry {\n\t\tconst url = new URL(baseUrl);\n\t\treturn url.pathname.split(\"/\").some((part) => /^v\\d+(?:beta\\d*)?$/.test(part));\n\t} catch {\n\t\treturn /(?:^|\\/)v\\d+(?:beta\\d*)?(?:\\/|$)/.test(baseUrl);\n\t}\n}\n\nfunction resolveApiKey(options?: GoogleVertexOptions): string | undefined {\n\tconst apiKey = options?.apiKey?.trim() || process.env.GOOGLE_CLOUD_API_KEY?.trim();\n\tif (!apiKey || apiKey === GCP_VERTEX_CREDENTIALS_MARKER || isPlaceholderApiKey(apiKey)) {\n\t\treturn undefined;\n\t}\n\treturn apiKey;\n}\n\nfunction isPlaceholderApiKey(apiKey: string): boolean {\n\treturn /^<[^>]+>$/.test(apiKey);\n}\n\nfunction resolveProject(options?: GoogleVertexOptions): string {\n\tconst project = options?.project || process.env.GOOGLE_CLOUD_PROJECT || process.env.GCLOUD_PROJECT;\n\tif (!project) {\n\t\tthrow new Error(\n\t\t\t\"Vertex AI requires a project ID. Set GOOGLE_CLOUD_PROJECT/GCLOUD_PROJECT or pass project in options.\",\n\t\t);\n\t}\n\treturn project;\n}\n\nfunction resolveLocation(options?: GoogleVertexOptions): string {\n\tconst location = options?.location || process.env.GOOGLE_CLOUD_LOCATION;\n\tif (!location) {\n\t\tthrow new Error(\"Vertex AI requires a location. Set GOOGLE_CLOUD_LOCATION or pass location in options.\");\n\t}\n\treturn location;\n}\n\nfunction buildParams(\n\tmodel: Model<\"google-vertex\">,\n\tcontext: Context,\n\toptions: GoogleVertexOptions = {},\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\tthinkingConfig.thinkingLevel = THINKING_LEVEL_MAP[options.thinking.level];\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<PiThinkingLevel, \"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 getDisabledThinkingConfig(model: Model<\"google-vertex\">): 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\tconst geminiModel = model as unknown as Model<\"google-generative-ai\">;\n\tif (isGemini3ProModel(geminiModel)) {\n\t\treturn { thinkingLevel: ThinkingLevel.LOW };\n\t}\n\tif (isGemini3FlashModel(geminiModel)) {\n\t\treturn { thinkingLevel: ThinkingLevel.MINIMAL };\n\t}\n\n\t// Gemini 2.x supports disabling via thinkingBudget = 0.\n\treturn { thinkingBudget: 0 };\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"]}