@midscene/core 1.0.1-beta-20251202152706.0 → 1.0.1-beta-20251203073716.0

package/dist/es/ai-model/service-caller/index.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"ai-model/service-caller/index.mjs","sources":["../../../../src/ai-model/service-caller/index.ts"],"sourcesContent":["import { AIResponseFormat, type AIUsageInfo } from '@/types';\nimport type { CodeGenerationChunk, StreamingCallback } from '@/types';\nimport {\n  type IModelConfig,\n  MIDSCENE_LANGFUSE_DEBUG,\n  MIDSCENE_LANGSMITH_DEBUG,\n  MIDSCENE_MODEL_MAX_TOKENS,\n  OPENAI_MAX_TOKENS,\n  type TVlModeTypes,\n  type UITarsModelVersion,\n  globalConfigManager,\n} from '@midscene/shared/env';\n\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert, ifInBrowser } from '@midscene/shared/utils';\nimport { HttpsProxyAgent } from 'https-proxy-agent';\nimport { jsonrepair } from 'jsonrepair';\nimport OpenAI from 'openai';\nimport type { ChatCompletionMessageParam } from 'openai/resources/index';\nimport type { Stream } from 'openai/streaming';\nimport { SocksProxyAgent } from 'socks-proxy-agent';\nimport { AIActionType, type AIArgs } from '../../common';\nimport { assertSchema } from '../prompt/assertion';\nimport { planSchema } from '../prompt/llm-planning';\n\nasync function createChatClient({\n  AIActionTypeValue,\n  modelConfig,\n}: {\n  AIActionTypeValue: AIActionType;\n  modelConfig: IModelConfig;\n}): Promise<{\n  completion: OpenAI.Chat.Completions;\n  modelName: string;\n  modelDescription: string;\n  uiTarsVersion?: UITarsModelVersion;\n  vlMode: TVlModeTypes | undefined;\n}> {\n  const {\n    socksProxy,\n    httpProxy,\n    modelName,\n    openaiBaseURL,\n    openaiApiKey,\n    openaiExtraConfig,\n    modelDescription,\n    uiTarsModelVersion: uiTarsVersion,\n    vlMode,\n    createOpenAIClient,\n  } = modelConfig;\n\n  let proxyAgent = undefined;\n  const debugProxy = getDebug('ai:call:proxy');\n  if (httpProxy) {\n    debugProxy('using http proxy', httpProxy);\n    proxyAgent = new HttpsProxyAgent(httpProxy);\n  } else if (socksProxy) {\n    debugProxy('using socks proxy', socksProxy);\n    proxyAgent = new SocksProxyAgent(socksProxy);\n  }\n\n  const openAIOptions = {\n    baseURL: openaiBaseURL,\n    apiKey: openaiApiKey,\n    ...(proxyAgent ? { httpAgent: proxyAgent as any } : {}),\n    ...openaiExtraConfig,\n    dangerouslyAllowBrowser: true,\n  };\n\n  const baseOpenAI = new OpenAI(openAIOptions);\n\n  let openai: OpenAI = baseOpenAI;\n\n  // LangSmith wrapper\n  if (\n    openai &&\n    globalConfigManager.getEnvConfigInBoolean(MIDSCENE_LANGSMITH_DEBUG)\n  ) {\n    if (ifInBrowser) {\n      throw new Error('langsmith is not supported in browser');\n    }\n    console.log('DEBUGGING MODE: langsmith wrapper enabled');\n    // Use variable to prevent static analysis by bundlers\n    const langsmithModule = 'langsmith/wrappers';\n    const { wrapOpenAI } = await import(langsmithModule);\n    openai = wrapOpenAI(openai);\n  }\n\n  // Langfuse wrapper\n  if (\n    openai &&\n    globalConfigManager.getEnvConfigInBoolean(MIDSCENE_LANGFUSE_DEBUG)\n  ) {\n    if (ifInBrowser) {\n      throw new Error('langfuse is not supported in browser');\n    }\n    console.log('DEBUGGING MODE: langfuse wrapper enabled');\n    // Use variable to prevent static analysis by bundlers\n    const langfuseModule = 'langfuse';\n    const { observeOpenAI } = await import(langfuseModule);\n    openai = observeOpenAI(openai);\n  }\n\n  if (createOpenAIClient) {\n    const wrappedClient = await createOpenAIClient(baseOpenAI, openAIOptions);\n\n    if (wrappedClient) {\n      openai = wrappedClient as OpenAI;\n    }\n  }\n\n  return {\n    completion: openai.chat.completions,\n    modelName,\n    modelDescription,\n    uiTarsVersion,\n    vlMode,\n  };\n}\n\nexport async function callAI(\n  messages: ChatCompletionMessageParam[],\n  AIActionTypeValue: AIActionType,\n  modelConfig: IModelConfig,\n  options?: {\n    stream?: boolean;\n    onChunk?: StreamingCallback;\n  },\n): Promise<{ content: string; usage?: AIUsageInfo; isStreamed: boolean }> {\n  const { completion, modelName, modelDescription, uiTarsVersion, vlMode } =\n    await createChatClient({\n      AIActionTypeValue,\n      modelConfig,\n    });\n\n  const responseFormat = getResponseFormat(modelName, AIActionTypeValue);\n\n  const maxTokens =\n    globalConfigManager.getEnvConfigValue(MIDSCENE_MODEL_MAX_TOKENS) ??\n    globalConfigManager.getEnvConfigValue(OPENAI_MAX_TOKENS);\n  const debugCall = getDebug('ai:call');\n  const debugProfileStats = getDebug('ai:profile:stats');\n  const debugProfileDetail = getDebug('ai:profile:detail');\n\n  const startTime = Date.now();\n\n  const isStreaming = options?.stream && options?.onChunk;\n  let content: string | undefined;\n  let accumulated = '';\n  let usage: OpenAI.CompletionUsage | undefined;\n  let timeCost: number | undefined;\n\n  const buildUsageInfo = (usageData?: OpenAI.CompletionUsage) => {\n    if (!usageData) return undefined;\n\n    const cachedInputTokens = (\n      usageData as { prompt_tokens_details?: { cached_tokens?: number } }\n    )?.prompt_tokens_details?.cached_tokens;\n\n    return {\n      prompt_tokens: usageData.prompt_tokens ?? 0,\n      completion_tokens: usageData.completion_tokens ?? 0,\n      total_tokens: usageData.total_tokens ?? 0,\n      cached_input: cachedInputTokens ?? 0,\n      time_cost: timeCost ?? 0,\n      model_name: modelName,\n      model_description: modelDescription,\n      intent: modelConfig.intent,\n    } satisfies AIUsageInfo;\n  };\n\n  const commonConfig = {\n    temperature: vlMode === 'vlm-ui-tars' ? 0.0 : undefined,\n    stream: !!isStreaming,\n    max_tokens: typeof maxTokens === 'number' ? maxTokens : undefined,\n    ...(vlMode === 'qwen2.5-vl' // qwen vl v2 specific config\n      ? {\n          vl_high_resolution_images: true,\n        }\n      : {}),\n  };\n\n  try {\n    debugCall(\n      `sending ${isStreaming ? 'streaming ' : ''}request to ${modelName}`,\n    );\n\n    if (isStreaming) {\n      const stream = (await completion.create(\n        {\n          model: modelName,\n          messages,\n          response_format: responseFormat,\n          ...commonConfig,\n        },\n        {\n          stream: true,\n        },\n      )) as Stream<OpenAI.Chat.Completions.ChatCompletionChunk> & {\n        _request_id?: string | null;\n      };\n\n      for await (const chunk of stream) {\n        const content = chunk.choices?.[0]?.delta?.content || '';\n        const reasoning_content =\n          (chunk.choices?.[0]?.delta as any)?.reasoning_content || '';\n\n        // Check for usage info in any chunk (OpenAI provides usage in separate chunks)\n        if (chunk.usage) {\n          usage = chunk.usage;\n        }\n\n        if (content || reasoning_content) {\n          accumulated += content;\n          const chunkData: CodeGenerationChunk = {\n            content,\n            reasoning_content,\n            accumulated,\n            isComplete: false,\n            usage: undefined,\n          };\n          options.onChunk!(chunkData);\n        }\n\n        // Check if stream is complete\n        if (chunk.choices?.[0]?.finish_reason) {\n          timeCost = Date.now() - startTime;\n\n          // If usage is not available from the stream, provide a basic usage info\n          if (!usage) {\n            // Estimate token counts based on content length (rough approximation)\n            const estimatedTokens = Math.max(\n              1,\n              Math.floor(accumulated.length / 4),\n            );\n            usage = {\n              prompt_tokens: estimatedTokens,\n              completion_tokens: estimatedTokens,\n              total_tokens: estimatedTokens * 2,\n            };\n          }\n\n          // Send final chunk\n          const finalChunk: CodeGenerationChunk = {\n            content: '',\n            accumulated,\n            reasoning_content: '',\n            isComplete: true,\n            usage: buildUsageInfo(usage),\n          };\n          options.onChunk!(finalChunk);\n          break;\n        }\n      }\n      content = accumulated;\n      debugProfileStats(\n        `streaming model, ${modelName}, mode, ${vlMode || 'default'}, cost-ms, ${timeCost}`,\n      );\n    } else {\n      const result = await completion.create({\n        model: modelName,\n        messages,\n        response_format: responseFormat,\n        ...commonConfig,\n      } as any);\n      timeCost = Date.now() - startTime;\n\n      debugProfileStats(\n        `model, ${modelName}, mode, ${vlMode || 'default'}, ui-tars-version, ${uiTarsVersion}, prompt-tokens, ${result.usage?.prompt_tokens || ''}, completion-tokens, ${result.usage?.completion_tokens || ''}, total-tokens, ${result.usage?.total_tokens || ''}, cost-ms, ${timeCost}, requestId, ${result._request_id || ''}`,\n      );\n\n      debugProfileDetail(`model usage detail: ${JSON.stringify(result.usage)}`);\n\n      assert(\n        result.choices,\n        `invalid response from LLM service: ${JSON.stringify(result)}`,\n      );\n      content = result.choices[0].message.content!;\n      usage = result.usage;\n    }\n\n    debugCall(`response: ${content}`);\n    assert(content, 'empty content');\n\n    // Ensure we always have usage info for streaming responses\n    if (isStreaming && !usage) {\n      // Estimate token counts based on content length (rough approximation)\n      const estimatedTokens = Math.max(\n        1,\n        Math.floor((content || '').length / 4),\n      );\n      usage = {\n        prompt_tokens: estimatedTokens,\n        completion_tokens: estimatedTokens,\n        total_tokens: estimatedTokens * 2,\n      } as OpenAI.CompletionUsage;\n    }\n\n    return {\n      content: content || '',\n      usage: buildUsageInfo(usage),\n      isStreamed: !!isStreaming,\n    };\n  } catch (e: any) {\n    console.error(' call AI error', e);\n    const newError = new Error(\n      `failed to call ${isStreaming ? 'streaming ' : ''}AI model service: ${e.message}. Trouble shooting: https://midscenejs.com/model-provider.html`,\n      {\n        cause: e,\n      },\n    );\n    throw newError;\n  }\n}\n\nexport const getResponseFormat = (\n  modelName: string,\n  AIActionTypeValue: AIActionType,\n):\n  | OpenAI.ChatCompletionCreateParams['response_format']\n  | OpenAI.ResponseFormatJSONObject => {\n  let responseFormat:\n    | OpenAI.ChatCompletionCreateParams['response_format']\n    | OpenAI.ResponseFormatJSONObject\n    | undefined;\n\n  if (modelName.includes('gpt-4')) {\n    switch (AIActionTypeValue) {\n      case AIActionType.ASSERT:\n        responseFormat = assertSchema;\n        break;\n      case AIActionType.PLAN:\n        responseFormat = planSchema;\n        break;\n      case AIActionType.EXTRACT_DATA:\n      case AIActionType.DESCRIBE_ELEMENT:\n        responseFormat = { type: AIResponseFormat.JSON };\n        break;\n      case AIActionType.TEXT:\n        // No response format for plain text - return as-is\n        responseFormat = undefined;\n        break;\n    }\n  }\n\n  // gpt-4o-2024-05-13 only supports json_object response format\n  // Skip for plain text to allow string output\n  if (\n    modelName === 'gpt-4o-2024-05-13' &&\n    AIActionTypeValue !== AIActionType.TEXT\n  ) {\n    responseFormat = { type: AIResponseFormat.JSON };\n  }\n\n  return responseFormat;\n};\n\nexport async function callAIWithObjectResponse<T>(\n  messages: ChatCompletionMessageParam[],\n  AIActionTypeValue: AIActionType,\n  modelConfig: IModelConfig,\n): Promise<{ content: T; contentString: string; usage?: AIUsageInfo }> {\n  const response = await callAI(messages, AIActionTypeValue, modelConfig);\n  assert(response, 'empty response');\n  const vlMode = modelConfig.vlMode;\n  const jsonContent = safeParseJson(response.content, vlMode);\n  return {\n    content: jsonContent,\n    contentString: response.content,\n    usage: response.usage,\n  };\n}\n\nexport async function callAIWithStringResponse(\n  msgs: AIArgs,\n  AIActionTypeValue: AIActionType,\n  modelConfig: IModelConfig,\n): Promise<{ content: string; usage?: AIUsageInfo }> {\n  const { content, usage } = await callAI(msgs, AIActionTypeValue, modelConfig);\n  return { content, usage };\n}\n\nexport function extractJSONFromCodeBlock(response: string) {\n  try {\n    // First, try to match a JSON object directly in the response\n    const jsonMatch = response.match(/^\\s*(\\{[\\s\\S]*\\})\\s*$/);\n    if (jsonMatch) {\n      return jsonMatch[1];\n    }\n\n    // If no direct JSON object is found, try to extract JSON from a code block\n    const codeBlockMatch = response.match(\n      /```(?:json)?\\s*(\\{[\\s\\S]*?\\})\\s*```/,\n    );\n    if (codeBlockMatch) {\n      return codeBlockMatch[1];\n    }\n\n    // If no code block is found, try to find a JSON-like structure in the text\n    const jsonLikeMatch = response.match(/\\{[\\s\\S]*\\}/);\n    if (jsonLikeMatch) {\n      return jsonLikeMatch[0];\n    }\n  } catch {}\n  // If no JSON-like structure is found, return the original response\n  return response;\n}\n\nexport function preprocessDoubaoBboxJson(input: string) {\n  if (input.includes('bbox')) {\n    // when its values like 940 445 969 490, replace all /\\d+\\s+\\d+/g with /$1,$2/g\n    while (/\\d+\\s+\\d+/.test(input)) {\n      input = input.replace(/(\\d+)\\s+(\\d+)/g, '$1,$2');\n    }\n  }\n  return input;\n}\n\n/**\n * Normalize a parsed JSON object by trimming whitespace from:\n * 1. All object keys (e.g., \" prompt \" -> \"prompt\")\n * 2. All string values (e.g., \" Tap \" -> \"Tap\")\n * This handles LLM output that may include leading/trailing spaces.\n */\nfunction normalizeJsonObject(obj: any): any {\n  // Handle null and undefined\n  if (obj === null || obj === undefined) {\n    return obj;\n  }\n\n  // Handle arrays - recursively normalize each element\n  if (Array.isArray(obj)) {\n    return obj.map((item) => normalizeJsonObject(item));\n  }\n\n  // Handle objects\n  if (typeof obj === 'object') {\n    const normalized: any = {};\n\n    for (const [key, value] of Object.entries(obj)) {\n      // Trim the key to remove leading/trailing spaces\n      const trimmedKey = key.trim();\n\n      // Recursively normalize the value\n      let normalizedValue = normalizeJsonObject(value);\n\n      // Trim all string values\n      if (typeof normalizedValue === 'string') {\n        normalizedValue = normalizedValue.trim();\n      }\n\n      normalized[trimmedKey] = normalizedValue;\n    }\n\n    return normalized;\n  }\n\n  // Handle primitive strings\n  if (typeof obj === 'string') {\n    return obj.trim();\n  }\n\n  // Return other primitives as-is\n  return obj;\n}\n\nexport function safeParseJson(input: string, vlMode: TVlModeTypes | undefined) {\n  const cleanJsonString = extractJSONFromCodeBlock(input);\n  // match the point\n  if (cleanJsonString?.match(/\\((\\d+),(\\d+)\\)/)) {\n    return cleanJsonString\n      .match(/\\((\\d+),(\\d+)\\)/)\n      ?.slice(1)\n      .map(Number);\n  }\n\n  let parsed: any;\n  try {\n    parsed = JSON.parse(cleanJsonString);\n    return normalizeJsonObject(parsed);\n  } catch {}\n  try {\n    parsed = JSON.parse(jsonrepair(cleanJsonString));\n    return normalizeJsonObject(parsed);\n  } catch (e) {}\n\n  if (vlMode === 'doubao-vision' || vlMode === 'vlm-ui-tars') {\n    const jsonString = preprocessDoubaoBboxJson(cleanJsonString);\n    parsed = JSON.parse(jsonrepair(jsonString));\n    return normalizeJsonObject(parsed);\n  }\n  throw Error(`failed to parse json response: ${input}`);\n}\n"],"names":["createChatClient","AIActionTypeValue","modelConfig","socksProxy","httpProxy","modelName","openaiBaseURL","openaiApiKey","openaiExtraConfig","modelDescription","uiTarsVersion","vlMode","createOpenAIClient","proxyAgent","debugProxy","getDebug","HttpsProxyAgent","SocksProxyAgent","openAIOptions","baseOpenAI","OpenAI","openai","globalConfigManager","MIDSCENE_LANGSMITH_DEBUG","ifInBrowser","Error","console","langsmithModule","wrapOpenAI","MIDSCENE_LANGFUSE_DEBUG","langfuseModule","observeOpenAI","wrappedClient","callAI","messages","options","completion","responseFormat","getResponseFormat","maxTokens","MIDSCENE_MODEL_MAX_TOKENS","OPENAI_MAX_TOKENS","debugCall","debugProfileStats","debugProfileDetail","startTime","Date","isStreaming","content","accumulated","usage","timeCost","buildUsageInfo","usageData","cachedInputTokens","commonConfig","undefined","stream","chunk","reasoning_content","chunkData","estimatedTokens","Math","finalChunk","result","JSON","assert","e","newError","AIActionType","assertSchema","planSchema","AIResponseFormat","callAIWithObjectResponse","response","jsonContent","safeParseJson","callAIWithStringResponse","msgs","extractJSONFromCodeBlock","jsonMatch","codeBlockMatch","jsonLikeMatch","preprocessDoubaoBboxJson","input","normalizeJsonObject","obj","Array","item","normalized","key","value","Object","trimmedKey","normalizedValue","cleanJsonString","Number","parsed","jsonrepair","jsonString"],"mappings":";;;;;;;;;;;AAyBA,eAAeA,iBAAiB,EAC9BC,iBAAiB,EACjBC,WAAW,EAIZ;IAOC,MAAM,EACJC,UAAU,EACVC,SAAS,EACTC,SAAS,EACTC,aAAa,EACbC,YAAY,EACZC,iBAAiB,EACjBC,gBAAgB,EAChB,oBAAoBC,aAAa,EACjCC,MAAM,EACNC,kBAAkB,EACnB,GAAGV;IAEJ,IAAIW;IACJ,MAAMC,aAAaC,SAAS;IAC5B,IAAIX,WAAW;QACbU,WAAW,oBAAoBV;QAC/BS,aAAa,IAAIG,gBAAgBZ;IACnC,OAAO,IAAID,YAAY;QACrBW,WAAW,qBAAqBX;QAChCU,aAAa,IAAII,gBAAgBd;IACnC;IAEA,MAAMe,gBAAgB;QACpB,SAASZ;QACT,QAAQC;QACR,GAAIM,aAAa;YAAE,WAAWA;QAAkB,IAAI,CAAC,CAAC;QACtD,GAAGL,iBAAiB;QACpB,yBAAyB;IAC3B;IAEA,MAAMW,aAAa,IAAIC,SAAOF;IAE9B,IAAIG,SAAiBF;IAGrB,IACEE,UACAC,oBAAoB,qBAAqB,CAACC,2BAC1C;QACA,IAAIC,aACF,MAAM,IAAIC,MAAM;QAElBC,QAAQ,GAAG,CAAC;QAEZ,MAAMC,kBAAkB;QACxB,MAAM,EAAEC,UAAU,EAAE,GAAG,MAAM,MAAM,CAACD;QACpCN,SAASO,WAAWP;IACtB;IAGA,IACEA,UACAC,oBAAoB,qBAAqB,CAACO,0BAC1C;QACA,IAAIL,aACF,MAAM,IAAIC,MAAM;QAElBC,QAAQ,GAAG,CAAC;QAEZ,MAAMI,iBAAiB;QACvB,MAAM,EAAEC,aAAa,EAAE,GAAG,MAAM,MAAM,CAACD;QACvCT,SAASU,cAAcV;IACzB;IAEA,IAAIT,oBAAoB;QACtB,MAAMoB,gBAAgB,MAAMpB,mBAAmBO,YAAYD;QAE3D,IAAIc,eACFX,SAASW;IAEb;IAEA,OAAO;QACL,YAAYX,OAAO,IAAI,CAAC,WAAW;QACnChB;QACAI;QACAC;QACAC;IACF;AACF;AAEO,eAAesB,OACpBC,QAAsC,EACtCjC,iBAA+B,EAC/BC,WAAyB,EACzBiC,OAGC;IAED,MAAM,EAAEC,UAAU,EAAE/B,SAAS,EAAEI,gBAAgB,EAAEC,aAAa,EAAEC,MAAM,EAAE,GACtE,MAAMX,iBAAiB;QACrBC;QACAC;IACF;IAEF,MAAMmC,iBAAiBC,kBAAkBjC,WAAWJ;IAEpD,MAAMsC,YACJjB,oBAAoB,iBAAiB,CAACkB,8BACtClB,oBAAoB,iBAAiB,CAACmB;IACxC,MAAMC,YAAY3B,SAAS;IAC3B,MAAM4B,oBAAoB5B,SAAS;IACnC,MAAM6B,qBAAqB7B,SAAS;IAEpC,MAAM8B,YAAYC,KAAK,GAAG;IAE1B,MAAMC,cAAcZ,SAAS,UAAUA,SAAS;IAChD,IAAIa;IACJ,IAAIC,cAAc;IAClB,IAAIC;IACJ,IAAIC;IAEJ,MAAMC,iBAAiB,CAACC;QACtB,IAAI,CAACA,WAAW;QAEhB,MAAMC,oBACJD,WACC,uBAAuB;QAE1B,OAAO;YACL,eAAeA,UAAU,aAAa,IAAI;YAC1C,mBAAmBA,UAAU,iBAAiB,IAAI;YAClD,cAAcA,UAAU,YAAY,IAAI;YACxC,cAAcC,qBAAqB;YACnC,WAAWH,YAAY;YACvB,YAAY9C;YACZ,mBAAmBI;YACnB,QAAQP,YAAY,MAAM;QAC5B;IACF;IAEA,MAAMqD,eAAe;QACnB,aAAa5C,AAAW,kBAAXA,SAA2B,MAAM6C;QAC9C,QAAQ,CAAC,CAACT;QACV,YAAY,AAAqB,YAArB,OAAOR,YAAyBA,YAAYiB;QACxD,GAAI7C,AAAW,iBAAXA,SACA;YACE,2BAA2B;QAC7B,IACA,CAAC,CAAC;IACR;IAEA,IAAI;QACF+B,UACE,CAAC,QAAQ,EAAEK,cAAc,eAAe,GAAG,WAAW,EAAE1C,WAAW;QAGrE,IAAI0C,aAAa;YACf,MAAMU,SAAU,MAAMrB,WAAW,MAAM,CACrC;gBACE,OAAO/B;gBACP6B;gBACA,iBAAiBG;gBACjB,GAAGkB,YAAY;YACjB,GACA;gBACE,QAAQ;YACV;YAKF,WAAW,MAAMG,SAASD,OAAQ;gBAChC,MAAMT,UAAUU,MAAM,OAAO,EAAE,CAAC,EAAE,EAAE,OAAO,WAAW;gBACtD,MAAMC,oBACHD,MAAM,OAAO,EAAE,CAAC,EAAE,EAAE,OAAe,qBAAqB;gBAG3D,IAAIA,MAAM,KAAK,EACbR,QAAQQ,MAAM,KAAK;gBAGrB,IAAIV,WAAWW,mBAAmB;oBAChCV,eAAeD;oBACf,MAAMY,YAAiC;wBACrCZ;wBACAW;wBACAV;wBACA,YAAY;wBACZ,OAAOO;oBACT;oBACArB,QAAQ,OAAO,CAAEyB;gBACnB;gBAGA,IAAIF,MAAM,OAAO,EAAE,CAAC,EAAE,EAAE,eAAe;oBACrCP,WAAWL,KAAK,GAAG,KAAKD;oBAGxB,IAAI,CAACK,OAAO;wBAEV,MAAMW,kBAAkBC,KAAK,GAAG,CAC9B,GACAA,KAAK,KAAK,CAACb,YAAY,MAAM,GAAG;wBAElCC,QAAQ;4BACN,eAAeW;4BACf,mBAAmBA;4BACnB,cAAcA,AAAkB,IAAlBA;wBAChB;oBACF;oBAGA,MAAME,aAAkC;wBACtC,SAAS;wBACTd;wBACA,mBAAmB;wBACnB,YAAY;wBACZ,OAAOG,eAAeF;oBACxB;oBACAf,QAAQ,OAAO,CAAE4B;oBACjB;gBACF;YACF;YACAf,UAAUC;YACVN,kBACE,CAAC,iBAAiB,EAAEtC,UAAU,QAAQ,EAAEM,UAAU,UAAU,WAAW,EAAEwC,UAAU;QAEvF,OAAO;YACL,MAAMa,SAAS,MAAM5B,WAAW,MAAM,CAAC;gBACrC,OAAO/B;gBACP6B;gBACA,iBAAiBG;gBACjB,GAAGkB,YAAY;YACjB;YACAJ,WAAWL,KAAK,GAAG,KAAKD;YAExBF,kBACE,CAAC,OAAO,EAAEtC,UAAU,QAAQ,EAAEM,UAAU,UAAU,mBAAmB,EAAED,cAAc,iBAAiB,EAAEsD,OAAO,KAAK,EAAE,iBAAiB,GAAG,qBAAqB,EAAEA,OAAO,KAAK,EAAE,qBAAqB,GAAG,gBAAgB,EAAEA,OAAO,KAAK,EAAE,gBAAgB,GAAG,WAAW,EAAEb,SAAS,aAAa,EAAEa,OAAO,WAAW,IAAI,IAAI;YAG3TpB,mBAAmB,CAAC,oBAAoB,EAAEqB,KAAK,SAAS,CAACD,OAAO,KAAK,GAAG;YAExEE,OACEF,OAAO,OAAO,EACd,CAAC,mCAAmC,EAAEC,KAAK,SAAS,CAACD,SAAS;YAEhEhB,UAAUgB,OAAO,OAAO,CAAC,EAAE,CAAC,OAAO,CAAC,OAAO;YAC3Cd,QAAQc,OAAO,KAAK;QACtB;QAEAtB,UAAU,CAAC,UAAU,EAAEM,SAAS;QAChCkB,OAAOlB,SAAS;QAGhB,IAAID,eAAe,CAACG,OAAO;YAEzB,MAAMW,kBAAkBC,KAAK,GAAG,CAC9B,GACAA,KAAK,KAAK,CAAEd,AAAAA,CAAAA,WAAW,EAAC,EAAG,MAAM,GAAG;YAEtCE,QAAQ;gBACN,eAAeW;gBACf,mBAAmBA;gBACnB,cAAcA,AAAkB,IAAlBA;YAChB;QACF;QAEA,OAAO;YACL,SAASb,WAAW;YACpB,OAAOI,eAAeF;YACtB,YAAY,CAAC,CAACH;QAChB;IACF,EAAE,OAAOoB,GAAQ;QACfzC,QAAQ,KAAK,CAAC,kBAAkByC;QAChC,MAAMC,WAAW,IAAI3C,MACnB,CAAC,eAAe,EAAEsB,cAAc,eAAe,GAAG,kBAAkB,EAAEoB,EAAE,OAAO,CAAC,8DAA8D,CAAC,EAC/I;YACE,OAAOA;QACT;QAEF,MAAMC;IACR;AACF;AAEO,MAAM9B,oBAAoB,CAC/BjC,WACAJ;IAIA,IAAIoC;IAKJ,IAAIhC,UAAU,QAAQ,CAAC,UACrB,OAAQJ;QACN,KAAKoE,aAAa,MAAM;YACtBhC,iBAAiBiC;YACjB;QACF,KAAKD,aAAa,IAAI;YACpBhC,iBAAiBkC;YACjB;QACF,KAAKF,aAAa,YAAY;QAC9B,KAAKA,aAAa,gBAAgB;YAChChC,iBAAiB;gBAAE,MAAMmC,iBAAiB,IAAI;YAAC;YAC/C;QACF,KAAKH,aAAa,IAAI;YAEpBhC,iBAAiBmB;YACjB;IACJ;IAKF,IACEnD,AAAc,wBAAdA,aACAJ,sBAAsBoE,aAAa,IAAI,EAEvChC,iBAAiB;QAAE,MAAMmC,iBAAiB,IAAI;IAAC;IAGjD,OAAOnC;AACT;AAEO,eAAeoC,yBACpBvC,QAAsC,EACtCjC,iBAA+B,EAC/BC,WAAyB;IAEzB,MAAMwE,WAAW,MAAMzC,OAAOC,UAAUjC,mBAAmBC;IAC3DgE,OAAOQ,UAAU;IACjB,MAAM/D,SAAST,YAAY,MAAM;IACjC,MAAMyE,cAAcC,cAAcF,SAAS,OAAO,EAAE/D;IACpD,OAAO;QACL,SAASgE;QACT,eAAeD,SAAS,OAAO;QAC/B,OAAOA,SAAS,KAAK;IACvB;AACF;AAEO,eAAeG,yBACpBC,IAAY,EACZ7E,iBAA+B,EAC/BC,WAAyB;IAEzB,MAAM,EAAE8C,OAAO,EAAEE,KAAK,EAAE,GAAG,MAAMjB,OAAO6C,MAAM7E,mBAAmBC;IACjE,OAAO;QAAE8C;QAASE;IAAM;AAC1B;AAEO,SAAS6B,yBAAyBL,QAAgB;IACvD,IAAI;QAEF,MAAMM,YAAYN,SAAS,KAAK,CAAC;QACjC,IAAIM,WACF,OAAOA,SAAS,CAAC,EAAE;QAIrB,MAAMC,iBAAiBP,SAAS,KAAK,CACnC;QAEF,IAAIO,gBACF,OAAOA,cAAc,CAAC,EAAE;QAI1B,MAAMC,gBAAgBR,SAAS,KAAK,CAAC;QACrC,IAAIQ,eACF,OAAOA,aAAa,CAAC,EAAE;IAE3B,EAAE,OAAM,CAAC;IAET,OAAOR;AACT;AAEO,SAASS,yBAAyBC,KAAa;IACpD,IAAIA,MAAM,QAAQ,CAAC,SAEjB,MAAO,YAAY,IAAI,CAACA,OACtBA,QAAQA,MAAM,OAAO,CAAC,kBAAkB;IAG5C,OAAOA;AACT;AAQA,SAASC,oBAAoBC,GAAQ;IAEnC,IAAIA,QAAAA,KACF,OAAOA;IAIT,IAAIC,MAAM,OAAO,CAACD,MAChB,OAAOA,IAAI,GAAG,CAAC,CAACE,OAASH,oBAAoBG;IAI/C,IAAI,AAAe,YAAf,OAAOF,KAAkB;QAC3B,MAAMG,aAAkB,CAAC;QAEzB,KAAK,MAAM,CAACC,KAAKC,MAAM,IAAIC,OAAO,OAAO,CAACN,KAAM;YAE9C,MAAMO,aAAaH,IAAI,IAAI;YAG3B,IAAII,kBAAkBT,oBAAoBM;YAG1C,IAAI,AAA2B,YAA3B,OAAOG,iBACTA,kBAAkBA,gBAAgB,IAAI;YAGxCL,UAAU,CAACI,WAAW,GAAGC;QAC3B;QAEA,OAAOL;IACT;IAGA,IAAI,AAAe,YAAf,OAAOH,KACT,OAAOA,IAAI,IAAI;IAIjB,OAAOA;AACT;AAEO,SAASV,cAAcQ,KAAa,EAAEzE,MAAgC;IAC3E,MAAMoF,kBAAkBhB,yBAAyBK;IAEjD,IAAIW,iBAAiB,MAAM,oBACzB,OAAOA,gBACJ,KAAK,CAAC,oBACL,MAAM,GACP,IAAIC;IAGT,IAAIC;IACJ,IAAI;QACFA,SAAShC,KAAK,KAAK,CAAC8B;QACpB,OAAOV,oBAAoBY;IAC7B,EAAE,OAAM,CAAC;IACT,IAAI;QACFA,SAAShC,KAAK,KAAK,CAACiC,WAAWH;QAC/B,OAAOV,oBAAoBY;IAC7B,EAAE,OAAO9B,GAAG,CAAC;IAEb,IAAIxD,AAAW,oBAAXA,UAA8BA,AAAW,kBAAXA,QAA0B;QAC1D,MAAMwF,aAAahB,yBAAyBY;QAC5CE,SAAShC,KAAK,KAAK,CAACiC,WAAWC;QAC/B,OAAOd,oBAAoBY;IAC7B;IACA,MAAMxE,MAAM,CAAC,+BAA+B,EAAE2D,OAAO;AACvD"}
+{"version":3,"file":"ai-model/service-caller/index.mjs","sources":["../../../../src/ai-model/service-caller/index.ts"],"sourcesContent":["import { AIResponseFormat, type AIUsageInfo } from '@/types';\nimport type { CodeGenerationChunk, StreamingCallback } from '@/types';\nimport {\n  type IModelConfig,\n  MIDSCENE_LANGFUSE_DEBUG,\n  MIDSCENE_LANGSMITH_DEBUG,\n  MIDSCENE_MODEL_MAX_TOKENS,\n  OPENAI_MAX_TOKENS,\n  type TVlModeTypes,\n  type UITarsModelVersion,\n  globalConfigManager,\n} from '@midscene/shared/env';\n\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert, ifInBrowser } from '@midscene/shared/utils';\nimport { HttpsProxyAgent } from 'https-proxy-agent';\nimport { jsonrepair } from 'jsonrepair';\nimport OpenAI from 'openai';\nimport type { ChatCompletionMessageParam } from 'openai/resources/index';\nimport type { Stream } from 'openai/streaming';\nimport { SocksProxyAgent } from 'socks-proxy-agent';\nimport { AIActionType, type AIArgs } from '../../common';\nimport { assertSchema } from '../prompt/assertion';\nimport { planSchema } from '../prompt/llm-planning';\n\nasync function createChatClient({\n  AIActionTypeValue,\n  modelConfig,\n}: {\n  AIActionTypeValue: AIActionType;\n  modelConfig: IModelConfig;\n}): Promise<{\n  completion: OpenAI.Chat.Completions;\n  modelName: string;\n  modelDescription: string;\n  uiTarsVersion?: UITarsModelVersion;\n  vlMode: TVlModeTypes | undefined;\n}> {\n  const {\n    socksProxy,\n    httpProxy,\n    modelName,\n    openaiBaseURL,\n    openaiApiKey,\n    openaiExtraConfig,\n    modelDescription,\n    uiTarsModelVersion: uiTarsVersion,\n    vlMode,\n    createOpenAIClient,\n    timeout,\n  } = modelConfig;\n\n  let proxyAgent = undefined;\n  const debugProxy = getDebug('ai:call:proxy');\n  if (httpProxy) {\n    debugProxy('using http proxy', httpProxy);\n    proxyAgent = new HttpsProxyAgent(httpProxy);\n  } else if (socksProxy) {\n    debugProxy('using socks proxy', socksProxy);\n    proxyAgent = new SocksProxyAgent(socksProxy);\n  }\n\n  const openAIOptions = {\n    baseURL: openaiBaseURL,\n    apiKey: openaiApiKey,\n    ...(proxyAgent ? { httpAgent: proxyAgent as any } : {}),\n    ...openaiExtraConfig,\n    ...(typeof timeout === 'number' ? { timeout } : {}),\n    dangerouslyAllowBrowser: true,\n  };\n\n  const baseOpenAI = new OpenAI(openAIOptions);\n\n  let openai: OpenAI = baseOpenAI;\n\n  // LangSmith wrapper\n  if (\n    openai &&\n    globalConfigManager.getEnvConfigInBoolean(MIDSCENE_LANGSMITH_DEBUG)\n  ) {\n    if (ifInBrowser) {\n      throw new Error('langsmith is not supported in browser');\n    }\n    console.log('DEBUGGING MODE: langsmith wrapper enabled');\n    // Use variable to prevent static analysis by bundlers\n    const langsmithModule = 'langsmith/wrappers';\n    const { wrapOpenAI } = await import(langsmithModule);\n    openai = wrapOpenAI(openai);\n  }\n\n  // Langfuse wrapper\n  if (\n    openai &&\n    globalConfigManager.getEnvConfigInBoolean(MIDSCENE_LANGFUSE_DEBUG)\n  ) {\n    if (ifInBrowser) {\n      throw new Error('langfuse is not supported in browser');\n    }\n    console.log('DEBUGGING MODE: langfuse wrapper enabled');\n    // Use variable to prevent static analysis by bundlers\n    const langfuseModule = 'langfuse';\n    const { observeOpenAI } = await import(langfuseModule);\n    openai = observeOpenAI(openai);\n  }\n\n  if (createOpenAIClient) {\n    const wrappedClient = await createOpenAIClient(baseOpenAI, openAIOptions);\n\n    if (wrappedClient) {\n      openai = wrappedClient as OpenAI;\n    }\n  }\n\n  return {\n    completion: openai.chat.completions,\n    modelName,\n    modelDescription,\n    uiTarsVersion,\n    vlMode,\n  };\n}\n\nexport async function callAI(\n  messages: ChatCompletionMessageParam[],\n  AIActionTypeValue: AIActionType,\n  modelConfig: IModelConfig,\n  options?: {\n    stream?: boolean;\n    onChunk?: StreamingCallback;\n  },\n): Promise<{ content: string; usage?: AIUsageInfo; isStreamed: boolean }> {\n  const { completion, modelName, modelDescription, uiTarsVersion, vlMode } =\n    await createChatClient({\n      AIActionTypeValue,\n      modelConfig,\n    });\n\n  const responseFormat = getResponseFormat(modelName, AIActionTypeValue);\n\n  const maxTokens =\n    globalConfigManager.getEnvConfigValue(MIDSCENE_MODEL_MAX_TOKENS) ??\n    globalConfigManager.getEnvConfigValue(OPENAI_MAX_TOKENS);\n  const debugCall = getDebug('ai:call');\n  const debugProfileStats = getDebug('ai:profile:stats');\n  const debugProfileDetail = getDebug('ai:profile:detail');\n\n  const startTime = Date.now();\n\n  const isStreaming = options?.stream && options?.onChunk;\n  let content: string | undefined;\n  let accumulated = '';\n  let usage: OpenAI.CompletionUsage | undefined;\n  let timeCost: number | undefined;\n\n  const buildUsageInfo = (usageData?: OpenAI.CompletionUsage) => {\n    if (!usageData) return undefined;\n\n    const cachedInputTokens = (\n      usageData as { prompt_tokens_details?: { cached_tokens?: number } }\n    )?.prompt_tokens_details?.cached_tokens;\n\n    return {\n      prompt_tokens: usageData.prompt_tokens ?? 0,\n      completion_tokens: usageData.completion_tokens ?? 0,\n      total_tokens: usageData.total_tokens ?? 0,\n      cached_input: cachedInputTokens ?? 0,\n      time_cost: timeCost ?? 0,\n      model_name: modelName,\n      model_description: modelDescription,\n      intent: modelConfig.intent,\n    } satisfies AIUsageInfo;\n  };\n\n  const commonConfig = {\n    temperature: vlMode === 'vlm-ui-tars' ? 0.0 : undefined,\n    stream: !!isStreaming,\n    max_tokens: typeof maxTokens === 'number' ? maxTokens : undefined,\n    ...(vlMode === 'qwen2.5-vl' // qwen vl v2 specific config\n      ? {\n          vl_high_resolution_images: true,\n        }\n      : {}),\n  };\n\n  try {\n    debugCall(\n      `sending ${isStreaming ? 'streaming ' : ''}request to ${modelName}`,\n    );\n\n    if (isStreaming) {\n      const stream = (await completion.create(\n        {\n          model: modelName,\n          messages,\n          response_format: responseFormat,\n          ...commonConfig,\n        },\n        {\n          stream: true,\n        },\n      )) as Stream<OpenAI.Chat.Completions.ChatCompletionChunk> & {\n        _request_id?: string | null;\n      };\n\n      for await (const chunk of stream) {\n        const content = chunk.choices?.[0]?.delta?.content || '';\n        const reasoning_content =\n          (chunk.choices?.[0]?.delta as any)?.reasoning_content || '';\n\n        // Check for usage info in any chunk (OpenAI provides usage in separate chunks)\n        if (chunk.usage) {\n          usage = chunk.usage;\n        }\n\n        if (content || reasoning_content) {\n          accumulated += content;\n          const chunkData: CodeGenerationChunk = {\n            content,\n            reasoning_content,\n            accumulated,\n            isComplete: false,\n            usage: undefined,\n          };\n          options.onChunk!(chunkData);\n        }\n\n        // Check if stream is complete\n        if (chunk.choices?.[0]?.finish_reason) {\n          timeCost = Date.now() - startTime;\n\n          // If usage is not available from the stream, provide a basic usage info\n          if (!usage) {\n            // Estimate token counts based on content length (rough approximation)\n            const estimatedTokens = Math.max(\n              1,\n              Math.floor(accumulated.length / 4),\n            );\n            usage = {\n              prompt_tokens: estimatedTokens,\n              completion_tokens: estimatedTokens,\n              total_tokens: estimatedTokens * 2,\n            };\n          }\n\n          // Send final chunk\n          const finalChunk: CodeGenerationChunk = {\n            content: '',\n            accumulated,\n            reasoning_content: '',\n            isComplete: true,\n            usage: buildUsageInfo(usage),\n          };\n          options.onChunk!(finalChunk);\n          break;\n        }\n      }\n      content = accumulated;\n      debugProfileStats(\n        `streaming model, ${modelName}, mode, ${vlMode || 'default'}, cost-ms, ${timeCost}`,\n      );\n    } else {\n      const result = await completion.create({\n        model: modelName,\n        messages,\n        response_format: responseFormat,\n        ...commonConfig,\n      } as any);\n      timeCost = Date.now() - startTime;\n\n      debugProfileStats(\n        `model, ${modelName}, mode, ${vlMode || 'default'}, ui-tars-version, ${uiTarsVersion}, prompt-tokens, ${result.usage?.prompt_tokens || ''}, completion-tokens, ${result.usage?.completion_tokens || ''}, total-tokens, ${result.usage?.total_tokens || ''}, cost-ms, ${timeCost}, requestId, ${result._request_id || ''}`,\n      );\n\n      debugProfileDetail(`model usage detail: ${JSON.stringify(result.usage)}`);\n\n      assert(\n        result.choices,\n        `invalid response from LLM service: ${JSON.stringify(result)}`,\n      );\n      content = result.choices[0].message.content!;\n      usage = result.usage;\n    }\n\n    debugCall(`response: ${content}`);\n    assert(content, 'empty content');\n\n    // Ensure we always have usage info for streaming responses\n    if (isStreaming && !usage) {\n      // Estimate token counts based on content length (rough approximation)\n      const estimatedTokens = Math.max(\n        1,\n        Math.floor((content || '').length / 4),\n      );\n      usage = {\n        prompt_tokens: estimatedTokens,\n        completion_tokens: estimatedTokens,\n        total_tokens: estimatedTokens * 2,\n      } as OpenAI.CompletionUsage;\n    }\n\n    return {\n      content: content || '',\n      usage: buildUsageInfo(usage),\n      isStreamed: !!isStreaming,\n    };\n  } catch (e: any) {\n    console.error(' call AI error', e);\n    const newError = new Error(\n      `failed to call ${isStreaming ? 'streaming ' : ''}AI model service: ${e.message}. Trouble shooting: https://midscenejs.com/model-provider.html`,\n      {\n        cause: e,\n      },\n    );\n    throw newError;\n  }\n}\n\nexport const getResponseFormat = (\n  modelName: string,\n  AIActionTypeValue: AIActionType,\n):\n  | OpenAI.ChatCompletionCreateParams['response_format']\n  | OpenAI.ResponseFormatJSONObject => {\n  let responseFormat:\n    | OpenAI.ChatCompletionCreateParams['response_format']\n    | OpenAI.ResponseFormatJSONObject\n    | undefined;\n\n  if (modelName.includes('gpt-4')) {\n    switch (AIActionTypeValue) {\n      case AIActionType.ASSERT:\n        responseFormat = assertSchema;\n        break;\n      case AIActionType.PLAN:\n        responseFormat = planSchema;\n        break;\n      case AIActionType.EXTRACT_DATA:\n      case AIActionType.DESCRIBE_ELEMENT:\n        responseFormat = { type: AIResponseFormat.JSON };\n        break;\n      case AIActionType.TEXT:\n        // No response format for plain text - return as-is\n        responseFormat = undefined;\n        break;\n    }\n  }\n\n  // gpt-4o-2024-05-13 only supports json_object response format\n  // Skip for plain text to allow string output\n  if (\n    modelName === 'gpt-4o-2024-05-13' &&\n    AIActionTypeValue !== AIActionType.TEXT\n  ) {\n    responseFormat = { type: AIResponseFormat.JSON };\n  }\n\n  return responseFormat;\n};\n\nexport async function callAIWithObjectResponse<T>(\n  messages: ChatCompletionMessageParam[],\n  AIActionTypeValue: AIActionType,\n  modelConfig: IModelConfig,\n): Promise<{ content: T; contentString: string; usage?: AIUsageInfo }> {\n  const response = await callAI(messages, AIActionTypeValue, modelConfig);\n  assert(response, 'empty response');\n  const vlMode = modelConfig.vlMode;\n  const jsonContent = safeParseJson(response.content, vlMode);\n  return {\n    content: jsonContent,\n    contentString: response.content,\n    usage: response.usage,\n  };\n}\n\nexport async function callAIWithStringResponse(\n  msgs: AIArgs,\n  AIActionTypeValue: AIActionType,\n  modelConfig: IModelConfig,\n): Promise<{ content: string; usage?: AIUsageInfo }> {\n  const { content, usage } = await callAI(msgs, AIActionTypeValue, modelConfig);\n  return { content, usage };\n}\n\nexport function extractJSONFromCodeBlock(response: string) {\n  try {\n    // First, try to match a JSON object directly in the response\n    const jsonMatch = response.match(/^\\s*(\\{[\\s\\S]*\\})\\s*$/);\n    if (jsonMatch) {\n      return jsonMatch[1];\n    }\n\n    // If no direct JSON object is found, try to extract JSON from a code block\n    const codeBlockMatch = response.match(\n      /```(?:json)?\\s*(\\{[\\s\\S]*?\\})\\s*```/,\n    );\n    if (codeBlockMatch) {\n      return codeBlockMatch[1];\n    }\n\n    // If no code block is found, try to find a JSON-like structure in the text\n    const jsonLikeMatch = response.match(/\\{[\\s\\S]*\\}/);\n    if (jsonLikeMatch) {\n      return jsonLikeMatch[0];\n    }\n  } catch {}\n  // If no JSON-like structure is found, return the original response\n  return response;\n}\n\nexport function preprocessDoubaoBboxJson(input: string) {\n  if (input.includes('bbox')) {\n    // when its values like 940 445 969 490, replace all /\\d+\\s+\\d+/g with /$1,$2/g\n    while (/\\d+\\s+\\d+/.test(input)) {\n      input = input.replace(/(\\d+)\\s+(\\d+)/g, '$1,$2');\n    }\n  }\n  return input;\n}\n\n/**\n * Normalize a parsed JSON object by trimming whitespace from:\n * 1. All object keys (e.g., \" prompt \" -> \"prompt\")\n * 2. All string values (e.g., \" Tap \" -> \"Tap\")\n * This handles LLM output that may include leading/trailing spaces.\n */\nfunction normalizeJsonObject(obj: any): any {\n  // Handle null and undefined\n  if (obj === null || obj === undefined) {\n    return obj;\n  }\n\n  // Handle arrays - recursively normalize each element\n  if (Array.isArray(obj)) {\n    return obj.map((item) => normalizeJsonObject(item));\n  }\n\n  // Handle objects\n  if (typeof obj === 'object') {\n    const normalized: any = {};\n\n    for (const [key, value] of Object.entries(obj)) {\n      // Trim the key to remove leading/trailing spaces\n      const trimmedKey = key.trim();\n\n      // Recursively normalize the value\n      let normalizedValue = normalizeJsonObject(value);\n\n      // Trim all string values\n      if (typeof normalizedValue === 'string') {\n        normalizedValue = normalizedValue.trim();\n      }\n\n      normalized[trimmedKey] = normalizedValue;\n    }\n\n    return normalized;\n  }\n\n  // Handle primitive strings\n  if (typeof obj === 'string') {\n    return obj.trim();\n  }\n\n  // Return other primitives as-is\n  return obj;\n}\n\nexport function safeParseJson(input: string, vlMode: TVlModeTypes | undefined) {\n  const cleanJsonString = extractJSONFromCodeBlock(input);\n  // match the point\n  if (cleanJsonString?.match(/\\((\\d+),(\\d+)\\)/)) {\n    return cleanJsonString\n      .match(/\\((\\d+),(\\d+)\\)/)\n      ?.slice(1)\n      .map(Number);\n  }\n\n  let parsed: any;\n  try {\n    parsed = JSON.parse(cleanJsonString);\n    return normalizeJsonObject(parsed);\n  } catch {}\n  try {\n    parsed = JSON.parse(jsonrepair(cleanJsonString));\n    return normalizeJsonObject(parsed);\n  } catch (e) {}\n\n  if (vlMode === 'doubao-vision' || vlMode === 'vlm-ui-tars') {\n    const jsonString = preprocessDoubaoBboxJson(cleanJsonString);\n    parsed = JSON.parse(jsonrepair(jsonString));\n    return normalizeJsonObject(parsed);\n  }\n  throw Error(`failed to parse json response: ${input}`);\n}\n"],"names":["createChatClient","AIActionTypeValue","modelConfig","socksProxy","httpProxy","modelName","openaiBaseURL","openaiApiKey","openaiExtraConfig","modelDescription","uiTarsVersion","vlMode","createOpenAIClient","timeout","proxyAgent","debugProxy","getDebug","HttpsProxyAgent","SocksProxyAgent","openAIOptions","baseOpenAI","OpenAI","openai","globalConfigManager","MIDSCENE_LANGSMITH_DEBUG","ifInBrowser","Error","console","langsmithModule","wrapOpenAI","MIDSCENE_LANGFUSE_DEBUG","langfuseModule","observeOpenAI","wrappedClient","callAI","messages","options","completion","responseFormat","getResponseFormat","maxTokens","MIDSCENE_MODEL_MAX_TOKENS","OPENAI_MAX_TOKENS","debugCall","debugProfileStats","debugProfileDetail","startTime","Date","isStreaming","content","accumulated","usage","timeCost","buildUsageInfo","usageData","cachedInputTokens","commonConfig","undefined","stream","chunk","reasoning_content","chunkData","estimatedTokens","Math","finalChunk","result","JSON","assert","e","newError","AIActionType","assertSchema","planSchema","AIResponseFormat","callAIWithObjectResponse","response","jsonContent","safeParseJson","callAIWithStringResponse","msgs","extractJSONFromCodeBlock","jsonMatch","codeBlockMatch","jsonLikeMatch","preprocessDoubaoBboxJson","input","normalizeJsonObject","obj","Array","item","normalized","key","value","Object","trimmedKey","normalizedValue","cleanJsonString","Number","parsed","jsonrepair","jsonString"],"mappings":";;;;;;;;;;;AAyBA,eAAeA,iBAAiB,EAC9BC,iBAAiB,EACjBC,WAAW,EAIZ;IAOC,MAAM,EACJC,UAAU,EACVC,SAAS,EACTC,SAAS,EACTC,aAAa,EACbC,YAAY,EACZC,iBAAiB,EACjBC,gBAAgB,EAChB,oBAAoBC,aAAa,EACjCC,MAAM,EACNC,kBAAkB,EAClBC,OAAO,EACR,GAAGX;IAEJ,IAAIY;IACJ,MAAMC,aAAaC,SAAS;IAC5B,IAAIZ,WAAW;QACbW,WAAW,oBAAoBX;QAC/BU,aAAa,IAAIG,gBAAgBb;IACnC,OAAO,IAAID,YAAY;QACrBY,WAAW,qBAAqBZ;QAChCW,aAAa,IAAII,gBAAgBf;IACnC;IAEA,MAAMgB,gBAAgB;QACpB,SAASb;QACT,QAAQC;QACR,GAAIO,aAAa;YAAE,WAAWA;QAAkB,IAAI,CAAC,CAAC;QACtD,GAAGN,iBAAiB;QACpB,GAAI,AAAmB,YAAnB,OAAOK,UAAuB;YAAEA;QAAQ,IAAI,CAAC,CAAC;QAClD,yBAAyB;IAC3B;IAEA,MAAMO,aAAa,IAAIC,SAAOF;IAE9B,IAAIG,SAAiBF;IAGrB,IACEE,UACAC,oBAAoB,qBAAqB,CAACC,2BAC1C;QACA,IAAIC,aACF,MAAM,IAAIC,MAAM;QAElBC,QAAQ,GAAG,CAAC;QAEZ,MAAMC,kBAAkB;QACxB,MAAM,EAAEC,UAAU,EAAE,GAAG,MAAM,MAAM,CAACD;QACpCN,SAASO,WAAWP;IACtB;IAGA,IACEA,UACAC,oBAAoB,qBAAqB,CAACO,0BAC1C;QACA,IAAIL,aACF,MAAM,IAAIC,MAAM;QAElBC,QAAQ,GAAG,CAAC;QAEZ,MAAMI,iBAAiB;QACvB,MAAM,EAAEC,aAAa,EAAE,GAAG,MAAM,MAAM,CAACD;QACvCT,SAASU,cAAcV;IACzB;IAEA,IAAIV,oBAAoB;QACtB,MAAMqB,gBAAgB,MAAMrB,mBAAmBQ,YAAYD;QAE3D,IAAIc,eACFX,SAASW;IAEb;IAEA,OAAO;QACL,YAAYX,OAAO,IAAI,CAAC,WAAW;QACnCjB;QACAI;QACAC;QACAC;IACF;AACF;AAEO,eAAeuB,OACpBC,QAAsC,EACtClC,iBAA+B,EAC/BC,WAAyB,EACzBkC,OAGC;IAED,MAAM,EAAEC,UAAU,EAAEhC,SAAS,EAAEI,gBAAgB,EAAEC,aAAa,EAAEC,MAAM,EAAE,GACtE,MAAMX,iBAAiB;QACrBC;QACAC;IACF;IAEF,MAAMoC,iBAAiBC,kBAAkBlC,WAAWJ;IAEpD,MAAMuC,YACJjB,oBAAoB,iBAAiB,CAACkB,8BACtClB,oBAAoB,iBAAiB,CAACmB;IACxC,MAAMC,YAAY3B,SAAS;IAC3B,MAAM4B,oBAAoB5B,SAAS;IACnC,MAAM6B,qBAAqB7B,SAAS;IAEpC,MAAM8B,YAAYC,KAAK,GAAG;IAE1B,MAAMC,cAAcZ,SAAS,UAAUA,SAAS;IAChD,IAAIa;IACJ,IAAIC,cAAc;IAClB,IAAIC;IACJ,IAAIC;IAEJ,MAAMC,iBAAiB,CAACC;QACtB,IAAI,CAACA,WAAW;QAEhB,MAAMC,oBACJD,WACC,uBAAuB;QAE1B,OAAO;YACL,eAAeA,UAAU,aAAa,IAAI;YAC1C,mBAAmBA,UAAU,iBAAiB,IAAI;YAClD,cAAcA,UAAU,YAAY,IAAI;YACxC,cAAcC,qBAAqB;YACnC,WAAWH,YAAY;YACvB,YAAY/C;YACZ,mBAAmBI;YACnB,QAAQP,YAAY,MAAM;QAC5B;IACF;IAEA,MAAMsD,eAAe;QACnB,aAAa7C,AAAW,kBAAXA,SAA2B,MAAM8C;QAC9C,QAAQ,CAAC,CAACT;QACV,YAAY,AAAqB,YAArB,OAAOR,YAAyBA,YAAYiB;QACxD,GAAI9C,AAAW,iBAAXA,SACA;YACE,2BAA2B;QAC7B,IACA,CAAC,CAAC;IACR;IAEA,IAAI;QACFgC,UACE,CAAC,QAAQ,EAAEK,cAAc,eAAe,GAAG,WAAW,EAAE3C,WAAW;QAGrE,IAAI2C,aAAa;YACf,MAAMU,SAAU,MAAMrB,WAAW,MAAM,CACrC;gBACE,OAAOhC;gBACP8B;gBACA,iBAAiBG;gBACjB,GAAGkB,YAAY;YACjB,GACA;gBACE,QAAQ;YACV;YAKF,WAAW,MAAMG,SAASD,OAAQ;gBAChC,MAAMT,UAAUU,MAAM,OAAO,EAAE,CAAC,EAAE,EAAE,OAAO,WAAW;gBACtD,MAAMC,oBACHD,MAAM,OAAO,EAAE,CAAC,EAAE,EAAE,OAAe,qBAAqB;gBAG3D,IAAIA,MAAM,KAAK,EACbR,QAAQQ,MAAM,KAAK;gBAGrB,IAAIV,WAAWW,mBAAmB;oBAChCV,eAAeD;oBACf,MAAMY,YAAiC;wBACrCZ;wBACAW;wBACAV;wBACA,YAAY;wBACZ,OAAOO;oBACT;oBACArB,QAAQ,OAAO,CAAEyB;gBACnB;gBAGA,IAAIF,MAAM,OAAO,EAAE,CAAC,EAAE,EAAE,eAAe;oBACrCP,WAAWL,KAAK,GAAG,KAAKD;oBAGxB,IAAI,CAACK,OAAO;wBAEV,MAAMW,kBAAkBC,KAAK,GAAG,CAC9B,GACAA,KAAK,KAAK,CAACb,YAAY,MAAM,GAAG;wBAElCC,QAAQ;4BACN,eAAeW;4BACf,mBAAmBA;4BACnB,cAAcA,AAAkB,IAAlBA;wBAChB;oBACF;oBAGA,MAAME,aAAkC;wBACtC,SAAS;wBACTd;wBACA,mBAAmB;wBACnB,YAAY;wBACZ,OAAOG,eAAeF;oBACxB;oBACAf,QAAQ,OAAO,CAAE4B;oBACjB;gBACF;YACF;YACAf,UAAUC;YACVN,kBACE,CAAC,iBAAiB,EAAEvC,UAAU,QAAQ,EAAEM,UAAU,UAAU,WAAW,EAAEyC,UAAU;QAEvF,OAAO;YACL,MAAMa,SAAS,MAAM5B,WAAW,MAAM,CAAC;gBACrC,OAAOhC;gBACP8B;gBACA,iBAAiBG;gBACjB,GAAGkB,YAAY;YACjB;YACAJ,WAAWL,KAAK,GAAG,KAAKD;YAExBF,kBACE,CAAC,OAAO,EAAEvC,UAAU,QAAQ,EAAEM,UAAU,UAAU,mBAAmB,EAAED,cAAc,iBAAiB,EAAEuD,OAAO,KAAK,EAAE,iBAAiB,GAAG,qBAAqB,EAAEA,OAAO,KAAK,EAAE,qBAAqB,GAAG,gBAAgB,EAAEA,OAAO,KAAK,EAAE,gBAAgB,GAAG,WAAW,EAAEb,SAAS,aAAa,EAAEa,OAAO,WAAW,IAAI,IAAI;YAG3TpB,mBAAmB,CAAC,oBAAoB,EAAEqB,KAAK,SAAS,CAACD,OAAO,KAAK,GAAG;YAExEE,OACEF,OAAO,OAAO,EACd,CAAC,mCAAmC,EAAEC,KAAK,SAAS,CAACD,SAAS;YAEhEhB,UAAUgB,OAAO,OAAO,CAAC,EAAE,CAAC,OAAO,CAAC,OAAO;YAC3Cd,QAAQc,OAAO,KAAK;QACtB;QAEAtB,UAAU,CAAC,UAAU,EAAEM,SAAS;QAChCkB,OAAOlB,SAAS;QAGhB,IAAID,eAAe,CAACG,OAAO;YAEzB,MAAMW,kBAAkBC,KAAK,GAAG,CAC9B,GACAA,KAAK,KAAK,CAAEd,AAAAA,CAAAA,WAAW,EAAC,EAAG,MAAM,GAAG;YAEtCE,QAAQ;gBACN,eAAeW;gBACf,mBAAmBA;gBACnB,cAAcA,AAAkB,IAAlBA;YAChB;QACF;QAEA,OAAO;YACL,SAASb,WAAW;YACpB,OAAOI,eAAeF;YACtB,YAAY,CAAC,CAACH;QAChB;IACF,EAAE,OAAOoB,GAAQ;QACfzC,QAAQ,KAAK,CAAC,kBAAkByC;QAChC,MAAMC,WAAW,IAAI3C,MACnB,CAAC,eAAe,EAAEsB,cAAc,eAAe,GAAG,kBAAkB,EAAEoB,EAAE,OAAO,CAAC,8DAA8D,CAAC,EAC/I;YACE,OAAOA;QACT;QAEF,MAAMC;IACR;AACF;AAEO,MAAM9B,oBAAoB,CAC/BlC,WACAJ;IAIA,IAAIqC;IAKJ,IAAIjC,UAAU,QAAQ,CAAC,UACrB,OAAQJ;QACN,KAAKqE,aAAa,MAAM;YACtBhC,iBAAiBiC;YACjB;QACF,KAAKD,aAAa,IAAI;YACpBhC,iBAAiBkC;YACjB;QACF,KAAKF,aAAa,YAAY;QAC9B,KAAKA,aAAa,gBAAgB;YAChChC,iBAAiB;gBAAE,MAAMmC,iBAAiB,IAAI;YAAC;YAC/C;QACF,KAAKH,aAAa,IAAI;YAEpBhC,iBAAiBmB;YACjB;IACJ;IAKF,IACEpD,AAAc,wBAAdA,aACAJ,sBAAsBqE,aAAa,IAAI,EAEvChC,iBAAiB;QAAE,MAAMmC,iBAAiB,IAAI;IAAC;IAGjD,OAAOnC;AACT;AAEO,eAAeoC,yBACpBvC,QAAsC,EACtClC,iBAA+B,EAC/BC,WAAyB;IAEzB,MAAMyE,WAAW,MAAMzC,OAAOC,UAAUlC,mBAAmBC;IAC3DiE,OAAOQ,UAAU;IACjB,MAAMhE,SAAST,YAAY,MAAM;IACjC,MAAM0E,cAAcC,cAAcF,SAAS,OAAO,EAAEhE;IACpD,OAAO;QACL,SAASiE;QACT,eAAeD,SAAS,OAAO;QAC/B,OAAOA,SAAS,KAAK;IACvB;AACF;AAEO,eAAeG,yBACpBC,IAAY,EACZ9E,iBAA+B,EAC/BC,WAAyB;IAEzB,MAAM,EAAE+C,OAAO,EAAEE,KAAK,EAAE,GAAG,MAAMjB,OAAO6C,MAAM9E,mBAAmBC;IACjE,OAAO;QAAE+C;QAASE;IAAM;AAC1B;AAEO,SAAS6B,yBAAyBL,QAAgB;IACvD,IAAI;QAEF,MAAMM,YAAYN,SAAS,KAAK,CAAC;QACjC,IAAIM,WACF,OAAOA,SAAS,CAAC,EAAE;QAIrB,MAAMC,iBAAiBP,SAAS,KAAK,CACnC;QAEF,IAAIO,gBACF,OAAOA,cAAc,CAAC,EAAE;QAI1B,MAAMC,gBAAgBR,SAAS,KAAK,CAAC;QACrC,IAAIQ,eACF,OAAOA,aAAa,CAAC,EAAE;IAE3B,EAAE,OAAM,CAAC;IAET,OAAOR;AACT;AAEO,SAASS,yBAAyBC,KAAa;IACpD,IAAIA,MAAM,QAAQ,CAAC,SAEjB,MAAO,YAAY,IAAI,CAACA,OACtBA,QAAQA,MAAM,OAAO,CAAC,kBAAkB;IAG5C,OAAOA;AACT;AAQA,SAASC,oBAAoBC,GAAQ;IAEnC,IAAIA,QAAAA,KACF,OAAOA;IAIT,IAAIC,MAAM,OAAO,CAACD,MAChB,OAAOA,IAAI,GAAG,CAAC,CAACE,OAASH,oBAAoBG;IAI/C,IAAI,AAAe,YAAf,OAAOF,KAAkB;QAC3B,MAAMG,aAAkB,CAAC;QAEzB,KAAK,MAAM,CAACC,KAAKC,MAAM,IAAIC,OAAO,OAAO,CAACN,KAAM;YAE9C,MAAMO,aAAaH,IAAI,IAAI;YAG3B,IAAII,kBAAkBT,oBAAoBM;YAG1C,IAAI,AAA2B,YAA3B,OAAOG,iBACTA,kBAAkBA,gBAAgB,IAAI;YAGxCL,UAAU,CAACI,WAAW,GAAGC;QAC3B;QAEA,OAAOL;IACT;IAGA,IAAI,AAAe,YAAf,OAAOH,KACT,OAAOA,IAAI,IAAI;IAIjB,OAAOA;AACT;AAEO,SAASV,cAAcQ,KAAa,EAAE1E,MAAgC;IAC3E,MAAMqF,kBAAkBhB,yBAAyBK;IAEjD,IAAIW,iBAAiB,MAAM,oBACzB,OAAOA,gBACJ,KAAK,CAAC,oBACL,MAAM,GACP,IAAIC;IAGT,IAAIC;IACJ,IAAI;QACFA,SAAShC,KAAK,KAAK,CAAC8B;QACpB,OAAOV,oBAAoBY;IAC7B,EAAE,OAAM,CAAC;IACT,IAAI;QACFA,SAAShC,KAAK,KAAK,CAACiC,WAAWH;QAC/B,OAAOV,oBAAoBY;IAC7B,EAAE,OAAO9B,GAAG,CAAC;IAEb,IAAIzD,AAAW,oBAAXA,UAA8BA,AAAW,kBAAXA,QAA0B;QAC1D,MAAMyF,aAAahB,yBAAyBY;QAC5CE,SAAShC,KAAK,KAAK,CAACiC,WAAWC;QAC/B,OAAOd,oBAAoBY;IAC7B;IACA,MAAMxE,MAAM,CAAC,+BAA+B,EAAE2D,OAAO;AACvD"}

package/dist/es/ai-model/ui-tars-planning.mjs

@@ -16,9 +16,11 @@ const pointToBbox = (point, width, height)=>[
         Math.round(Math.min(point.y + bboxSize / 2, height))
     ];
 async function uiTarsPlanning(userInstruction, options) {
-    const { conversationHistory, context, modelConfig } = options;
+    const { conversationHistory, context, modelConfig, actionContext } = options;
     const { uiTarsModelVersion } = modelConfig;
-    const systemPrompt = getUiTarsPlanningPrompt() + userInstruction;
+    let instruction = userInstruction;
+    if (actionContext) instruction = `<high_priority_knowledge>${actionContext}</high_priority_knowledge>\n<user_instruction>${userInstruction}</user_instruction>`;
+    const systemPrompt = getUiTarsPlanningPrompt() + instruction;
     const imagePayload = await resizeImageForUiTars(context.screenshotBase64, context.size, uiTarsModelVersion);
     conversationHistory.append({
         role: 'user',
@@ -54,6 +56,7 @@ async function uiTarsPlanning(userInstruction, options) {
     });
     debug('ui-tars modelVer', uiTarsModelVersion, ', parsed', JSON.stringify(parsed));
     const transformActions = [];
+    const unhandledActions = [];
     let shouldContinue = true;
     parsed.forEach((action)=>{
         const actionType = (action.action_type || '').toLowerCase();
@@ -72,6 +75,38 @@ async function uiTarsPlanning(userInstruction, options) {
                     }
                 }
             });
+        } else if ('left_double' === actionType) {
+            assert(action.action_inputs.start_box, 'start_box is required');
+            const point = getPoint(action.action_inputs.start_box, size);
+            transformActions.push({
+                type: 'DoubleClick',
+                param: {
+                    locate: {
+                        prompt: action.thought || '',
+                        bbox: pointToBbox({
+                            x: point[0],
+                            y: point[1]
+                        }, size.width, size.height)
+                    }
+                },
+                thought: action.thought || ''
+            });
+        } else if ('right_single' === actionType) {
+            assert(action.action_inputs.start_box, 'start_box is required');
+            const point = getPoint(action.action_inputs.start_box, size);
+            transformActions.push({
+                type: 'RightClick',
+                param: {
+                    locate: {
+                        prompt: action.thought || '',
+                        bbox: pointToBbox({
+                            x: point[0],
+                            y: point[1]
+                        }, size.width, size.height)
+                    }
+                },
+                thought: action.thought || ''
+            });
         } else if ('drag' === actionType) {
             assert(action.action_inputs.start_box, 'start_box is required');
             assert(action.action_inputs.end_box, 'end_box is required');
@@ -135,13 +170,39 @@ async function uiTarsPlanning(userInstruction, options) {
             },
             thought: action.thought || ''
         });
-    });
-    if (0 === transformActions.length) throw new Error(`No actions found, response: ${res.content}`, {
-        cause: {
-            prediction: res.content,
-            parsed
+        else if (actionType) {
+            unhandledActions.push({
+                type: actionType,
+                thought: action.thought || ''
+            });
+            debug('Unhandled action type:', actionType, 'thought:', action.thought);
         }
     });
+    if (0 === transformActions.length) {
+        const errorDetails = [];
+        if (0 === parsed.length) {
+            errorDetails.push('Action parser returned no actions');
+            if (res.content.includes('Thought:') && !res.content.includes('Action:')) errorDetails.push('Response contains "Thought:" but missing "Action:" line');
+            else errorDetails.push('Response may be malformed or empty');
+        }
+        if (unhandledActions.length > 0) {
+            const types = unhandledActions.map((a)=>a.type).join(', ');
+            errorDetails.push(`Unhandled action types: ${types}`);
+        }
+        const errorMessage = [
+            'No actions found in UI-TARS response.',
+            ...errorDetails,
+            `\nRaw response: ${res.content}`
+        ].join('\n');
+        throw new Error(errorMessage, {
+            cause: {
+                prediction: res.content,
+                parsed,
+                unhandledActions,
+                convertedText
+            }
+        });
+    }
     debug('transformActions', JSON.stringify(transformActions, null, 2));
     const log = getSummary(res.content);
     conversationHistory.append({

package/dist/es/ai-model/ui-tars-planning.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"ai-model/ui-tars-planning.mjs","sources":["../../../src/ai-model/ui-tars-planning.ts"],"sourcesContent":["import type {\n  PlanningAIResponse,\n  PlanningAction,\n  Size,\n  UIContext,\n} from '@/types';\nimport { type IModelConfig, UITarsModelVersion } from '@midscene/shared/env';\nimport { resizeImgBase64 } from '@midscene/shared/img';\nimport { getDebug } from '@midscene/shared/logger';\nimport { transformHotkeyInput } from '@midscene/shared/us-keyboard-layout';\nimport { assert } from '@midscene/shared/utils';\nimport { actionParser } from '@ui-tars/action-parser';\nimport { AIActionType } from '../common';\nimport type { ConversationHistory } from './conversation-history';\nimport { getSummary, getUiTarsPlanningPrompt } from './prompt/ui-tars-planning';\nimport { callAIWithStringResponse } from './service-caller/index';\ntype ActionType =\n  | 'click'\n  | 'drag'\n  | 'type'\n  | 'hotkey'\n  | 'finished'\n  | 'scroll'\n  | 'wait';\n\nconst debug = getDebug('ui-tars-planning');\nconst bboxSize = 10;\nconst pointToBbox = (\n  point: { x: number; y: number },\n  width: number,\n  height: number,\n): [number, number, number, number] => {\n  return [\n    Math.round(Math.max(point.x - bboxSize / 2, 0)),\n    Math.round(Math.max(point.y - bboxSize / 2, 0)),\n    Math.round(Math.min(point.x + bboxSize / 2, width)),\n    Math.round(Math.min(point.y + bboxSize / 2, height)),\n  ];\n};\n\nexport async function uiTarsPlanning(\n  userInstruction: string,\n  options: {\n    conversationHistory: ConversationHistory;\n    context: UIContext;\n    modelConfig: IModelConfig;\n  },\n): Promise<PlanningAIResponse> {\n  const { conversationHistory, context, modelConfig } = options;\n  const { uiTarsModelVersion } = modelConfig;\n  const systemPrompt = getUiTarsPlanningPrompt() + userInstruction;\n\n  const imagePayload = await resizeImageForUiTars(\n    context.screenshotBase64,\n    context.size,\n    uiTarsModelVersion,\n  );\n\n  conversationHistory.append({\n    role: 'user',\n    content: [\n      {\n        type: 'image_url',\n        image_url: {\n          url: imagePayload,\n        },\n      },\n    ],\n  });\n\n  const res = await callAIWithStringResponse(\n    [\n      {\n        role: 'user',\n        content: systemPrompt,\n      },\n      ...conversationHistory.snapshot(),\n    ],\n    AIActionType.INSPECT_ELEMENT,\n    modelConfig,\n  );\n  const convertedText = convertBboxToCoordinates(res.content);\n\n  const { size } = context;\n  const { parsed } = actionParser({\n    prediction: convertedText,\n    factor: [1000, 1000],\n    screenContext: {\n      width: size.width,\n      height: size.height,\n    },\n    modelVer: uiTarsModelVersion,\n  });\n\n  debug(\n    'ui-tars modelVer',\n    uiTarsModelVersion,\n    ', parsed',\n    JSON.stringify(parsed),\n  );\n\n  const transformActions: PlanningAction[] = [];\n  let shouldContinue = true;\n  parsed.forEach((action) => {\n    const actionType = (action.action_type || '').toLowerCase();\n    if (actionType === 'click') {\n      assert(action.action_inputs.start_box, 'start_box is required');\n      const point = getPoint(action.action_inputs.start_box, size);\n      transformActions.push({\n        type: 'Tap',\n        param: {\n          locate: {\n            prompt: action.thought || '',\n            bbox: pointToBbox(\n              { x: point[0], y: point[1] },\n              size.width,\n              size.height,\n            ),\n          },\n        },\n      });\n    } else if (actionType === 'drag') {\n      assert(action.action_inputs.start_box, 'start_box is required');\n      assert(action.action_inputs.end_box, 'end_box is required');\n      const startPoint = getPoint(action.action_inputs.start_box, size);\n      const endPoint = getPoint(action.action_inputs.end_box, size);\n      transformActions.push({\n        type: 'DragAndDrop',\n        param: {\n          from: {\n            prompt: action.thought || '',\n            bbox: pointToBbox(\n              { x: startPoint[0], y: startPoint[1] },\n              size.width,\n              size.height,\n            ),\n          },\n          to: {\n            prompt: action.thought || '',\n            bbox: pointToBbox(\n              { x: endPoint[0], y: endPoint[1] },\n              size.width,\n              size.height,\n            ),\n          },\n        },\n        thought: action.thought || '',\n      });\n    } else if (actionType === 'type') {\n      transformActions.push({\n        type: 'Input',\n        param: {\n          value: action.action_inputs.content,\n        },\n        thought: action.thought || '',\n      });\n    } else if (actionType === 'scroll') {\n      transformActions.push({\n        type: 'Scroll',\n        param: {\n          direction: action.action_inputs.direction,\n        },\n        thought: action.thought || '',\n      });\n    } else if (actionType === 'finished') {\n      shouldContinue = false;\n      transformActions.push({\n        type: 'Finished',\n        param: {},\n        thought: action.thought || '',\n      });\n    } else if (actionType === 'hotkey') {\n      if (!action.action_inputs.key) {\n        console.warn(\n          'No key found in action: hotkey. Will not perform action.',\n        );\n      } else {\n        const keys = transformHotkeyInput(action.action_inputs.key);\n\n        transformActions.push({\n          type: 'KeyboardPress',\n          param: {\n            keyName: keys,\n          },\n          thought: action.thought || '',\n        });\n      }\n    } else if (actionType === 'wait') {\n      transformActions.push({\n        type: 'Sleep',\n        param: {\n          timeMs: 1000,\n        },\n        thought: action.thought || '',\n      });\n    }\n  });\n\n  if (transformActions.length === 0) {\n    throw new Error(`No actions found, response: ${res.content}`, {\n      cause: {\n        prediction: res.content,\n        parsed,\n      },\n    });\n  }\n\n  debug('transformActions', JSON.stringify(transformActions, null, 2));\n  const log = getSummary(res.content);\n\n  conversationHistory.append({\n    role: 'assistant',\n    content: log,\n  });\n\n  return {\n    actions: transformActions,\n    log,\n    usage: res.usage,\n    rawResponse: JSON.stringify(res.content, undefined, 2),\n    more_actions_needed_by_instruction: shouldContinue,\n  };\n}\n\n/**\n * Converts bounding box notation to coordinate points\n * @param text - The text containing bbox tags to be converted\n * @returns The text with bbox tags replaced by coordinate points\n */\nfunction convertBboxToCoordinates(text: string): string {\n  // Match the four numbers after <bbox>\n  const pattern = /<bbox>(\\d+)\\s+(\\d+)\\s+(\\d+)\\s+(\\d+)<\\/bbox>/g;\n\n  function replaceMatch(\n    match: string,\n    x1: string,\n    y1: string,\n    x2: string,\n    y2: string,\n  ): string {\n    // Convert strings to numbers and calculate center point\n    const x1Num = Number.parseInt(x1, 10);\n    const y1Num = Number.parseInt(y1, 10);\n    const x2Num = Number.parseInt(x2, 10);\n    const y2Num = Number.parseInt(y2, 10);\n\n    // Use Math.floor to truncate and calculate center point\n    const x = Math.floor((x1Num + x2Num) / 2);\n    const y = Math.floor((y1Num + y2Num) / 2);\n\n    // Return formatted coordinate string\n    return `(${x},${y})`;\n  }\n\n  // Remove [EOS] and replace <bbox> coordinates\n  const cleanedText = text.replace(/\\[EOS\\]/g, '');\n  return cleanedText.replace(pattern, replaceMatch).trim();\n}\n\nfunction getPoint(startBox: string, size: { width: number; height: number }) {\n  const [x, y] = JSON.parse(startBox);\n  return [x * size.width, y * size.height];\n}\n\ninterface BaseAction {\n  action_type: ActionType;\n  action_inputs: Record<string, any>;\n  reflection: string | null;\n  thought: string | null;\n}\n\ninterface ClickAction extends BaseAction {\n  action_type: 'click';\n  action_inputs: {\n    start_box: string; // JSON string of [x, y] coordinates\n  };\n}\n\ninterface DragAction extends BaseAction {\n  action_type: 'drag';\n  action_inputs: {\n    start_box: string; // JSON string of [x, y] coordinates\n    end_box: string; // JSON string of [x, y] coordinates\n  };\n}\n\ninterface WaitAction extends BaseAction {\n  action_type: 'wait';\n  action_inputs: {\n    time: string; // JSON string of [x, y] coordinates\n  };\n}\n\ninterface TypeAction extends BaseAction {\n  action_type: 'type';\n  action_inputs: {\n    content: string;\n  };\n}\n\ninterface HotkeyAction extends BaseAction {\n  action_type: 'hotkey';\n  action_inputs: {\n    key: string;\n  };\n}\n\ninterface ScrollAction extends BaseAction {\n  action_type: 'scroll';\n  action_inputs: {\n    direction: 'up' | 'down';\n  };\n}\n\ninterface FinishedAction extends BaseAction {\n  action_type: 'finished';\n  action_inputs: Record<string, never>;\n}\n\nexport type Action =\n  | ClickAction\n  | DragAction\n  | TypeAction\n  | HotkeyAction\n  | ScrollAction\n  | FinishedAction\n  | WaitAction;\n\nexport async function resizeImageForUiTars(\n  imageBase64: string,\n  size: Size,\n  uiTarsVersion: UITarsModelVersion | undefined,\n) {\n  if (uiTarsVersion === UITarsModelVersion.V1_5) {\n    debug('ui-tars-v1.5, will check image size', size);\n    const currentPixels = size.width * size.height;\n    const maxPixels = 16384 * 28 * 28; //\n    if (currentPixels > maxPixels) {\n      const resizeFactor = Math.sqrt(maxPixels / currentPixels);\n      const newWidth = Math.floor(size.width * resizeFactor);\n      const newHeight = Math.floor(size.height * resizeFactor);\n      debug(\n        'resize image for ui-tars, new width: %s, new height: %s',\n        newWidth,\n        newHeight,\n      );\n      const resizedImage = await resizeImgBase64(imageBase64, {\n        width: newWidth,\n        height: newHeight,\n      });\n      return resizedImage;\n    }\n  }\n  return imageBase64;\n}\n"],"names":["debug","getDebug","bboxSize","pointToBbox","point","width","height","Math","uiTarsPlanning","userInstruction","options","conversationHistory","context","modelConfig","uiTarsModelVersion","systemPrompt","getUiTarsPlanningPrompt","imagePayload","resizeImageForUiTars","res","callAIWithStringResponse","AIActionType","convertedText","convertBboxToCoordinates","size","parsed","actionParser","JSON","transformActions","shouldContinue","action","actionType","assert","getPoint","startPoint","endPoint","keys","transformHotkeyInput","console","Error","log","getSummary","undefined","text","pattern","replaceMatch","match","x1","y1","x2","y2","x1Num","Number","y1Num","x2Num","y2Num","x","y","cleanedText","startBox","imageBase64","uiTarsVersion","UITarsModelVersion","currentPixels","maxPixels","resizeFactor","newWidth","newHeight","resizedImage","resizeImgBase64"],"mappings":";;;;;;;;;AAyBA,MAAMA,QAAQC,SAAS;AACvB,MAAMC,WAAW;AACjB,MAAMC,cAAc,CAClBC,OACAC,OACAC,SAEO;QACLC,KAAK,KAAK,CAACA,KAAK,GAAG,CAACH,MAAM,CAAC,GAAGF,WAAW,GAAG;QAC5CK,KAAK,KAAK,CAACA,KAAK,GAAG,CAACH,MAAM,CAAC,GAAGF,WAAW,GAAG;QAC5CK,KAAK,KAAK,CAACA,KAAK,GAAG,CAACH,MAAM,CAAC,GAAGF,WAAW,GAAGG;QAC5CE,KAAK,KAAK,CAACA,KAAK,GAAG,CAACH,MAAM,CAAC,GAAGF,WAAW,GAAGI;KAC7C;AAGI,eAAeE,eACpBC,eAAuB,EACvBC,OAIC;IAED,MAAM,EAAEC,mBAAmB,EAAEC,OAAO,EAAEC,WAAW,EAAE,GAAGH;IACtD,MAAM,EAAEI,kBAAkB,EAAE,GAAGD;IAC/B,MAAME,eAAeC,4BAA4BP;IAEjD,MAAMQ,eAAe,MAAMC,qBACzBN,QAAQ,gBAAgB,EACxBA,QAAQ,IAAI,EACZE;IAGFH,oBAAoB,MAAM,CAAC;QACzB,MAAM;QACN,SAAS;YACP;gBACE,MAAM;gBACN,WAAW;oBACT,KAAKM;gBACP;YACF;SACD;IACH;IAEA,MAAME,MAAM,MAAMC,yBAChB;QACE;YACE,MAAM;YACN,SAASL;QACX;WACGJ,oBAAoB,QAAQ;KAChC,EACDU,aAAa,eAAe,EAC5BR;IAEF,MAAMS,gBAAgBC,yBAAyBJ,IAAI,OAAO;IAE1D,MAAM,EAAEK,IAAI,EAAE,GAAGZ;IACjB,MAAM,EAAEa,MAAM,EAAE,GAAGC,aAAa;QAC9B,YAAYJ;QACZ,QAAQ;YAAC;YAAM;SAAK;QACpB,eAAe;YACb,OAAOE,KAAK,KAAK;YACjB,QAAQA,KAAK,MAAM;QACrB;QACA,UAAUV;IACZ;IAEAd,MACE,oBACAc,oBACA,YACAa,KAAK,SAAS,CAACF;IAGjB,MAAMG,mBAAqC,EAAE;IAC7C,IAAIC,iBAAiB;IACrBJ,OAAO,OAAO,CAAC,CAACK;QACd,MAAMC,aAAcD,AAAAA,CAAAA,OAAO,WAAW,IAAI,EAAC,EAAG,WAAW;QACzD,IAAIC,AAAe,YAAfA,YAAwB;YAC1BC,OAAOF,OAAO,aAAa,CAAC,SAAS,EAAE;YACvC,MAAM1B,QAAQ6B,SAASH,OAAO,aAAa,CAAC,SAAS,EAAEN;YACvDI,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO;oBACL,QAAQ;wBACN,QAAQE,OAAO,OAAO,IAAI;wBAC1B,MAAM3B,YACJ;4BAAE,GAAGC,KAAK,CAAC,EAAE;4BAAE,GAAGA,KAAK,CAAC,EAAE;wBAAC,GAC3BoB,KAAK,KAAK,EACVA,KAAK,MAAM;oBAEf;gBACF;YACF;QACF,OAAO,IAAIO,AAAe,WAAfA,YAAuB;YAChCC,OAAOF,OAAO,aAAa,CAAC,SAAS,EAAE;YACvCE,OAAOF,OAAO,aAAa,CAAC,OAAO,EAAE;YACrC,MAAMI,aAAaD,SAASH,OAAO,aAAa,CAAC,SAAS,EAAEN;YAC5D,MAAMW,WAAWF,SAASH,OAAO,aAAa,CAAC,OAAO,EAAEN;YACxDI,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO;oBACL,MAAM;wBACJ,QAAQE,OAAO,OAAO,IAAI;wBAC1B,MAAM3B,YACJ;4BAAE,GAAG+B,UAAU,CAAC,EAAE;4BAAE,GAAGA,UAAU,CAAC,EAAE;wBAAC,GACrCV,KAAK,KAAK,EACVA,KAAK,MAAM;oBAEf;oBACA,IAAI;wBACF,QAAQM,OAAO,OAAO,IAAI;wBAC1B,MAAM3B,YACJ;4BAAE,GAAGgC,QAAQ,CAAC,EAAE;4BAAE,GAAGA,QAAQ,CAAC,EAAE;wBAAC,GACjCX,KAAK,KAAK,EACVA,KAAK,MAAM;oBAEf;gBACF;gBACA,SAASM,OAAO,OAAO,IAAI;YAC7B;QACF,OAAO,IAAIC,AAAe,WAAfA,YACTH,iBAAiB,IAAI,CAAC;YACpB,MAAM;YACN,OAAO;gBACL,OAAOE,OAAO,aAAa,CAAC,OAAO;YACrC;YACA,SAASA,OAAO,OAAO,IAAI;QAC7B;aACK,IAAIC,AAAe,aAAfA,YACTH,iBAAiB,IAAI,CAAC;YACpB,MAAM;YACN,OAAO;gBACL,WAAWE,OAAO,aAAa,CAAC,SAAS;YAC3C;YACA,SAASA,OAAO,OAAO,IAAI;QAC7B;aACK,IAAIC,AAAe,eAAfA,YAA2B;YACpCF,iBAAiB;YACjBD,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO,CAAC;gBACR,SAASE,OAAO,OAAO,IAAI;YAC7B;QACF,OAAO,IAAIC,AAAe,aAAfA,YACT,IAAKD,OAAO,aAAa,CAAC,GAAG,EAItB;YACL,MAAMM,OAAOC,qBAAqBP,OAAO,aAAa,CAAC,GAAG;YAE1DF,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO;oBACL,SAASQ;gBACX;gBACA,SAASN,OAAO,OAAO,IAAI;YAC7B;QACF,OAbEQ,QAAQ,IAAI,CACV;aAaC,IAAIP,AAAe,WAAfA,YACTH,iBAAiB,IAAI,CAAC;YACpB,MAAM;YACN,OAAO;gBACL,QAAQ;YACV;YACA,SAASE,OAAO,OAAO,IAAI;QAC7B;IAEJ;IAEA,IAAIF,AAA4B,MAA5BA,iBAAiB,MAAM,EACzB,MAAM,IAAIW,MAAM,CAAC,4BAA4B,EAAEpB,IAAI,OAAO,EAAE,EAAE;QAC5D,OAAO;YACL,YAAYA,IAAI,OAAO;YACvBM;QACF;IACF;IAGFzB,MAAM,oBAAoB2B,KAAK,SAAS,CAACC,kBAAkB,MAAM;IACjE,MAAMY,MAAMC,WAAWtB,IAAI,OAAO;IAElCR,oBAAoB,MAAM,CAAC;QACzB,MAAM;QACN,SAAS6B;IACX;IAEA,OAAO;QACL,SAASZ;QACTY;QACA,OAAOrB,IAAI,KAAK;QAChB,aAAaQ,KAAK,SAAS,CAACR,IAAI,OAAO,EAAEuB,QAAW;QACpD,oCAAoCb;IACtC;AACF;AAOA,SAASN,yBAAyBoB,IAAY;IAE5C,MAAMC,UAAU;IAEhB,SAASC,aACPC,KAAa,EACbC,EAAU,EACVC,EAAU,EACVC,EAAU,EACVC,EAAU;QAGV,MAAMC,QAAQC,OAAO,QAAQ,CAACL,IAAI;QAClC,MAAMM,QAAQD,OAAO,QAAQ,CAACJ,IAAI;QAClC,MAAMM,QAAQF,OAAO,QAAQ,CAACH,IAAI;QAClC,MAAMM,QAAQH,OAAO,QAAQ,CAACF,IAAI;QAGlC,MAAMM,IAAIjD,KAAK,KAAK,CAAE4C,AAAAA,CAAAA,QAAQG,KAAI,IAAK;QACvC,MAAMG,IAAIlD,KAAK,KAAK,CAAE8C,AAAAA,CAAAA,QAAQE,KAAI,IAAK;QAGvC,OAAO,CAAC,CAAC,EAAEC,EAAE,CAAC,EAAEC,EAAE,CAAC,CAAC;IACtB;IAGA,MAAMC,cAAcf,KAAK,OAAO,CAAC,YAAY;IAC7C,OAAOe,YAAY,OAAO,CAACd,SAASC,cAAc,IAAI;AACxD;AAEA,SAASZ,SAAS0B,QAAgB,EAAEnC,IAAuC;IACzE,MAAM,CAACgC,GAAGC,EAAE,GAAG9B,KAAK,KAAK,CAACgC;IAC1B,OAAO;QAACH,IAAIhC,KAAK,KAAK;QAAEiC,IAAIjC,KAAK,MAAM;KAAC;AAC1C;AAkEO,eAAeN,qBACpB0C,WAAmB,EACnBpC,IAAU,EACVqC,aAA6C;IAE7C,IAAIA,kBAAkBC,mBAAmB,IAAI,EAAE;QAC7C9D,MAAM,uCAAuCwB;QAC7C,MAAMuC,gBAAgBvC,KAAK,KAAK,GAAGA,KAAK,MAAM;QAC9C,MAAMwC,YAAY;QAClB,IAAID,gBAAgBC,WAAW;YAC7B,MAAMC,eAAe1D,KAAK,IAAI,CAACyD,YAAYD;YAC3C,MAAMG,WAAW3D,KAAK,KAAK,CAACiB,KAAK,KAAK,GAAGyC;YACzC,MAAME,YAAY5D,KAAK,KAAK,CAACiB,KAAK,MAAM,GAAGyC;YAC3CjE,MACE,2DACAkE,UACAC;YAEF,MAAMC,eAAe,MAAMC,gBAAgBT,aAAa;gBACtD,OAAOM;gBACP,QAAQC;YACV;YACA,OAAOC;QACT;IACF;IACA,OAAOR;AACT"}
+{"version":3,"file":"ai-model/ui-tars-planning.mjs","sources":["../../../src/ai-model/ui-tars-planning.ts"],"sourcesContent":["import type {\n  PlanningAIResponse,\n  PlanningAction,\n  Size,\n  UIContext,\n} from '@/types';\nimport { type IModelConfig, UITarsModelVersion } from '@midscene/shared/env';\nimport { resizeImgBase64 } from '@midscene/shared/img';\nimport { getDebug } from '@midscene/shared/logger';\nimport { transformHotkeyInput } from '@midscene/shared/us-keyboard-layout';\nimport { assert } from '@midscene/shared/utils';\nimport { actionParser } from '@ui-tars/action-parser';\nimport { AIActionType } from '../common';\nimport type { ConversationHistory } from './conversation-history';\nimport { getSummary, getUiTarsPlanningPrompt } from './prompt/ui-tars-planning';\nimport { callAIWithStringResponse } from './service-caller/index';\ntype ActionType =\n  | 'click'\n  | 'left_double'\n  | 'right_single'\n  | 'drag'\n  | 'type'\n  | 'hotkey'\n  | 'finished'\n  | 'scroll'\n  | 'wait';\n\nconst debug = getDebug('ui-tars-planning');\nconst bboxSize = 10;\nconst pointToBbox = (\n  point: { x: number; y: number },\n  width: number,\n  height: number,\n): [number, number, number, number] => {\n  return [\n    Math.round(Math.max(point.x - bboxSize / 2, 0)),\n    Math.round(Math.max(point.y - bboxSize / 2, 0)),\n    Math.round(Math.min(point.x + bboxSize / 2, width)),\n    Math.round(Math.min(point.y + bboxSize / 2, height)),\n  ];\n};\n\nexport async function uiTarsPlanning(\n  userInstruction: string,\n  options: {\n    conversationHistory: ConversationHistory;\n    context: UIContext;\n    modelConfig: IModelConfig;\n    actionContext?: string;\n  },\n): Promise<PlanningAIResponse> {\n  const { conversationHistory, context, modelConfig, actionContext } = options;\n  const { uiTarsModelVersion } = modelConfig;\n\n  let instruction = userInstruction;\n  if (actionContext) {\n    instruction = `<high_priority_knowledge>${actionContext}</high_priority_knowledge>\\n<user_instruction>${userInstruction}</user_instruction>`;\n  }\n\n  const systemPrompt = getUiTarsPlanningPrompt() + instruction;\n\n  const imagePayload = await resizeImageForUiTars(\n    context.screenshotBase64,\n    context.size,\n    uiTarsModelVersion,\n  );\n\n  conversationHistory.append({\n    role: 'user',\n    content: [\n      {\n        type: 'image_url',\n        image_url: {\n          url: imagePayload,\n        },\n      },\n    ],\n  });\n\n  const res = await callAIWithStringResponse(\n    [\n      {\n        role: 'user',\n        content: systemPrompt,\n      },\n      ...conversationHistory.snapshot(),\n    ],\n    AIActionType.INSPECT_ELEMENT,\n    modelConfig,\n  );\n  const convertedText = convertBboxToCoordinates(res.content);\n\n  const { size } = context;\n  const { parsed } = actionParser({\n    prediction: convertedText,\n    factor: [1000, 1000],\n    screenContext: {\n      width: size.width,\n      height: size.height,\n    },\n    modelVer: uiTarsModelVersion,\n  });\n\n  debug(\n    'ui-tars modelVer',\n    uiTarsModelVersion,\n    ', parsed',\n    JSON.stringify(parsed),\n  );\n\n  const transformActions: PlanningAction[] = [];\n  const unhandledActions: Array<{ type: string; thought: string }> = [];\n  let shouldContinue = true;\n  parsed.forEach((action) => {\n    const actionType = (action.action_type || '').toLowerCase();\n    if (actionType === 'click') {\n      assert(action.action_inputs.start_box, 'start_box is required');\n      const point = getPoint(action.action_inputs.start_box, size);\n      transformActions.push({\n        type: 'Tap',\n        param: {\n          locate: {\n            prompt: action.thought || '',\n            bbox: pointToBbox(\n              { x: point[0], y: point[1] },\n              size.width,\n              size.height,\n            ),\n          },\n        },\n      });\n    } else if (actionType === 'left_double') {\n      assert(action.action_inputs.start_box, 'start_box is required');\n      const point = getPoint(action.action_inputs.start_box, size);\n      transformActions.push({\n        type: 'DoubleClick',\n        param: {\n          locate: {\n            prompt: action.thought || '',\n            bbox: pointToBbox(\n              { x: point[0], y: point[1] },\n              size.width,\n              size.height,\n            ),\n          },\n        },\n        thought: action.thought || '',\n      });\n    } else if (actionType === 'right_single') {\n      assert(action.action_inputs.start_box, 'start_box is required');\n      const point = getPoint(action.action_inputs.start_box, size);\n      transformActions.push({\n        type: 'RightClick',\n        param: {\n          locate: {\n            prompt: action.thought || '',\n            bbox: pointToBbox(\n              { x: point[0], y: point[1] },\n              size.width,\n              size.height,\n            ),\n          },\n        },\n        thought: action.thought || '',\n      });\n    } else if (actionType === 'drag') {\n      assert(action.action_inputs.start_box, 'start_box is required');\n      assert(action.action_inputs.end_box, 'end_box is required');\n      const startPoint = getPoint(action.action_inputs.start_box, size);\n      const endPoint = getPoint(action.action_inputs.end_box, size);\n      transformActions.push({\n        type: 'DragAndDrop',\n        param: {\n          from: {\n            prompt: action.thought || '',\n            bbox: pointToBbox(\n              { x: startPoint[0], y: startPoint[1] },\n              size.width,\n              size.height,\n            ),\n          },\n          to: {\n            prompt: action.thought || '',\n            bbox: pointToBbox(\n              { x: endPoint[0], y: endPoint[1] },\n              size.width,\n              size.height,\n            ),\n          },\n        },\n        thought: action.thought || '',\n      });\n    } else if (actionType === 'type') {\n      transformActions.push({\n        type: 'Input',\n        param: {\n          value: action.action_inputs.content,\n        },\n        thought: action.thought || '',\n      });\n    } else if (actionType === 'scroll') {\n      transformActions.push({\n        type: 'Scroll',\n        param: {\n          direction: action.action_inputs.direction,\n        },\n        thought: action.thought || '',\n      });\n    } else if (actionType === 'finished') {\n      shouldContinue = false;\n      transformActions.push({\n        type: 'Finished',\n        param: {},\n        thought: action.thought || '',\n      });\n    } else if (actionType === 'hotkey') {\n      if (!action.action_inputs.key) {\n        console.warn(\n          'No key found in action: hotkey. Will not perform action.',\n        );\n      } else {\n        const keys = transformHotkeyInput(action.action_inputs.key);\n\n        transformActions.push({\n          type: 'KeyboardPress',\n          param: {\n            keyName: keys,\n          },\n          thought: action.thought || '',\n        });\n      }\n    } else if (actionType === 'wait') {\n      transformActions.push({\n        type: 'Sleep',\n        param: {\n          timeMs: 1000,\n        },\n        thought: action.thought || '',\n      });\n    } else if (actionType) {\n      // Track unhandled action types\n      unhandledActions.push({\n        type: actionType,\n        thought: action.thought || '',\n      });\n      debug('Unhandled action type:', actionType, 'thought:', action.thought);\n    }\n  });\n\n  if (transformActions.length === 0) {\n    const errorDetails: string[] = [];\n\n    // Check if parsing failed\n    if (parsed.length === 0) {\n      errorDetails.push('Action parser returned no actions');\n\n      // Check if response has Thought but no Action\n      if (\n        res.content.includes('Thought:') &&\n        !res.content.includes('Action:')\n      ) {\n        errorDetails.push(\n          'Response contains \"Thought:\" but missing \"Action:\" line',\n        );\n      } else {\n        errorDetails.push('Response may be malformed or empty');\n      }\n    }\n\n    // Check if we have unhandled action types\n    if (unhandledActions.length > 0) {\n      const types = unhandledActions.map((a) => a.type).join(', ');\n      errorDetails.push(`Unhandled action types: ${types}`);\n    }\n\n    const errorMessage = [\n      'No actions found in UI-TARS response.',\n      ...errorDetails,\n      `\\nRaw response: ${res.content}`,\n    ].join('\\n');\n\n    throw new Error(errorMessage, {\n      cause: {\n        prediction: res.content,\n        parsed,\n        unhandledActions,\n        convertedText,\n      },\n    });\n  }\n\n  debug('transformActions', JSON.stringify(transformActions, null, 2));\n  const log = getSummary(res.content);\n\n  conversationHistory.append({\n    role: 'assistant',\n    content: log,\n  });\n\n  return {\n    actions: transformActions,\n    log,\n    usage: res.usage,\n    rawResponse: JSON.stringify(res.content, undefined, 2),\n    more_actions_needed_by_instruction: shouldContinue,\n  };\n}\n\n/**\n * Converts bounding box notation to coordinate points\n * @param text - The text containing bbox tags to be converted\n * @returns The text with bbox tags replaced by coordinate points\n */\nfunction convertBboxToCoordinates(text: string): string {\n  // Match the four numbers after <bbox>\n  const pattern = /<bbox>(\\d+)\\s+(\\d+)\\s+(\\d+)\\s+(\\d+)<\\/bbox>/g;\n\n  function replaceMatch(\n    match: string,\n    x1: string,\n    y1: string,\n    x2: string,\n    y2: string,\n  ): string {\n    // Convert strings to numbers and calculate center point\n    const x1Num = Number.parseInt(x1, 10);\n    const y1Num = Number.parseInt(y1, 10);\n    const x2Num = Number.parseInt(x2, 10);\n    const y2Num = Number.parseInt(y2, 10);\n\n    // Use Math.floor to truncate and calculate center point\n    const x = Math.floor((x1Num + x2Num) / 2);\n    const y = Math.floor((y1Num + y2Num) / 2);\n\n    // Return formatted coordinate string\n    return `(${x},${y})`;\n  }\n\n  // Remove [EOS] and replace <bbox> coordinates\n  const cleanedText = text.replace(/\\[EOS\\]/g, '');\n  return cleanedText.replace(pattern, replaceMatch).trim();\n}\n\nfunction getPoint(startBox: string, size: { width: number; height: number }) {\n  const [x, y] = JSON.parse(startBox);\n  return [x * size.width, y * size.height];\n}\n\ninterface BaseAction {\n  action_type: ActionType;\n  action_inputs: Record<string, any>;\n  reflection: string | null;\n  thought: string | null;\n}\n\ninterface ClickAction extends BaseAction {\n  action_type: 'click';\n  action_inputs: {\n    start_box: string; // JSON string of [x, y] coordinates\n  };\n}\n\ninterface DragAction extends BaseAction {\n  action_type: 'drag';\n  action_inputs: {\n    start_box: string; // JSON string of [x, y] coordinates\n    end_box: string; // JSON string of [x, y] coordinates\n  };\n}\n\ninterface WaitAction extends BaseAction {\n  action_type: 'wait';\n  action_inputs: {\n    time: string; // JSON string of [x, y] coordinates\n  };\n}\n\ninterface LeftDoubleAction extends BaseAction {\n  action_type: 'left_double';\n  action_inputs: {\n    start_box: string; // JSON string of [x, y] coordinates\n  };\n}\n\ninterface RightSingleAction extends BaseAction {\n  action_type: 'right_single';\n  action_inputs: {\n    start_box: string; // JSON string of [x, y] coordinates\n  };\n}\n\ninterface TypeAction extends BaseAction {\n  action_type: 'type';\n  action_inputs: {\n    content: string;\n  };\n}\n\ninterface HotkeyAction extends BaseAction {\n  action_type: 'hotkey';\n  action_inputs: {\n    key: string;\n  };\n}\n\ninterface ScrollAction extends BaseAction {\n  action_type: 'scroll';\n  action_inputs: {\n    direction: 'up' | 'down';\n  };\n}\n\ninterface FinishedAction extends BaseAction {\n  action_type: 'finished';\n  action_inputs: Record<string, never>;\n}\n\nexport type Action =\n  | ClickAction\n  | LeftDoubleAction\n  | RightSingleAction\n  | DragAction\n  | TypeAction\n  | HotkeyAction\n  | ScrollAction\n  | FinishedAction\n  | WaitAction;\n\nexport async function resizeImageForUiTars(\n  imageBase64: string,\n  size: Size,\n  uiTarsVersion: UITarsModelVersion | undefined,\n) {\n  if (uiTarsVersion === UITarsModelVersion.V1_5) {\n    debug('ui-tars-v1.5, will check image size', size);\n    const currentPixels = size.width * size.height;\n    const maxPixels = 16384 * 28 * 28; //\n    if (currentPixels > maxPixels) {\n      const resizeFactor = Math.sqrt(maxPixels / currentPixels);\n      const newWidth = Math.floor(size.width * resizeFactor);\n      const newHeight = Math.floor(size.height * resizeFactor);\n      debug(\n        'resize image for ui-tars, new width: %s, new height: %s',\n        newWidth,\n        newHeight,\n      );\n      const resizedImage = await resizeImgBase64(imageBase64, {\n        width: newWidth,\n        height: newHeight,\n      });\n      return resizedImage;\n    }\n  }\n  return imageBase64;\n}\n"],"names":["debug","getDebug","bboxSize","pointToBbox","point","width","height","Math","uiTarsPlanning","userInstruction","options","conversationHistory","context","modelConfig","actionContext","uiTarsModelVersion","instruction","systemPrompt","getUiTarsPlanningPrompt","imagePayload","resizeImageForUiTars","res","callAIWithStringResponse","AIActionType","convertedText","convertBboxToCoordinates","size","parsed","actionParser","JSON","transformActions","unhandledActions","shouldContinue","action","actionType","assert","getPoint","startPoint","endPoint","keys","transformHotkeyInput","console","errorDetails","types","a","errorMessage","Error","log","getSummary","undefined","text","pattern","replaceMatch","match","x1","y1","x2","y2","x1Num","Number","y1Num","x2Num","y2Num","x","y","cleanedText","startBox","imageBase64","uiTarsVersion","UITarsModelVersion","currentPixels","maxPixels","resizeFactor","newWidth","newHeight","resizedImage","resizeImgBase64"],"mappings":";;;;;;;;;AA2BA,MAAMA,QAAQC,SAAS;AACvB,MAAMC,WAAW;AACjB,MAAMC,cAAc,CAClBC,OACAC,OACAC,SAEO;QACLC,KAAK,KAAK,CAACA,KAAK,GAAG,CAACH,MAAM,CAAC,GAAGF,WAAW,GAAG;QAC5CK,KAAK,KAAK,CAACA,KAAK,GAAG,CAACH,MAAM,CAAC,GAAGF,WAAW,GAAG;QAC5CK,KAAK,KAAK,CAACA,KAAK,GAAG,CAACH,MAAM,CAAC,GAAGF,WAAW,GAAGG;QAC5CE,KAAK,KAAK,CAACA,KAAK,GAAG,CAACH,MAAM,CAAC,GAAGF,WAAW,GAAGI;KAC7C;AAGI,eAAeE,eACpBC,eAAuB,EACvBC,OAKC;IAED,MAAM,EAAEC,mBAAmB,EAAEC,OAAO,EAAEC,WAAW,EAAEC,aAAa,EAAE,GAAGJ;IACrE,MAAM,EAAEK,kBAAkB,EAAE,GAAGF;IAE/B,IAAIG,cAAcP;IAClB,IAAIK,eACFE,cAAc,CAAC,yBAAyB,EAAEF,cAAc,8CAA8C,EAAEL,gBAAgB,mBAAmB,CAAC;IAG9I,MAAMQ,eAAeC,4BAA4BF;IAEjD,MAAMG,eAAe,MAAMC,qBACzBR,QAAQ,gBAAgB,EACxBA,QAAQ,IAAI,EACZG;IAGFJ,oBAAoB,MAAM,CAAC;QACzB,MAAM;QACN,SAAS;YACP;gBACE,MAAM;gBACN,WAAW;oBACT,KAAKQ;gBACP;YACF;SACD;IACH;IAEA,MAAME,MAAM,MAAMC,yBAChB;QACE;YACE,MAAM;YACN,SAASL;QACX;WACGN,oBAAoB,QAAQ;KAChC,EACDY,aAAa,eAAe,EAC5BV;IAEF,MAAMW,gBAAgBC,yBAAyBJ,IAAI,OAAO;IAE1D,MAAM,EAAEK,IAAI,EAAE,GAAGd;IACjB,MAAM,EAAEe,MAAM,EAAE,GAAGC,aAAa;QAC9B,YAAYJ;QACZ,QAAQ;YAAC;YAAM;SAAK;QACpB,eAAe;YACb,OAAOE,KAAK,KAAK;YACjB,QAAQA,KAAK,MAAM;QACrB;QACA,UAAUX;IACZ;IAEAf,MACE,oBACAe,oBACA,YACAc,KAAK,SAAS,CAACF;IAGjB,MAAMG,mBAAqC,EAAE;IAC7C,MAAMC,mBAA6D,EAAE;IACrE,IAAIC,iBAAiB;IACrBL,OAAO,OAAO,CAAC,CAACM;QACd,MAAMC,aAAcD,AAAAA,CAAAA,OAAO,WAAW,IAAI,EAAC,EAAG,WAAW;QACzD,IAAIC,AAAe,YAAfA,YAAwB;YAC1BC,OAAOF,OAAO,aAAa,CAAC,SAAS,EAAE;YACvC,MAAM7B,QAAQgC,SAASH,OAAO,aAAa,CAAC,SAAS,EAAEP;YACvDI,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO;oBACL,QAAQ;wBACN,QAAQG,OAAO,OAAO,IAAI;wBAC1B,MAAM9B,YACJ;4BAAE,GAAGC,KAAK,CAAC,EAAE;4BAAE,GAAGA,KAAK,CAAC,EAAE;wBAAC,GAC3BsB,KAAK,KAAK,EACVA,KAAK,MAAM;oBAEf;gBACF;YACF;QACF,OAAO,IAAIQ,AAAe,kBAAfA,YAA8B;YACvCC,OAAOF,OAAO,aAAa,CAAC,SAAS,EAAE;YACvC,MAAM7B,QAAQgC,SAASH,OAAO,aAAa,CAAC,SAAS,EAAEP;YACvDI,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO;oBACL,QAAQ;wBACN,QAAQG,OAAO,OAAO,IAAI;wBAC1B,MAAM9B,YACJ;4BAAE,GAAGC,KAAK,CAAC,EAAE;4BAAE,GAAGA,KAAK,CAAC,EAAE;wBAAC,GAC3BsB,KAAK,KAAK,EACVA,KAAK,MAAM;oBAEf;gBACF;gBACA,SAASO,OAAO,OAAO,IAAI;YAC7B;QACF,OAAO,IAAIC,AAAe,mBAAfA,YAA+B;YACxCC,OAAOF,OAAO,aAAa,CAAC,SAAS,EAAE;YACvC,MAAM7B,QAAQgC,SAASH,OAAO,aAAa,CAAC,SAAS,EAAEP;YACvDI,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO;oBACL,QAAQ;wBACN,QAAQG,OAAO,OAAO,IAAI;wBAC1B,MAAM9B,YACJ;4BAAE,GAAGC,KAAK,CAAC,EAAE;4BAAE,GAAGA,KAAK,CAAC,EAAE;wBAAC,GAC3BsB,KAAK,KAAK,EACVA,KAAK,MAAM;oBAEf;gBACF;gBACA,SAASO,OAAO,OAAO,IAAI;YAC7B;QACF,OAAO,IAAIC,AAAe,WAAfA,YAAuB;YAChCC,OAAOF,OAAO,aAAa,CAAC,SAAS,EAAE;YACvCE,OAAOF,OAAO,aAAa,CAAC,OAAO,EAAE;YACrC,MAAMI,aAAaD,SAASH,OAAO,aAAa,CAAC,SAAS,EAAEP;YAC5D,MAAMY,WAAWF,SAASH,OAAO,aAAa,CAAC,OAAO,EAAEP;YACxDI,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO;oBACL,MAAM;wBACJ,QAAQG,OAAO,OAAO,IAAI;wBAC1B,MAAM9B,YACJ;4BAAE,GAAGkC,UAAU,CAAC,EAAE;4BAAE,GAAGA,UAAU,CAAC,EAAE;wBAAC,GACrCX,KAAK,KAAK,EACVA,KAAK,MAAM;oBAEf;oBACA,IAAI;wBACF,QAAQO,OAAO,OAAO,IAAI;wBAC1B,MAAM9B,YACJ;4BAAE,GAAGmC,QAAQ,CAAC,EAAE;4BAAE,GAAGA,QAAQ,CAAC,EAAE;wBAAC,GACjCZ,KAAK,KAAK,EACVA,KAAK,MAAM;oBAEf;gBACF;gBACA,SAASO,OAAO,OAAO,IAAI;YAC7B;QACF,OAAO,IAAIC,AAAe,WAAfA,YACTJ,iBAAiB,IAAI,CAAC;YACpB,MAAM;YACN,OAAO;gBACL,OAAOG,OAAO,aAAa,CAAC,OAAO;YACrC;YACA,SAASA,OAAO,OAAO,IAAI;QAC7B;aACK,IAAIC,AAAe,aAAfA,YACTJ,iBAAiB,IAAI,CAAC;YACpB,MAAM;YACN,OAAO;gBACL,WAAWG,OAAO,aAAa,CAAC,SAAS;YAC3C;YACA,SAASA,OAAO,OAAO,IAAI;QAC7B;aACK,IAAIC,AAAe,eAAfA,YAA2B;YACpCF,iBAAiB;YACjBF,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO,CAAC;gBACR,SAASG,OAAO,OAAO,IAAI;YAC7B;QACF,OAAO,IAAIC,AAAe,aAAfA,YACT,IAAKD,OAAO,aAAa,CAAC,GAAG,EAItB;YACL,MAAMM,OAAOC,qBAAqBP,OAAO,aAAa,CAAC,GAAG;YAE1DH,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO;oBACL,SAASS;gBACX;gBACA,SAASN,OAAO,OAAO,IAAI;YAC7B;QACF,OAbEQ,QAAQ,IAAI,CACV;aAaC,IAAIP,AAAe,WAAfA,YACTJ,iBAAiB,IAAI,CAAC;YACpB,MAAM;YACN,OAAO;gBACL,QAAQ;YACV;YACA,SAASG,OAAO,OAAO,IAAI;QAC7B;aACK,IAAIC,YAAY;YAErBH,iBAAiB,IAAI,CAAC;gBACpB,MAAMG;gBACN,SAASD,OAAO,OAAO,IAAI;YAC7B;YACAjC,MAAM,0BAA0BkC,YAAY,YAAYD,OAAO,OAAO;QACxE;IACF;IAEA,IAAIH,AAA4B,MAA5BA,iBAAiB,MAAM,EAAQ;QACjC,MAAMY,eAAyB,EAAE;QAGjC,IAAIf,AAAkB,MAAlBA,OAAO,MAAM,EAAQ;YACvBe,aAAa,IAAI,CAAC;YAGlB,IACErB,IAAI,OAAO,CAAC,QAAQ,CAAC,eACrB,CAACA,IAAI,OAAO,CAAC,QAAQ,CAAC,YAEtBqB,aAAa,IAAI,CACf;iBAGFA,aAAa,IAAI,CAAC;QAEtB;QAGA,IAAIX,iBAAiB,MAAM,GAAG,GAAG;YAC/B,MAAMY,QAAQZ,iBAAiB,GAAG,CAAC,CAACa,IAAMA,EAAE,IAAI,EAAE,IAAI,CAAC;YACvDF,aAAa,IAAI,CAAC,CAAC,wBAAwB,EAAEC,OAAO;QACtD;QAEA,MAAME,eAAe;YACnB;eACGH;YACH,CAAC,gBAAgB,EAAErB,IAAI,OAAO,EAAE;SACjC,CAAC,IAAI,CAAC;QAEP,MAAM,IAAIyB,MAAMD,cAAc;YAC5B,OAAO;gBACL,YAAYxB,IAAI,OAAO;gBACvBM;gBACAI;gBACAP;YACF;QACF;IACF;IAEAxB,MAAM,oBAAoB6B,KAAK,SAAS,CAACC,kBAAkB,MAAM;IACjE,MAAMiB,MAAMC,WAAW3B,IAAI,OAAO;IAElCV,oBAAoB,MAAM,CAAC;QACzB,MAAM;QACN,SAASoC;IACX;IAEA,OAAO;QACL,SAASjB;QACTiB;QACA,OAAO1B,IAAI,KAAK;QAChB,aAAaQ,KAAK,SAAS,CAACR,IAAI,OAAO,EAAE4B,QAAW;QACpD,oCAAoCjB;IACtC;AACF;AAOA,SAASP,yBAAyByB,IAAY;IAE5C,MAAMC,UAAU;IAEhB,SAASC,aACPC,KAAa,EACbC,EAAU,EACVC,EAAU,EACVC,EAAU,EACVC,EAAU;QAGV,MAAMC,QAAQC,OAAO,QAAQ,CAACL,IAAI;QAClC,MAAMM,QAAQD,OAAO,QAAQ,CAACJ,IAAI;QAClC,MAAMM,QAAQF,OAAO,QAAQ,CAACH,IAAI;QAClC,MAAMM,QAAQH,OAAO,QAAQ,CAACF,IAAI;QAGlC,MAAMM,IAAIxD,KAAK,KAAK,CAAEmD,AAAAA,CAAAA,QAAQG,KAAI,IAAK;QACvC,MAAMG,IAAIzD,KAAK,KAAK,CAAEqD,AAAAA,CAAAA,QAAQE,KAAI,IAAK;QAGvC,OAAO,CAAC,CAAC,EAAEC,EAAE,CAAC,EAAEC,EAAE,CAAC,CAAC;IACtB;IAGA,MAAMC,cAAcf,KAAK,OAAO,CAAC,YAAY;IAC7C,OAAOe,YAAY,OAAO,CAACd,SAASC,cAAc,IAAI;AACxD;AAEA,SAAShB,SAAS8B,QAAgB,EAAExC,IAAuC;IACzE,MAAM,CAACqC,GAAGC,EAAE,GAAGnC,KAAK,KAAK,CAACqC;IAC1B,OAAO;QAACH,IAAIrC,KAAK,KAAK;QAAEsC,IAAItC,KAAK,MAAM;KAAC;AAC1C;AAkFO,eAAeN,qBACpB+C,WAAmB,EACnBzC,IAAU,EACV0C,aAA6C;IAE7C,IAAIA,kBAAkBC,mBAAmB,IAAI,EAAE;QAC7CrE,MAAM,uCAAuC0B;QAC7C,MAAM4C,gBAAgB5C,KAAK,KAAK,GAAGA,KAAK,MAAM;QAC9C,MAAM6C,YAAY;QAClB,IAAID,gBAAgBC,WAAW;YAC7B,MAAMC,eAAejE,KAAK,IAAI,CAACgE,YAAYD;YAC3C,MAAMG,WAAWlE,KAAK,KAAK,CAACmB,KAAK,KAAK,GAAG8C;YACzC,MAAME,YAAYnE,KAAK,KAAK,CAACmB,KAAK,MAAM,GAAG8C;YAC3CxE,MACE,2DACAyE,UACAC;YAEF,MAAMC,eAAe,MAAMC,gBAAgBT,aAAa;gBACtD,OAAOM;gBACP,QAAQC;YACV;YACA,OAAOC;QACT;IACF;IACA,OAAOR;AACT"}