@midscene/core 0.28.2-beta-20250910072710.0 → 0.28.2

package/dist/es/ai-model/prompt/yaml-generator.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"ai-model/prompt/yaml-generator.mjs","sources":["webpack://@midscene/core/./src/ai-model/prompt/yaml-generator.ts"],"sourcesContent":["import type {\n  StreamingAIResponse,\n  StreamingCodeGenerationOptions,\n} from '@/types';\nimport { YAML_EXAMPLE_CODE } from '@midscene/shared/constants';\nimport type { IModelConfig } from '@midscene/shared/env';\nimport {\n  AIActionType,\n  type ChatCompletionMessageParam,\n  callAI,\n} from '../index';\n\n// Common interfaces for test generation (shared between YAML and Playwright)\nexport interface EventCounts {\n  navigation: number;\n  click: number;\n  input: number;\n  scroll: number;\n  total: number;\n}\n\nexport interface InputDescription {\n  description: string;\n  value: string;\n}\n\nexport interface ProcessedEvent {\n  type: string;\n  timestamp: number;\n  url?: string;\n  title?: string;\n  elementDescription?: string;\n  value?: string;\n  pageInfo?: any;\n  elementRect?: any;\n}\n\nexport interface EventSummary {\n  testName: string;\n  startUrl: string;\n  eventCounts: EventCounts;\n  urls: string[];\n  clickDescriptions: string[];\n  inputDescriptions: InputDescription[];\n  events: ProcessedEvent[];\n}\n\n// Common ChromeRecordedEvent interface\nexport interface ChromeRecordedEvent {\n  type: string;\n  timestamp: number;\n  url?: string;\n  title?: string;\n  elementDescription?: string;\n  value?: string;\n  pageInfo?: any;\n  elementRect?: any;\n  screenshotBefore?: string;\n  screenshotAfter?: string;\n  screenshotWithBox?: string;\n}\n\nexport interface YamlGenerationOptions {\n  testName?: string;\n  includeTimestamps?: boolean;\n  maxScreenshots?: number;\n  description?: string;\n}\n\nexport interface FilteredEvents {\n  navigationEvents: ChromeRecordedEvent[];\n  clickEvents: ChromeRecordedEvent[];\n  inputEvents: ChromeRecordedEvent[];\n  scrollEvents: ChromeRecordedEvent[];\n}\n\n// Common utility functions (shared between YAML and Playwright generators)\n\n/**\n * Get screenshots from events for LLM context\n */\nexport const getScreenshotsForLLM = (\n  events: ChromeRecordedEvent[],\n  maxScreenshots = 1,\n): string[] => {\n  // Find events with screenshots, prioritizing navigation and click events\n  const eventsWithScreenshots = events.filter(\n    (event) =>\n      event.screenshotBefore ||\n      event.screenshotAfter ||\n      event.screenshotWithBox,\n  );\n\n  // Sort them by priority (navigation first, then clicks, then others)\n  const sortedEvents = [...eventsWithScreenshots].sort((a, b) => {\n    if (a.type === 'navigation' && b.type !== 'navigation') return -1;\n    if (a.type !== 'navigation' && b.type === 'navigation') return 1;\n    if (a.type === 'click' && b.type !== 'click') return -1;\n    if (a.type !== 'click' && b.type === 'click') return 1;\n    return 0;\n  });\n\n  // Extract up to maxScreenshots screenshots\n  const screenshots: string[] = [];\n  for (const event of sortedEvents) {\n    // Prefer the most informative screenshot\n    const screenshot =\n      event.screenshotWithBox ||\n      event.screenshotAfter ||\n      event.screenshotBefore;\n    if (screenshot && !screenshots.includes(screenshot)) {\n      screenshots.push(screenshot);\n      if (screenshots.length >= maxScreenshots) break;\n    }\n  }\n\n  return screenshots;\n};\n\n/**\n * Filter events by type for easier processing\n */\nexport const filterEventsByType = (\n  events: ChromeRecordedEvent[],\n): FilteredEvents => {\n  return {\n    navigationEvents: events.filter((event) => event.type === 'navigation'),\n    clickEvents: events.filter((event) => event.type === 'click'),\n    inputEvents: events.filter((event) => event.type === 'input'),\n    scrollEvents: events.filter((event) => event.type === 'scroll'),\n  };\n};\n\n/**\n * Create event counts summary\n */\nexport const createEventCounts = (\n  filteredEvents: FilteredEvents,\n  totalEvents: number,\n): EventCounts => {\n  return {\n    navigation: filteredEvents.navigationEvents.length,\n    click: filteredEvents.clickEvents.length,\n    input: filteredEvents.inputEvents.length,\n    scroll: filteredEvents.scrollEvents.length,\n    total: totalEvents,\n  };\n};\n\n/**\n * Extract input descriptions from input events\n */\nexport const extractInputDescriptions = (\n  inputEvents: ChromeRecordedEvent[],\n): InputDescription[] => {\n  return inputEvents\n    .map((event) => ({\n      description: event.elementDescription || '',\n      value: event.value || '',\n    }))\n    .filter((item) => item.description && item.value);\n};\n\n/**\n * Process events for LLM consumption\n */\nexport const processEventsForLLM = (\n  events: ChromeRecordedEvent[],\n): ProcessedEvent[] => {\n  return events.map((event) => ({\n    type: event.type,\n    timestamp: event.timestamp,\n    url: event.url,\n    title: event.title,\n    elementDescription: event.elementDescription,\n    value: event.value,\n    pageInfo: event.pageInfo,\n    elementRect: event.elementRect,\n  }));\n};\n\n/**\n * Prepare comprehensive event summary for LLM\n */\nexport const prepareEventSummary = (\n  events: ChromeRecordedEvent[],\n  options: { testName?: string; maxScreenshots?: number } = {},\n): EventSummary => {\n  const filteredEvents = filterEventsByType(events);\n  const eventCounts = createEventCounts(filteredEvents, events.length);\n\n  // Extract useful information from events\n  const startUrl =\n    filteredEvents.navigationEvents.length > 0\n      ? filteredEvents.navigationEvents[0].url || ''\n      : '';\n\n  const clickDescriptions = filteredEvents.clickEvents\n    .map((event) => event.elementDescription)\n    .filter((desc): desc is string => Boolean(desc))\n    .slice(0, 10);\n\n  const inputDescriptions = extractInputDescriptions(\n    filteredEvents.inputEvents,\n  ).slice(0, 10);\n\n  const urls = filteredEvents.navigationEvents\n    .map((e) => e.url)\n    .filter((url): url is string => Boolean(url))\n    .slice(0, 5);\n\n  const processedEvents = processEventsForLLM(events);\n\n  return {\n    testName: options.testName || 'Automated test from recorded events',\n    startUrl,\n    eventCounts,\n    urls,\n    clickDescriptions,\n    inputDescriptions,\n    events: processedEvents,\n  };\n};\n\n/**\n * Create message content for LLM with optional screenshots\n */\nexport const createMessageContent = (\n  promptText: string,\n  screenshots: string[] = [],\n  includeScreenshots = true,\n) => {\n  const messageContent: any[] = [\n    {\n      type: 'text',\n      text: promptText,\n    },\n  ];\n\n  // Add screenshots if available and requested\n  if (includeScreenshots && screenshots.length > 0) {\n    messageContent.unshift({\n      type: 'text',\n      text: 'Here are screenshots from the recording session to help you understand the context:',\n    });\n\n    screenshots.forEach((screenshot) => {\n      messageContent.push({\n        type: 'image_url',\n        image_url: {\n          url: screenshot,\n        },\n      });\n    });\n  }\n\n  return messageContent;\n};\n\n/**\n * Validate events before processing\n */\nexport const validateEvents = (events: ChromeRecordedEvent[]): void => {\n  if (!events.length) {\n    throw new Error('No events provided for test generation');\n  }\n};\n\n// YAML-specific generation functions\n\n/**\n * Generates YAML test configuration from recorded events using AI\n */\nexport const generateYamlTest = async (\n  events: ChromeRecordedEvent[],\n  options: YamlGenerationOptions,\n  modelConfig: IModelConfig,\n): Promise<string> => {\n  try {\n    // Validate input\n    validateEvents(events);\n\n    // Prepare event summary using shared utilities\n    const summary = prepareEventSummary(events, {\n      testName: options.testName,\n      maxScreenshots: options.maxScreenshots || 3,\n    });\n\n    // Add YAML-specific options to summary\n    const yamlSummary = {\n      ...summary,\n      includeTimestamps: options.includeTimestamps || false,\n    };\n\n    // Get screenshots for visual context\n    const screenshots = getScreenshotsForLLM(\n      events,\n      options.maxScreenshots || 3,\n    );\n\n    // Use LLM to generate the YAML test configuration\n    const prompt: ChatCompletionMessageParam[] = [\n      {\n        role: 'system',\n        content: `You are an expert in Midscene.js YAML test generation. Generate clean, accurate YAML following these rules: ${YAML_EXAMPLE_CODE}`,\n      },\n      {\n        role: 'user',\n        content: `Generate YAML test for Midscene.js automation from recorded browser events.\n\nEvent Summary:\n${JSON.stringify(yamlSummary, null, 2)}\n\nConvert events:\n- navigation → target.url\n- click → aiTap with element description\n- input → aiInput with value and locate\n- scroll → aiScroll with appropriate direction\n- Add aiAssert for important state changes\n\nRespond with YAML only, no explanations.`,\n      },\n    ];\n\n    // Add screenshots if available and requested\n    if (screenshots.length > 0) {\n      prompt.push({\n        role: 'user',\n        content:\n          'Here are screenshots from the recording session to help you understand the context:',\n      });\n\n      prompt.push({\n        role: 'user',\n        content: screenshots.map((screenshot) => ({\n          type: 'image_url',\n          image_url: {\n            url: screenshot,\n          },\n        })),\n      });\n    }\n\n    const response = await callAI(\n      prompt,\n      AIActionType.EXTRACT_DATA,\n      modelConfig,\n    );\n\n    if (response?.content && typeof response.content === 'string') {\n      return response.content;\n    }\n\n    throw new Error('Failed to generate YAML test configuration');\n  } catch (error) {\n    throw new Error(`Failed to generate YAML test: ${error}`);\n  }\n};\n\n/**\n * Generates YAML test configuration from recorded events using AI with streaming support\n */\nexport const generateYamlTestStream = async (\n  events: ChromeRecordedEvent[],\n  options: YamlGenerationOptions & StreamingCodeGenerationOptions,\n  modelConfig: IModelConfig,\n): Promise<StreamingAIResponse> => {\n  try {\n    // Validate input\n    validateEvents(events);\n\n    // Prepare event summary using shared utilities\n    const summary = prepareEventSummary(events, {\n      testName: options.testName,\n      maxScreenshots: options.maxScreenshots || 3,\n    });\n\n    // Add YAML-specific options to summary\n    const yamlSummary = {\n      ...summary,\n      includeTimestamps: options.includeTimestamps || false,\n    };\n\n    // Get screenshots for visual context\n    const screenshots = getScreenshotsForLLM(\n      events,\n      options.maxScreenshots || 3,\n    );\n\n    // Use LLM to generate the YAML test configuration\n    const prompt: ChatCompletionMessageParam[] = [\n      {\n        role: 'system',\n        content: `You are an expert in Midscene.js YAML test generation. Generate clean, accurate YAML following these rules: ${YAML_EXAMPLE_CODE}`,\n      },\n      {\n        role: 'user',\n        content: `Generate YAML test for Midscene.js automation from recorded browser events.\n\nEvent Summary:\n${JSON.stringify(yamlSummary, null, 2)}\n\nConvert events:\n- navigation → target.url\n- click → aiTap with element description\n- input → aiInput with value and locate\n- scroll → aiScroll with appropriate direction\n- Add aiAssert for important state changes\n\nRespond with YAML only, no explanations.`,\n      },\n    ];\n\n    // Add screenshots if available and requested\n    if (screenshots.length > 0) {\n      prompt.push({\n        role: 'user',\n        content:\n          'Here are screenshots from the recording session to help you understand the context:',\n      });\n\n      prompt.push({\n        role: 'user',\n        content: screenshots.map((screenshot) => ({\n          type: 'image_url',\n          image_url: {\n            url: screenshot,\n          },\n        })),\n      });\n    }\n\n    if (options.stream && options.onChunk) {\n      // Use streaming\n      return await callAI(prompt, AIActionType.EXTRACT_DATA, modelConfig, {\n        stream: true,\n        onChunk: options.onChunk,\n      });\n    } else {\n      // Fallback to non-streaming\n      const response = await callAI(\n        prompt,\n        AIActionType.EXTRACT_DATA,\n        modelConfig,\n      );\n\n      if (response?.content && typeof response.content === 'string') {\n        return {\n          content: response.content,\n          usage: response.usage,\n          isStreamed: false,\n        };\n      }\n\n      throw new Error('Failed to generate YAML test configuration');\n    }\n  } catch (error) {\n    throw new Error(`Failed to generate YAML test: ${error}`);\n  }\n};\n"],"names":["getScreenshotsForLLM","events","maxScreenshots","eventsWithScreenshots","event","sortedEvents","a","b","screenshots","screenshot","filterEventsByType","createEventCounts","filteredEvents","totalEvents","extractInputDescriptions","inputEvents","item","processEventsForLLM","prepareEventSummary","options","eventCounts","startUrl","clickDescriptions","desc","Boolean","inputDescriptions","urls","e","url","processedEvents","createMessageContent","promptText","includeScreenshots","messageContent","validateEvents","Error","generateYamlTest","modelConfig","summary","yamlSummary","prompt","YAML_EXAMPLE_CODE","JSON","response","callAI","AIActionType","error","generateYamlTestStream"],"mappings":";;AAiFO,MAAMA,uBAAuB,CAClCC,QACAC,iBAAiB,CAAC;IAGlB,MAAMC,wBAAwBF,OAAO,MAAM,CACzC,CAACG,QACCA,MAAM,gBAAgB,IACtBA,MAAM,eAAe,IACrBA,MAAM,iBAAiB;IAI3B,MAAMC,eAAe;WAAIF;KAAsB,CAAC,IAAI,CAAC,CAACG,GAAGC;QACvD,IAAID,AAAW,iBAAXA,EAAE,IAAI,IAAqBC,AAAW,iBAAXA,EAAE,IAAI,EAAmB,OAAO;QAC/D,IAAID,AAAW,iBAAXA,EAAE,IAAI,IAAqBC,AAAW,iBAAXA,EAAE,IAAI,EAAmB,OAAO;QAC/D,IAAID,AAAW,YAAXA,EAAE,IAAI,IAAgBC,AAAW,YAAXA,EAAE,IAAI,EAAc,OAAO;QACrD,IAAID,AAAW,YAAXA,EAAE,IAAI,IAAgBC,AAAW,YAAXA,EAAE,IAAI,EAAc,OAAO;QACrD,OAAO;IACT;IAGA,MAAMC,cAAwB,EAAE;IAChC,KAAK,MAAMJ,SAASC,aAAc;QAEhC,MAAMI,aACJL,MAAM,iBAAiB,IACvBA,MAAM,eAAe,IACrBA,MAAM,gBAAgB;QACxB,IAAIK,cAAc,CAACD,YAAY,QAAQ,CAACC,aAAa;YACnDD,YAAY,IAAI,CAACC;YACjB,IAAID,YAAY,MAAM,IAAIN,gBAAgB;QAC5C;IACF;IAEA,OAAOM;AACT;AAKO,MAAME,qBAAqB,CAChCT,SAEO;QACL,kBAAkBA,OAAO,MAAM,CAAC,CAACG,QAAUA,AAAe,iBAAfA,MAAM,IAAI;QACrD,aAAaH,OAAO,MAAM,CAAC,CAACG,QAAUA,AAAe,YAAfA,MAAM,IAAI;QAChD,aAAaH,OAAO,MAAM,CAAC,CAACG,QAAUA,AAAe,YAAfA,MAAM,IAAI;QAChD,cAAcH,OAAO,MAAM,CAAC,CAACG,QAAUA,AAAe,aAAfA,MAAM,IAAI;IACnD;AAMK,MAAMO,oBAAoB,CAC/BC,gBACAC,cAEO;QACL,YAAYD,eAAe,gBAAgB,CAAC,MAAM;QAClD,OAAOA,eAAe,WAAW,CAAC,MAAM;QACxC,OAAOA,eAAe,WAAW,CAAC,MAAM;QACxC,QAAQA,eAAe,YAAY,CAAC,MAAM;QAC1C,OAAOC;IACT;AAMK,MAAMC,2BAA2B,CACtCC,cAEOA,YACJ,GAAG,CAAC,CAACX,QAAW;YACf,aAAaA,MAAM,kBAAkB,IAAI;YACzC,OAAOA,MAAM,KAAK,IAAI;QACxB,IACC,MAAM,CAAC,CAACY,OAASA,KAAK,WAAW,IAAIA,KAAK,KAAK;AAM7C,MAAMC,sBAAsB,CACjChB,SAEOA,OAAO,GAAG,CAAC,CAACG,QAAW;YAC5B,MAAMA,MAAM,IAAI;YAChB,WAAWA,MAAM,SAAS;YAC1B,KAAKA,MAAM,GAAG;YACd,OAAOA,MAAM,KAAK;YAClB,oBAAoBA,MAAM,kBAAkB;YAC5C,OAAOA,MAAM,KAAK;YAClB,UAAUA,MAAM,QAAQ;YACxB,aAAaA,MAAM,WAAW;QAChC;AAMK,MAAMc,sBAAsB,CACjCjB,QACAkB,UAA0D,CAAC,CAAC;IAE5D,MAAMP,iBAAiBF,mBAAmBT;IAC1C,MAAMmB,cAAcT,kBAAkBC,gBAAgBX,OAAO,MAAM;IAGnE,MAAMoB,WACJT,eAAe,gBAAgB,CAAC,MAAM,GAAG,IACrCA,eAAe,gBAAgB,CAAC,EAAE,CAAC,GAAG,IAAI,KAC1C;IAEN,MAAMU,oBAAoBV,eAAe,WAAW,CACjD,GAAG,CAAC,CAACR,QAAUA,MAAM,kBAAkB,EACvC,MAAM,CAAC,CAACmB,OAAyBC,QAAQD,OACzC,KAAK,CAAC,GAAG;IAEZ,MAAME,oBAAoBX,yBACxBF,eAAe,WAAW,EAC1B,KAAK,CAAC,GAAG;IAEX,MAAMc,OAAOd,eAAe,gBAAgB,CACzC,GAAG,CAAC,CAACe,IAAMA,EAAE,GAAG,EAChB,MAAM,CAAC,CAACC,MAAuBJ,QAAQI,MACvC,KAAK,CAAC,GAAG;IAEZ,MAAMC,kBAAkBZ,oBAAoBhB;IAE5C,OAAO;QACL,UAAUkB,QAAQ,QAAQ,IAAI;QAC9BE;QACAD;QACAM;QACAJ;QACAG;QACA,QAAQI;IACV;AACF;AAKO,MAAMC,uBAAuB,CAClCC,YACAvB,cAAwB,EAAE,EAC1BwB,qBAAqB,IAAI;IAEzB,MAAMC,iBAAwB;QAC5B;YACE,MAAM;YACN,MAAMF;QACR;KACD;IAGD,IAAIC,sBAAsBxB,YAAY,MAAM,GAAG,GAAG;QAChDyB,eAAe,OAAO,CAAC;YACrB,MAAM;YACN,MAAM;QACR;QAEAzB,YAAY,OAAO,CAAC,CAACC;YACnBwB,eAAe,IAAI,CAAC;gBAClB,MAAM;gBACN,WAAW;oBACT,KAAKxB;gBACP;YACF;QACF;IACF;IAEA,OAAOwB;AACT;AAKO,MAAMC,iBAAiB,CAACjC;IAC7B,IAAI,CAACA,OAAO,MAAM,EAChB,MAAM,IAAIkC,MAAM;AAEpB;AAOO,MAAMC,mBAAmB,OAC9BnC,QACAkB,SACAkB;IAEA,IAAI;QAEFH,eAAejC;QAGf,MAAMqC,UAAUpB,oBAAoBjB,QAAQ;YAC1C,UAAUkB,QAAQ,QAAQ;YAC1B,gBAAgBA,QAAQ,cAAc,IAAI;QAC5C;QAGA,MAAMoB,cAAc;YAClB,GAAGD,OAAO;YACV,mBAAmBnB,QAAQ,iBAAiB,IAAI;QAClD;QAGA,MAAMX,cAAcR,qBAClBC,QACAkB,QAAQ,cAAc,IAAI;QAI5B,MAAMqB,SAAuC;YAC3C;gBACE,MAAM;gBACN,SAAS,CAAC,4GAA4G,EAAEC,mBAAmB;YAC7I;YACA;gBACE,MAAM;gBACN,SAAS,CAAC;;;AAGlB,EAAEC,KAAK,SAAS,CAACH,aAAa,MAAM,GAAG;;;;;;;;;wCASC,CAAC;YACnC;SACD;QAGD,IAAI/B,YAAY,MAAM,GAAG,GAAG;YAC1BgC,OAAO,IAAI,CAAC;gBACV,MAAM;gBACN,SACE;YACJ;YAEAA,OAAO,IAAI,CAAC;gBACV,MAAM;gBACN,SAAShC,YAAY,GAAG,CAAC,CAACC,aAAgB;wBACxC,MAAM;wBACN,WAAW;4BACT,KAAKA;wBACP;oBACF;YACF;QACF;QAEA,MAAMkC,WAAW,MAAMC,OACrBJ,QACAK,aAAa,YAAY,EACzBR;QAGF,IAAIM,AAAAA,CAAAA,QAAAA,WAAAA,KAAAA,IAAAA,SAAU,OAAO,AAAD,KAAK,AAA4B,YAA5B,OAAOA,SAAS,OAAO,EAC9C,OAAOA,SAAS,OAAO;QAGzB,MAAM,IAAIR,MAAM;IAClB,EAAE,OAAOW,OAAO;QACd,MAAM,IAAIX,MAAM,CAAC,8BAA8B,EAAEW,OAAO;IAC1D;AACF;AAKO,MAAMC,yBAAyB,OACpC9C,QACAkB,SACAkB;IAEA,IAAI;QAEFH,eAAejC;QAGf,MAAMqC,UAAUpB,oBAAoBjB,QAAQ;YAC1C,UAAUkB,QAAQ,QAAQ;YAC1B,gBAAgBA,QAAQ,cAAc,IAAI;QAC5C;QAGA,MAAMoB,cAAc;YAClB,GAAGD,OAAO;YACV,mBAAmBnB,QAAQ,iBAAiB,IAAI;QAClD;QAGA,MAAMX,cAAcR,qBAClBC,QACAkB,QAAQ,cAAc,IAAI;QAI5B,MAAMqB,SAAuC;YAC3C;gBACE,MAAM;gBACN,SAAS,CAAC,4GAA4G,EAAEC,mBAAmB;YAC7I;YACA;gBACE,MAAM;gBACN,SAAS,CAAC;;;AAGlB,EAAEC,KAAK,SAAS,CAACH,aAAa,MAAM,GAAG;;;;;;;;;wCASC,CAAC;YACnC;SACD;QAGD,IAAI/B,YAAY,MAAM,GAAG,GAAG;YAC1BgC,OAAO,IAAI,CAAC;gBACV,MAAM;gBACN,SACE;YACJ;YAEAA,OAAO,IAAI,CAAC;gBACV,MAAM;gBACN,SAAShC,YAAY,GAAG,CAAC,CAACC,aAAgB;wBACxC,MAAM;wBACN,WAAW;4BACT,KAAKA;wBACP;oBACF;YACF;QACF;QAEA,IAAIU,QAAQ,MAAM,IAAIA,QAAQ,OAAO,EAEnC,OAAO,MAAMyB,OAAOJ,QAAQK,aAAa,YAAY,EAAER,aAAa;YAClE,QAAQ;YACR,SAASlB,QAAQ,OAAO;QAC1B;QACK;YAEL,MAAMwB,WAAW,MAAMC,OACrBJ,QACAK,aAAa,YAAY,EACzBR;YAGF,IAAIM,AAAAA,CAAAA,QAAAA,WAAAA,KAAAA,IAAAA,SAAU,OAAO,AAAD,KAAK,AAA4B,YAA5B,OAAOA,SAAS,OAAO,EAC9C,OAAO;gBACL,SAASA,SAAS,OAAO;gBACzB,OAAOA,SAAS,KAAK;gBACrB,YAAY;YACd;YAGF,MAAM,IAAIR,MAAM;QAClB;IACF,EAAE,OAAOW,OAAO;QACd,MAAM,IAAIX,MAAM,CAAC,8BAA8B,EAAEW,OAAO;IAC1D;AACF"}
+{"version":3,"file":"ai-model/prompt/yaml-generator.mjs","sources":["webpack://@midscene/core/./src/ai-model/prompt/yaml-generator.ts"],"sourcesContent":["import type {\n  StreamingAIResponse,\n  StreamingCodeGenerationOptions,\n} from '@/types';\nimport { YAML_EXAMPLE_CODE } from '@midscene/shared/constants';\nimport {\n  AIActionType,\n  type ChatCompletionMessageParam,\n  callAi,\n} from '../index';\n\n// Common interfaces for test generation (shared between YAML and Playwright)\nexport interface EventCounts {\n  navigation: number;\n  click: number;\n  input: number;\n  scroll: number;\n  total: number;\n}\n\nexport interface InputDescription {\n  description: string;\n  value: string;\n}\n\nexport interface ProcessedEvent {\n  type: string;\n  timestamp: number;\n  url?: string;\n  title?: string;\n  elementDescription?: string;\n  value?: string;\n  pageInfo?: any;\n  elementRect?: any;\n}\n\nexport interface EventSummary {\n  testName: string;\n  startUrl: string;\n  eventCounts: EventCounts;\n  urls: string[];\n  clickDescriptions: string[];\n  inputDescriptions: InputDescription[];\n  events: ProcessedEvent[];\n}\n\n// Common ChromeRecordedEvent interface\nexport interface ChromeRecordedEvent {\n  type: string;\n  timestamp: number;\n  url?: string;\n  title?: string;\n  elementDescription?: string;\n  value?: string;\n  pageInfo?: any;\n  elementRect?: any;\n  screenshotBefore?: string;\n  screenshotAfter?: string;\n  screenshotWithBox?: string;\n}\n\nexport interface YamlGenerationOptions {\n  testName?: string;\n  includeTimestamps?: boolean;\n  maxScreenshots?: number;\n  description?: string;\n}\n\nexport interface FilteredEvents {\n  navigationEvents: ChromeRecordedEvent[];\n  clickEvents: ChromeRecordedEvent[];\n  inputEvents: ChromeRecordedEvent[];\n  scrollEvents: ChromeRecordedEvent[];\n}\n\n// Common utility functions (shared between YAML and Playwright generators)\n\n/**\n * Get screenshots from events for LLM context\n */\nexport const getScreenshotsForLLM = (\n  events: ChromeRecordedEvent[],\n  maxScreenshots = 1,\n): string[] => {\n  // Find events with screenshots, prioritizing navigation and click events\n  const eventsWithScreenshots = events.filter(\n    (event) =>\n      event.screenshotBefore ||\n      event.screenshotAfter ||\n      event.screenshotWithBox,\n  );\n\n  // Sort them by priority (navigation first, then clicks, then others)\n  const sortedEvents = [...eventsWithScreenshots].sort((a, b) => {\n    if (a.type === 'navigation' && b.type !== 'navigation') return -1;\n    if (a.type !== 'navigation' && b.type === 'navigation') return 1;\n    if (a.type === 'click' && b.type !== 'click') return -1;\n    if (a.type !== 'click' && b.type === 'click') return 1;\n    return 0;\n  });\n\n  // Extract up to maxScreenshots screenshots\n  const screenshots: string[] = [];\n  for (const event of sortedEvents) {\n    // Prefer the most informative screenshot\n    const screenshot =\n      event.screenshotWithBox ||\n      event.screenshotAfter ||\n      event.screenshotBefore;\n    if (screenshot && !screenshots.includes(screenshot)) {\n      screenshots.push(screenshot);\n      if (screenshots.length >= maxScreenshots) break;\n    }\n  }\n\n  return screenshots;\n};\n\n/**\n * Filter events by type for easier processing\n */\nexport const filterEventsByType = (\n  events: ChromeRecordedEvent[],\n): FilteredEvents => {\n  return {\n    navigationEvents: events.filter((event) => event.type === 'navigation'),\n    clickEvents: events.filter((event) => event.type === 'click'),\n    inputEvents: events.filter((event) => event.type === 'input'),\n    scrollEvents: events.filter((event) => event.type === 'scroll'),\n  };\n};\n\n/**\n * Create event counts summary\n */\nexport const createEventCounts = (\n  filteredEvents: FilteredEvents,\n  totalEvents: number,\n): EventCounts => {\n  return {\n    navigation: filteredEvents.navigationEvents.length,\n    click: filteredEvents.clickEvents.length,\n    input: filteredEvents.inputEvents.length,\n    scroll: filteredEvents.scrollEvents.length,\n    total: totalEvents,\n  };\n};\n\n/**\n * Extract input descriptions from input events\n */\nexport const extractInputDescriptions = (\n  inputEvents: ChromeRecordedEvent[],\n): InputDescription[] => {\n  return inputEvents\n    .map((event) => ({\n      description: event.elementDescription || '',\n      value: event.value || '',\n    }))\n    .filter((item) => item.description && item.value);\n};\n\n/**\n * Process events for LLM consumption\n */\nexport const processEventsForLLM = (\n  events: ChromeRecordedEvent[],\n): ProcessedEvent[] => {\n  return events.map((event) => ({\n    type: event.type,\n    timestamp: event.timestamp,\n    url: event.url,\n    title: event.title,\n    elementDescription: event.elementDescription,\n    value: event.value,\n    pageInfo: event.pageInfo,\n    elementRect: event.elementRect,\n  }));\n};\n\n/**\n * Prepare comprehensive event summary for LLM\n */\nexport const prepareEventSummary = (\n  events: ChromeRecordedEvent[],\n  options: { testName?: string; maxScreenshots?: number } = {},\n): EventSummary => {\n  const filteredEvents = filterEventsByType(events);\n  const eventCounts = createEventCounts(filteredEvents, events.length);\n\n  // Extract useful information from events\n  const startUrl =\n    filteredEvents.navigationEvents.length > 0\n      ? filteredEvents.navigationEvents[0].url || ''\n      : '';\n\n  const clickDescriptions = filteredEvents.clickEvents\n    .map((event) => event.elementDescription)\n    .filter((desc): desc is string => Boolean(desc))\n    .slice(0, 10);\n\n  const inputDescriptions = extractInputDescriptions(\n    filteredEvents.inputEvents,\n  ).slice(0, 10);\n\n  const urls = filteredEvents.navigationEvents\n    .map((e) => e.url)\n    .filter((url): url is string => Boolean(url))\n    .slice(0, 5);\n\n  const processedEvents = processEventsForLLM(events);\n\n  return {\n    testName: options.testName || 'Automated test from recorded events',\n    startUrl,\n    eventCounts,\n    urls,\n    clickDescriptions,\n    inputDescriptions,\n    events: processedEvents,\n  };\n};\n\n/**\n * Create message content for LLM with optional screenshots\n */\nexport const createMessageContent = (\n  promptText: string,\n  screenshots: string[] = [],\n  includeScreenshots = true,\n) => {\n  const messageContent: any[] = [\n    {\n      type: 'text',\n      text: promptText,\n    },\n  ];\n\n  // Add screenshots if available and requested\n  if (includeScreenshots && screenshots.length > 0) {\n    messageContent.unshift({\n      type: 'text',\n      text: 'Here are screenshots from the recording session to help you understand the context:',\n    });\n\n    screenshots.forEach((screenshot) => {\n      messageContent.push({\n        type: 'image_url',\n        image_url: {\n          url: screenshot,\n        },\n      });\n    });\n  }\n\n  return messageContent;\n};\n\n/**\n * Validate events before processing\n */\nexport const validateEvents = (events: ChromeRecordedEvent[]): void => {\n  if (!events.length) {\n    throw new Error('No events provided for test generation');\n  }\n};\n\n// YAML-specific generation functions\n\n/**\n * Generates YAML test configuration from recorded events using AI\n */\nexport const generateYamlTest = async (\n  events: ChromeRecordedEvent[],\n  options: YamlGenerationOptions = {},\n): Promise<string> => {\n  try {\n    // Validate input\n    validateEvents(events);\n\n    // Prepare event summary using shared utilities\n    const summary = prepareEventSummary(events, {\n      testName: options.testName,\n      maxScreenshots: options.maxScreenshots || 3,\n    });\n\n    // Add YAML-specific options to summary\n    const yamlSummary = {\n      ...summary,\n      includeTimestamps: options.includeTimestamps || false,\n    };\n\n    // Get screenshots for visual context\n    const screenshots = getScreenshotsForLLM(\n      events,\n      options.maxScreenshots || 3,\n    );\n\n    // Use LLM to generate the YAML test configuration\n    const prompt: ChatCompletionMessageParam[] = [\n      {\n        role: 'system',\n        content: `You are an expert in Midscene.js YAML test generation. Generate clean, accurate YAML following these rules: ${YAML_EXAMPLE_CODE}`,\n      },\n      {\n        role: 'user',\n        content: `Generate YAML test for Midscene.js automation from recorded browser events.\n\nEvent Summary:\n${JSON.stringify(yamlSummary, null, 2)}\n\nConvert events:\n- navigation → target.url\n- click → aiTap with element description\n- input → aiInput with value and locate\n- scroll → aiScroll with appropriate direction\n- Add aiAssert for important state changes\n\nRespond with YAML only, no explanations.`,\n      },\n    ];\n\n    // Add screenshots if available and requested\n    if (screenshots.length > 0) {\n      prompt.push({\n        role: 'user',\n        content:\n          'Here are screenshots from the recording session to help you understand the context:',\n      });\n\n      prompt.push({\n        role: 'user',\n        content: screenshots.map((screenshot) => ({\n          type: 'image_url',\n          image_url: {\n            url: screenshot,\n          },\n        })),\n      });\n    }\n\n    const response = await callAi(prompt, AIActionType.EXTRACT_DATA, {\n      intent: 'default',\n    });\n\n    if (response?.content && typeof response.content === 'string') {\n      return response.content;\n    }\n\n    throw new Error('Failed to generate YAML test configuration');\n  } catch (error) {\n    throw new Error(`Failed to generate YAML test: ${error}`);\n  }\n};\n\n/**\n * Generates YAML test configuration from recorded events using AI with streaming support\n */\nexport const generateYamlTestStream = async (\n  events: ChromeRecordedEvent[],\n  options: YamlGenerationOptions & StreamingCodeGenerationOptions = {},\n): Promise<StreamingAIResponse> => {\n  try {\n    // Validate input\n    validateEvents(events);\n\n    // Prepare event summary using shared utilities\n    const summary = prepareEventSummary(events, {\n      testName: options.testName,\n      maxScreenshots: options.maxScreenshots || 3,\n    });\n\n    // Add YAML-specific options to summary\n    const yamlSummary = {\n      ...summary,\n      includeTimestamps: options.includeTimestamps || false,\n    };\n\n    // Get screenshots for visual context\n    const screenshots = getScreenshotsForLLM(\n      events,\n      options.maxScreenshots || 3,\n    );\n\n    // Use LLM to generate the YAML test configuration\n    const prompt: ChatCompletionMessageParam[] = [\n      {\n        role: 'system',\n        content: `You are an expert in Midscene.js YAML test generation. Generate clean, accurate YAML following these rules: ${YAML_EXAMPLE_CODE}`,\n      },\n      {\n        role: 'user',\n        content: `Generate YAML test for Midscene.js automation from recorded browser events.\n\nEvent Summary:\n${JSON.stringify(yamlSummary, null, 2)}\n\nConvert events:\n- navigation → target.url\n- click → aiTap with element description\n- input → aiInput with value and locate\n- scroll → aiScroll with appropriate direction\n- Add aiAssert for important state changes\n\nRespond with YAML only, no explanations.`,\n      },\n    ];\n\n    // Add screenshots if available and requested\n    if (screenshots.length > 0) {\n      prompt.push({\n        role: 'user',\n        content:\n          'Here are screenshots from the recording session to help you understand the context:',\n      });\n\n      prompt.push({\n        role: 'user',\n        content: screenshots.map((screenshot) => ({\n          type: 'image_url',\n          image_url: {\n            url: screenshot,\n          },\n        })),\n      });\n    }\n\n    if (options.stream && options.onChunk) {\n      // Use streaming\n      return await callAi(\n        prompt,\n        AIActionType.EXTRACT_DATA,\n        {\n          intent: 'default',\n        },\n        {\n          stream: true,\n          onChunk: options.onChunk,\n        },\n      );\n    } else {\n      // Fallback to non-streaming\n      const response = await callAi(prompt, AIActionType.EXTRACT_DATA, {\n        intent: 'default',\n      });\n\n      if (response?.content && typeof response.content === 'string') {\n        return {\n          content: response.content,\n          usage: response.usage,\n          isStreamed: false,\n        };\n      }\n\n      throw new Error('Failed to generate YAML test configuration');\n    }\n  } catch (error) {\n    throw new Error(`Failed to generate YAML test: ${error}`);\n  }\n};\n"],"names":["getScreenshotsForLLM","events","maxScreenshots","eventsWithScreenshots","event","sortedEvents","a","b","screenshots","screenshot","filterEventsByType","createEventCounts","filteredEvents","totalEvents","extractInputDescriptions","inputEvents","item","processEventsForLLM","prepareEventSummary","options","eventCounts","startUrl","clickDescriptions","desc","Boolean","inputDescriptions","urls","e","url","processedEvents","createMessageContent","promptText","includeScreenshots","messageContent","validateEvents","Error","generateYamlTest","summary","yamlSummary","prompt","YAML_EXAMPLE_CODE","JSON","response","callAi","AIActionType","error","generateYamlTestStream"],"mappings":";;AAgFO,MAAMA,uBAAuB,CAClCC,QACAC,iBAAiB,CAAC;IAGlB,MAAMC,wBAAwBF,OAAO,MAAM,CACzC,CAACG,QACCA,MAAM,gBAAgB,IACtBA,MAAM,eAAe,IACrBA,MAAM,iBAAiB;IAI3B,MAAMC,eAAe;WAAIF;KAAsB,CAAC,IAAI,CAAC,CAACG,GAAGC;QACvD,IAAID,AAAW,iBAAXA,EAAE,IAAI,IAAqBC,AAAW,iBAAXA,EAAE,IAAI,EAAmB,OAAO;QAC/D,IAAID,AAAW,iBAAXA,EAAE,IAAI,IAAqBC,AAAW,iBAAXA,EAAE,IAAI,EAAmB,OAAO;QAC/D,IAAID,AAAW,YAAXA,EAAE,IAAI,IAAgBC,AAAW,YAAXA,EAAE,IAAI,EAAc,OAAO;QACrD,IAAID,AAAW,YAAXA,EAAE,IAAI,IAAgBC,AAAW,YAAXA,EAAE,IAAI,EAAc,OAAO;QACrD,OAAO;IACT;IAGA,MAAMC,cAAwB,EAAE;IAChC,KAAK,MAAMJ,SAASC,aAAc;QAEhC,MAAMI,aACJL,MAAM,iBAAiB,IACvBA,MAAM,eAAe,IACrBA,MAAM,gBAAgB;QACxB,IAAIK,cAAc,CAACD,YAAY,QAAQ,CAACC,aAAa;YACnDD,YAAY,IAAI,CAACC;YACjB,IAAID,YAAY,MAAM,IAAIN,gBAAgB;QAC5C;IACF;IAEA,OAAOM;AACT;AAKO,MAAME,qBAAqB,CAChCT,SAEO;QACL,kBAAkBA,OAAO,MAAM,CAAC,CAACG,QAAUA,AAAe,iBAAfA,MAAM,IAAI;QACrD,aAAaH,OAAO,MAAM,CAAC,CAACG,QAAUA,AAAe,YAAfA,MAAM,IAAI;QAChD,aAAaH,OAAO,MAAM,CAAC,CAACG,QAAUA,AAAe,YAAfA,MAAM,IAAI;QAChD,cAAcH,OAAO,MAAM,CAAC,CAACG,QAAUA,AAAe,aAAfA,MAAM,IAAI;IACnD;AAMK,MAAMO,oBAAoB,CAC/BC,gBACAC,cAEO;QACL,YAAYD,eAAe,gBAAgB,CAAC,MAAM;QAClD,OAAOA,eAAe,WAAW,CAAC,MAAM;QACxC,OAAOA,eAAe,WAAW,CAAC,MAAM;QACxC,QAAQA,eAAe,YAAY,CAAC,MAAM;QAC1C,OAAOC;IACT;AAMK,MAAMC,2BAA2B,CACtCC,cAEOA,YACJ,GAAG,CAAC,CAACX,QAAW;YACf,aAAaA,MAAM,kBAAkB,IAAI;YACzC,OAAOA,MAAM,KAAK,IAAI;QACxB,IACC,MAAM,CAAC,CAACY,OAASA,KAAK,WAAW,IAAIA,KAAK,KAAK;AAM7C,MAAMC,sBAAsB,CACjChB,SAEOA,OAAO,GAAG,CAAC,CAACG,QAAW;YAC5B,MAAMA,MAAM,IAAI;YAChB,WAAWA,MAAM,SAAS;YAC1B,KAAKA,MAAM,GAAG;YACd,OAAOA,MAAM,KAAK;YAClB,oBAAoBA,MAAM,kBAAkB;YAC5C,OAAOA,MAAM,KAAK;YAClB,UAAUA,MAAM,QAAQ;YACxB,aAAaA,MAAM,WAAW;QAChC;AAMK,MAAMc,sBAAsB,CACjCjB,QACAkB,UAA0D,CAAC,CAAC;IAE5D,MAAMP,iBAAiBF,mBAAmBT;IAC1C,MAAMmB,cAAcT,kBAAkBC,gBAAgBX,OAAO,MAAM;IAGnE,MAAMoB,WACJT,eAAe,gBAAgB,CAAC,MAAM,GAAG,IACrCA,eAAe,gBAAgB,CAAC,EAAE,CAAC,GAAG,IAAI,KAC1C;IAEN,MAAMU,oBAAoBV,eAAe,WAAW,CACjD,GAAG,CAAC,CAACR,QAAUA,MAAM,kBAAkB,EACvC,MAAM,CAAC,CAACmB,OAAyBC,QAAQD,OACzC,KAAK,CAAC,GAAG;IAEZ,MAAME,oBAAoBX,yBACxBF,eAAe,WAAW,EAC1B,KAAK,CAAC,GAAG;IAEX,MAAMc,OAAOd,eAAe,gBAAgB,CACzC,GAAG,CAAC,CAACe,IAAMA,EAAE,GAAG,EAChB,MAAM,CAAC,CAACC,MAAuBJ,QAAQI,MACvC,KAAK,CAAC,GAAG;IAEZ,MAAMC,kBAAkBZ,oBAAoBhB;IAE5C,OAAO;QACL,UAAUkB,QAAQ,QAAQ,IAAI;QAC9BE;QACAD;QACAM;QACAJ;QACAG;QACA,QAAQI;IACV;AACF;AAKO,MAAMC,uBAAuB,CAClCC,YACAvB,cAAwB,EAAE,EAC1BwB,qBAAqB,IAAI;IAEzB,MAAMC,iBAAwB;QAC5B;YACE,MAAM;YACN,MAAMF;QACR;KACD;IAGD,IAAIC,sBAAsBxB,YAAY,MAAM,GAAG,GAAG;QAChDyB,eAAe,OAAO,CAAC;YACrB,MAAM;YACN,MAAM;QACR;QAEAzB,YAAY,OAAO,CAAC,CAACC;YACnBwB,eAAe,IAAI,CAAC;gBAClB,MAAM;gBACN,WAAW;oBACT,KAAKxB;gBACP;YACF;QACF;IACF;IAEA,OAAOwB;AACT;AAKO,MAAMC,iBAAiB,CAACjC;IAC7B,IAAI,CAACA,OAAO,MAAM,EAChB,MAAM,IAAIkC,MAAM;AAEpB;AAOO,MAAMC,mBAAmB,OAC9BnC,QACAkB,UAAiC,CAAC,CAAC;IAEnC,IAAI;QAEFe,eAAejC;QAGf,MAAMoC,UAAUnB,oBAAoBjB,QAAQ;YAC1C,UAAUkB,QAAQ,QAAQ;YAC1B,gBAAgBA,QAAQ,cAAc,IAAI;QAC5C;QAGA,MAAMmB,cAAc;YAClB,GAAGD,OAAO;YACV,mBAAmBlB,QAAQ,iBAAiB,IAAI;QAClD;QAGA,MAAMX,cAAcR,qBAClBC,QACAkB,QAAQ,cAAc,IAAI;QAI5B,MAAMoB,SAAuC;YAC3C;gBACE,MAAM;gBACN,SAAS,CAAC,4GAA4G,EAAEC,mBAAmB;YAC7I;YACA;gBACE,MAAM;gBACN,SAAS,CAAC;;;AAGlB,EAAEC,KAAK,SAAS,CAACH,aAAa,MAAM,GAAG;;;;;;;;;wCASC,CAAC;YACnC;SACD;QAGD,IAAI9B,YAAY,MAAM,GAAG,GAAG;YAC1B+B,OAAO,IAAI,CAAC;gBACV,MAAM;gBACN,SACE;YACJ;YAEAA,OAAO,IAAI,CAAC;gBACV,MAAM;gBACN,SAAS/B,YAAY,GAAG,CAAC,CAACC,aAAgB;wBACxC,MAAM;wBACN,WAAW;4BACT,KAAKA;wBACP;oBACF;YACF;QACF;QAEA,MAAMiC,WAAW,MAAMC,OAAOJ,QAAQK,aAAa,YAAY,EAAE;YAC/D,QAAQ;QACV;QAEA,IAAIF,AAAAA,CAAAA,QAAAA,WAAAA,KAAAA,IAAAA,SAAU,OAAO,AAAD,KAAK,AAA4B,YAA5B,OAAOA,SAAS,OAAO,EAC9C,OAAOA,SAAS,OAAO;QAGzB,MAAM,IAAIP,MAAM;IAClB,EAAE,OAAOU,OAAO;QACd,MAAM,IAAIV,MAAM,CAAC,8BAA8B,EAAEU,OAAO;IAC1D;AACF;AAKO,MAAMC,yBAAyB,OACpC7C,QACAkB,UAAkE,CAAC,CAAC;IAEpE,IAAI;QAEFe,eAAejC;QAGf,MAAMoC,UAAUnB,oBAAoBjB,QAAQ;YAC1C,UAAUkB,QAAQ,QAAQ;YAC1B,gBAAgBA,QAAQ,cAAc,IAAI;QAC5C;QAGA,MAAMmB,cAAc;YAClB,GAAGD,OAAO;YACV,mBAAmBlB,QAAQ,iBAAiB,IAAI;QAClD;QAGA,MAAMX,cAAcR,qBAClBC,QACAkB,QAAQ,cAAc,IAAI;QAI5B,MAAMoB,SAAuC;YAC3C;gBACE,MAAM;gBACN,SAAS,CAAC,4GAA4G,EAAEC,mBAAmB;YAC7I;YACA;gBACE,MAAM;gBACN,SAAS,CAAC;;;AAGlB,EAAEC,KAAK,SAAS,CAACH,aAAa,MAAM,GAAG;;;;;;;;;wCASC,CAAC;YACnC;SACD;QAGD,IAAI9B,YAAY,MAAM,GAAG,GAAG;YAC1B+B,OAAO,IAAI,CAAC;gBACV,MAAM;gBACN,SACE;YACJ;YAEAA,OAAO,IAAI,CAAC;gBACV,MAAM;gBACN,SAAS/B,YAAY,GAAG,CAAC,CAACC,aAAgB;wBACxC,MAAM;wBACN,WAAW;4BACT,KAAKA;wBACP;oBACF;YACF;QACF;QAEA,IAAIU,QAAQ,MAAM,IAAIA,QAAQ,OAAO,EAEnC,OAAO,MAAMwB,OACXJ,QACAK,aAAa,YAAY,EACzB;YACE,QAAQ;QACV,GACA;YACE,QAAQ;YACR,SAASzB,QAAQ,OAAO;QAC1B;QAEG;YAEL,MAAMuB,WAAW,MAAMC,OAAOJ,QAAQK,aAAa,YAAY,EAAE;gBAC/D,QAAQ;YACV;YAEA,IAAIF,AAAAA,CAAAA,QAAAA,WAAAA,KAAAA,IAAAA,SAAU,OAAO,AAAD,KAAK,AAA4B,YAA5B,OAAOA,SAAS,OAAO,EAC9C,OAAO;gBACL,SAASA,SAAS,OAAO;gBACzB,OAAOA,SAAS,KAAK;gBACrB,YAAY;YACd;YAGF,MAAM,IAAIP,MAAM;QAClB;IACF,EAAE,OAAOU,OAAO;QACd,MAAM,IAAIV,MAAM,CAAC,8BAA8B,EAAEU,OAAO;IAC1D;AACF"}

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

@@ -1,7 +1,7 @@
 import { AIResponseFormat } from "../../types.mjs";
 import { Anthropic } from "@anthropic-ai/sdk";
 import { DefaultAzureCredential, getBearerTokenProvider } from "@azure/identity";
-import { MIDSCENE_API_TYPE, MIDSCENE_LANGSMITH_DEBUG, OPENAI_MAX_TOKENS, globalConfigManager } from "@midscene/shared/env";
+import { MIDSCENE_API_TYPE, MIDSCENE_LANGSMITH_DEBUG, OPENAI_MAX_TOKENS, globalConfigManager, uiTarsModelVersion, vlLocateMode } from "@midscene/shared/env";
 import { parseBase64 } from "@midscene/shared/img";
 import { getDebug } from "@midscene/shared/logger";
 import { assert, ifInBrowser } from "@midscene/shared/utils";
@@ -13,8 +13,8 @@ import { AIActionType } from "../common.mjs";
 import { assertSchema } from "../prompt/assertion.mjs";
 import { locatorSchema } from "../prompt/llm-locator.mjs";
 import { planSchema } from "../prompt/llm-planning.mjs";
-async function createChatClient({ AIActionTypeValue, modelConfig }) {
-    const { socksProxy, httpProxy, modelName, openaiBaseURL, openaiApiKey, openaiExtraConfig, openaiUseAzureDeprecated, useAzureOpenai, azureOpenaiScope, azureOpenaiKey, azureOpenaiEndpoint, azureOpenaiApiVersion, azureOpenaiDeployment, azureExtraConfig, useAnthropicSdk, anthropicApiKey, modelDescription, uiTarsModelVersion: uiTarsVersion, vlMode } = modelConfig;
+async function createChatClient({ AIActionTypeValue, modelPreferences }) {
+    const { socksProxy, httpProxy, modelName, openaiBaseURL, openaiApiKey, openaiExtraConfig, openaiUseAzureDeprecated, useAzureOpenai, azureOpenaiScope, azureOpenaiKey, azureOpenaiEndpoint, azureOpenaiApiVersion, azureOpenaiDeployment, azureExtraConfig, useAnthropicSdk, anthropicApiKey, modelDescription } = globalConfigManager.getModelConfigByIntent(modelPreferences.intent);
     let openai;
     let proxyAgent;
     const debugProxy = getDebug('ai:call:proxy');
@@ -76,9 +76,7 @@ async function createChatClient({ AIActionTypeValue, modelConfig }) {
         completion: openai.chat.completions,
         style: 'openai',
         modelName,
-        modelDescription,
-        uiTarsVersion,
-        vlMode
+        modelDescription
     };
     if (useAnthropicSdk) openai = new Anthropic({
         apiKey: anthropicApiKey,
@@ -89,16 +87,14 @@ async function createChatClient({ AIActionTypeValue, modelConfig }) {
         completion: openai.messages,
         style: 'anthropic',
         modelName,
-        modelDescription,
-        uiTarsVersion,
-        vlMode
+        modelDescription
     };
     throw new Error('Openai SDK or Anthropic SDK is not initialized');
 }
-async function callAI(messages, AIActionTypeValue, modelConfig, options) {
-    const { completion, style, modelName, modelDescription, uiTarsVersion, vlMode } = await createChatClient({
+async function call(messages, AIActionTypeValue, modelPreferences, options) {
+    const { completion, style, modelName, modelDescription } = await createChatClient({
         AIActionTypeValue,
-        modelConfig
+        modelPreferences
     });
     const responseFormat = getResponseFormat(modelName, AIActionTypeValue);
     const maxTokens = globalConfigManager.getEnvConfigValue(OPENAI_MAX_TOKENS);
@@ -112,10 +108,10 @@ async function callAI(messages, AIActionTypeValue, modelConfig, options) {
     let usage;
     let timeCost;
     const commonConfig = {
-        temperature: 'vlm-ui-tars' === vlMode ? 0.0 : 0.1,
+        temperature: 'vlm-ui-tars' === vlLocateMode(modelPreferences) ? 0.0 : 0.1,
         stream: !!isStreaming,
         max_tokens: 'number' == typeof maxTokens ? maxTokens : Number.parseInt(maxTokens || '2048', 10),
-        ...'qwen-vl' === vlMode ? {
+        ...'qwen-vl' === vlLocateMode(modelPreferences) ? {
             vl_high_resolution_images: true
         } : {}
     };
@@ -169,7 +165,7 @@ async function callAI(messages, AIActionTypeValue, modelConfig, options) {
                                 time_cost: timeCost ?? 0,
                                 model_name: modelName,
                                 model_description: modelDescription,
-                                intent: modelConfig.intent
+                                intent: modelPreferences.intent
                             }
                         };
                         options.onChunk(finalChunk);
@@ -177,7 +173,7 @@ async function callAI(messages, AIActionTypeValue, modelConfig, options) {
                     }
                 }
                 content = accumulated;
-                debugProfileStats(`streaming model, ${modelName}, mode, ${vlMode || 'default'}, cost-ms, ${timeCost}`);
+                debugProfileStats(`streaming model, ${modelName}, mode, ${vlLocateMode(modelPreferences) || 'default'}, cost-ms, ${timeCost}`);
             } else {
                 var _result_usage, _result_usage1, _result_usage2;
                 const result = await completion.create({
@@ -187,7 +183,7 @@ async function callAI(messages, AIActionTypeValue, modelConfig, options) {
                     ...commonConfig
                 });
                 timeCost = Date.now() - startTime;
-                debugProfileStats(`model, ${modelName}, mode, ${vlMode || 'default'}, ui-tars-version, ${uiTarsVersion}, prompt-tokens, ${(null == (_result_usage = result.usage) ? void 0 : _result_usage.prompt_tokens) || ''}, completion-tokens, ${(null == (_result_usage1 = result.usage) ? void 0 : _result_usage1.completion_tokens) || ''}, total-tokens, ${(null == (_result_usage2 = result.usage) ? void 0 : _result_usage2.total_tokens) || ''}, cost-ms, ${timeCost}, requestId, ${result._request_id || ''}`);
+                debugProfileStats(`model, ${modelName}, mode, ${vlLocateMode(modelPreferences) || 'default'}, ui-tars-version, ${uiTarsModelVersion(modelPreferences)}, prompt-tokens, ${(null == (_result_usage = result.usage) ? void 0 : _result_usage.prompt_tokens) || ''}, completion-tokens, ${(null == (_result_usage1 = result.usage) ? void 0 : _result_usage1.completion_tokens) || ''}, total-tokens, ${(null == (_result_usage2 = result.usage) ? void 0 : _result_usage2.total_tokens) || ''}, cost-ms, ${timeCost}, requestId, ${result._request_id || ''}`);
                 debugProfileDetail(`model usage detail: ${JSON.stringify(result.usage)}`);
                 assert(result.choices, `invalid response from LLM service: ${JSON.stringify(result)}`);
                 content = result.choices[0].message.content;
@@ -252,7 +248,7 @@ async function callAI(messages, AIActionTypeValue, modelConfig, options) {
                                 time_cost: timeCost ?? 0,
                                 model_name: modelName,
                                 model_description: modelDescription,
-                                intent: modelConfig.intent
+                                intent: modelPreferences.intent
                             } : void 0
                         };
                         options.onChunk(finalChunk);
@@ -294,7 +290,7 @@ async function callAI(messages, AIActionTypeValue, modelConfig, options) {
                 time_cost: timeCost ?? 0,
                 model_name: modelName,
                 model_description: modelDescription,
-                intent: modelConfig.intent
+                intent: modelPreferences.intent
             } : void 0,
             isStreamed: !!isStreaming
         };
@@ -330,18 +326,17 @@ const getResponseFormat = (modelName, AIActionTypeValue)=>{
     };
     return responseFormat;
 };
-async function callAIWithObjectResponse(messages, AIActionTypeValue, modelConfig) {
-    const response = await callAI(messages, AIActionTypeValue, modelConfig);
+async function callToGetJSONObject(messages, AIActionTypeValue, modelPreferences) {
+    const response = await call(messages, AIActionTypeValue, modelPreferences);
     assert(response, 'empty response');
-    const vlMode = modelConfig.vlMode;
-    const jsonContent = safeParseJson(response.content, vlMode);
+    const jsonContent = safeParseJson(response.content, modelPreferences);
     return {
         content: jsonContent,
         usage: response.usage
     };
 }
-async function callAIWithStringResponse(msgs, AIActionTypeValue, modelConfig) {
-    const { content, usage } = await callAI(msgs, AIActionTypeValue, modelConfig);
+async function callAiFnWithStringResponse(msgs, AIActionTypeValue, modelPreferences) {
+    const { content, usage } = await call(msgs, AIActionTypeValue, modelPreferences);
     return {
         content,
         usage
@@ -362,7 +357,7 @@ function preprocessDoubaoBboxJson(input) {
     if (input.includes('bbox')) while(/\d+\s+\d+/.test(input))input = input.replace(/(\d+)\s+(\d+)/g, '$1,$2');
     return input;
 }
-function safeParseJson(input, vlMode) {
+function safeParseJson(input, modelPreferences) {
     const cleanJsonString = extractJSONFromCodeBlock(input);
     if (null == cleanJsonString ? void 0 : cleanJsonString.match(/\((\d+),(\d+)\)/)) {
         var _cleanJsonString_match;
@@ -374,12 +369,12 @@ function safeParseJson(input, vlMode) {
     try {
         return JSON.parse(jsonrepair(cleanJsonString));
     } catch (e) {}
-    if ('doubao-vision' === vlMode || 'vlm-ui-tars' === vlMode) {
+    if ('doubao-vision' === vlLocateMode(modelPreferences) || 'vlm-ui-tars' === vlLocateMode(modelPreferences)) {
         const jsonString = preprocessDoubaoBboxJson(cleanJsonString);
         return JSON.parse(jsonrepair(jsonString));
     }
     throw Error(`failed to parse json response: ${input}`);
 }
-export { callAI, callAIWithObjectResponse, callAIWithStringResponse, extractJSONFromCodeBlock, getResponseFormat, preprocessDoubaoBboxJson, safeParseJson };
+export { call, callAiFnWithStringResponse, callToGetJSONObject, extractJSONFromCodeBlock, getResponseFormat, preprocessDoubaoBboxJson, safeParseJson };
 
 //# sourceMappingURL=index.mjs.map

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

@@ -1 +1 @@
-{"version":3,"file":"ai-model/service-caller/index.mjs","sources":["webpack://@midscene/core/./src/ai-model/service-caller/index.ts"],"sourcesContent":["import { AIResponseFormat, type AIUsageInfo } from '@/types';\nimport type { CodeGenerationChunk, StreamingCallback } from '@/types';\nimport { Anthropic } from '@anthropic-ai/sdk';\nimport {\n  DefaultAzureCredential,\n  getBearerTokenProvider,\n} from '@azure/identity';\nimport {\n  type IModelConfig,\n  MIDSCENE_API_TYPE,\n  MIDSCENE_LANGSMITH_DEBUG,\n  OPENAI_MAX_TOKENS,\n  type TVlModeTypes,\n  type UITarsModelVersion,\n  globalConfigManager,\n} from '@midscene/shared/env';\n\nimport { parseBase64 } from '@midscene/shared/img';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport { ifInBrowser } from '@midscene/shared/utils';\nimport { HttpsProxyAgent } from 'https-proxy-agent';\nimport { jsonrepair } from 'jsonrepair';\nimport OpenAI, { AzureOpenAI } 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 { locatorSchema } from '../prompt/llm-locator';\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  style: 'openai' | 'anthropic';\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    openaiUseAzureDeprecated,\n    useAzureOpenai,\n    azureOpenaiScope,\n    azureOpenaiKey,\n    azureOpenaiEndpoint,\n    azureOpenaiApiVersion,\n    azureOpenaiDeployment,\n    azureExtraConfig,\n    useAnthropicSdk,\n    anthropicApiKey,\n    modelDescription,\n    uiTarsModelVersion: uiTarsVersion,\n    vlMode,\n  } = modelConfig;\n\n  let openai: OpenAI | AzureOpenAI | undefined;\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  if (openaiUseAzureDeprecated) {\n    // this is deprecated\n    openai = new AzureOpenAI({\n      baseURL: openaiBaseURL,\n      apiKey: openaiApiKey,\n      httpAgent: proxyAgent,\n      ...openaiExtraConfig,\n      dangerouslyAllowBrowser: true,\n    }) as OpenAI;\n  } else if (useAzureOpenai) {\n    // https://learn.microsoft.com/en-us/azure/ai-services/openai/chatgpt-quickstart?tabs=bash%2Cjavascript-key%2Ctypescript-keyless%2Cpython&pivots=programming-language-javascript#rest-api\n    // keyless authentication\n    let tokenProvider: any = undefined;\n    if (azureOpenaiScope) {\n      assert(\n        !ifInBrowser,\n        'Azure OpenAI is not supported in browser with Midscene.',\n      );\n      const credential = new DefaultAzureCredential();\n\n      tokenProvider = getBearerTokenProvider(credential, azureOpenaiScope);\n\n      openai = new AzureOpenAI({\n        azureADTokenProvider: tokenProvider,\n        endpoint: azureOpenaiEndpoint,\n        apiVersion: azureOpenaiApiVersion,\n        deployment: azureOpenaiDeployment,\n        ...openaiExtraConfig,\n        ...azureExtraConfig,\n      });\n    } else {\n      // endpoint, apiKey, apiVersion, deployment\n      openai = new AzureOpenAI({\n        apiKey: azureOpenaiKey,\n        endpoint: azureOpenaiEndpoint,\n        apiVersion: azureOpenaiApiVersion,\n        deployment: azureOpenaiDeployment,\n        dangerouslyAllowBrowser: true,\n        ...openaiExtraConfig,\n        ...azureExtraConfig,\n      });\n    }\n  } else if (!useAnthropicSdk) {\n    openai = new OpenAI({\n      baseURL: openaiBaseURL,\n      apiKey: openaiApiKey,\n      httpAgent: proxyAgent,\n      ...openaiExtraConfig,\n      defaultHeaders: {\n        ...(openaiExtraConfig?.defaultHeaders || {}),\n        [MIDSCENE_API_TYPE]: AIActionTypeValue.toString(),\n      },\n      dangerouslyAllowBrowser: true,\n    });\n  }\n\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    const { wrapOpenAI } = await import('langsmith/wrappers');\n    openai = wrapOpenAI(openai);\n  }\n\n  if (typeof openai !== 'undefined') {\n    return {\n      completion: openai.chat.completions,\n      style: 'openai',\n      modelName,\n      modelDescription,\n      uiTarsVersion,\n      vlMode,\n    };\n  }\n\n  // Anthropic\n  if (useAnthropicSdk) {\n    openai = new Anthropic({\n      apiKey: anthropicApiKey,\n      httpAgent: proxyAgent,\n      dangerouslyAllowBrowser: true,\n    }) as any;\n  }\n\n  if (typeof openai !== 'undefined' && (openai as any).messages) {\n    return {\n      completion: (openai as any).messages,\n      style: 'anthropic',\n      modelName,\n      modelDescription,\n      uiTarsVersion,\n      vlMode,\n    };\n  }\n\n  throw new Error('Openai SDK or Anthropic SDK is not initialized');\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 {\n    completion,\n    style,\n    modelName,\n    modelDescription,\n    uiTarsVersion,\n    vlMode,\n  } = await createChatClient({\n    AIActionTypeValue,\n    modelConfig,\n  });\n\n  const responseFormat = getResponseFormat(modelName, AIActionTypeValue);\n\n  const maxTokens = 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 commonConfig = {\n    temperature: vlMode === 'vlm-ui-tars' ? 0.0 : 0.1,\n    stream: !!isStreaming,\n    max_tokens:\n      typeof maxTokens === 'number'\n        ? maxTokens\n        : Number.parseInt(maxTokens || '2048', 10),\n    ...(vlMode === 'qwen-vl' // qwen specific config\n      ? {\n          vl_high_resolution_images: true,\n        }\n      : {}),\n  };\n\n  try {\n    if (style === 'openai') {\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: {\n                prompt_tokens: usage.prompt_tokens ?? 0,\n                completion_tokens: usage.completion_tokens ?? 0,\n                total_tokens: usage.total_tokens ?? 0,\n                time_cost: timeCost ?? 0,\n                model_name: modelName,\n                model_description: modelDescription,\n                intent: modelConfig.intent,\n              },\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(\n          `model usage detail: ${JSON.stringify(result.usage)}`,\n        );\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    } else if (style === 'anthropic') {\n      const convertImageContent = (content: any) => {\n        if (content.type === 'image_url') {\n          const imgBase64 = content.image_url.url;\n          assert(imgBase64, 'image_url is required');\n          const { mimeType, body } = parseBase64(content.image_url.url);\n          return {\n            source: {\n              type: 'base64',\n              media_type: mimeType,\n              data: body,\n            },\n            type: 'image',\n          };\n        }\n        return content;\n      };\n\n      if (isStreaming) {\n        const stream = (await completion.create({\n          model: modelName,\n          system: 'You are a versatile professional in software UI automation',\n          messages: messages.map((m) => ({\n            role: 'user',\n            content: Array.isArray(m.content)\n              ? (m.content as any).map(convertImageContent)\n              : m.content,\n          })),\n          response_format: responseFormat,\n          ...commonConfig,\n        } as any)) as any;\n\n        for await (const chunk of stream) {\n          const content = chunk.delta?.text || '';\n          if (content) {\n            accumulated += content;\n            const chunkData: CodeGenerationChunk = {\n              content,\n              accumulated,\n              reasoning_content: '',\n              isComplete: false,\n              usage: undefined,\n            };\n            options.onChunk!(chunkData);\n          }\n\n          // Check if stream is complete\n          if (chunk.type === 'message_stop') {\n            timeCost = Date.now() - startTime;\n            const anthropicUsage = chunk.usage;\n\n            // Send final chunk\n            const finalChunk: CodeGenerationChunk = {\n              content: '',\n              accumulated,\n              reasoning_content: '',\n              isComplete: true,\n              usage: anthropicUsage\n                ? {\n                    prompt_tokens: anthropicUsage.input_tokens ?? 0,\n                    completion_tokens: anthropicUsage.output_tokens ?? 0,\n                    total_tokens:\n                      (anthropicUsage.input_tokens ?? 0) +\n                      (anthropicUsage.output_tokens ?? 0),\n                    time_cost: timeCost ?? 0,\n                    model_name: modelName,\n                    model_description: modelDescription,\n                    intent: modelConfig.intent,\n                  }\n                : undefined,\n            };\n            options.onChunk!(finalChunk);\n            break;\n          }\n        }\n        content = accumulated;\n      } else {\n        const result = await completion.create({\n          model: modelName,\n          system: 'You are a versatile professional in software UI automation',\n          messages: messages.map((m) => ({\n            role: 'user',\n            content: Array.isArray(m.content)\n              ? (m.content as any).map(convertImageContent)\n              : m.content,\n          })),\n          response_format: responseFormat,\n          ...commonConfig,\n        } as any);\n        timeCost = Date.now() - startTime;\n        content = (result as any).content[0].text as string;\n        usage = result.usage;\n      }\n\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      };\n    }\n\n    return {\n      content: content || '',\n      usage: usage\n        ? {\n            prompt_tokens: usage.prompt_tokens ?? 0,\n            completion_tokens: usage.completion_tokens ?? 0,\n            total_tokens: usage.total_tokens ?? 0,\n            time_cost: timeCost ?? 0,\n            model_name: modelName,\n            model_description: modelDescription,\n            intent: modelConfig.intent,\n          }\n        : undefined,\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.INSPECT_ELEMENT:\n        responseFormat = locatorSchema;\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    }\n  }\n\n  // gpt-4o-2024-05-13 only supports json_object response format\n  if (modelName === 'gpt-4o-2024-05-13') {\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; 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 { content: jsonContent, usage: response.usage };\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\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  try {\n    return JSON.parse(cleanJsonString);\n  } catch {}\n  try {\n    return JSON.parse(jsonrepair(cleanJsonString));\n  } catch (e) {}\n\n  if (vlMode === 'doubao-vision' || vlMode === 'vlm-ui-tars') {\n    const jsonString = preprocessDoubaoBboxJson(cleanJsonString);\n    return JSON.parse(jsonrepair(jsonString));\n  }\n  throw Error(`failed to parse json response: ${input}`);\n}\n"],"names":["createChatClient","AIActionTypeValue","modelConfig","socksProxy","httpProxy","modelName","openaiBaseURL","openaiApiKey","openaiExtraConfig","openaiUseAzureDeprecated","useAzureOpenai","azureOpenaiScope","azureOpenaiKey","azureOpenaiEndpoint","azureOpenaiApiVersion","azureOpenaiDeployment","azureExtraConfig","useAnthropicSdk","anthropicApiKey","modelDescription","uiTarsVersion","vlMode","openai","proxyAgent","debugProxy","getDebug","HttpsProxyAgent","SocksProxyAgent","AzureOpenAI","tokenProvider","assert","ifInBrowser","credential","DefaultAzureCredential","getBearerTokenProvider","OpenAI","MIDSCENE_API_TYPE","globalConfigManager","MIDSCENE_LANGSMITH_DEBUG","Error","console","wrapOpenAI","Anthropic","callAI","messages","options","completion","style","responseFormat","getResponseFormat","maxTokens","OPENAI_MAX_TOKENS","debugCall","debugProfileStats","debugProfileDetail","startTime","Date","isStreaming","content","accumulated","usage","timeCost","commonConfig","Number","stream","chunk","_chunk_choices__delta","_chunk_choices__delta1","_chunk_choices_2","reasoning_content","chunkData","undefined","estimatedTokens","Math","finalChunk","_result_usage","_result_usage1","_result_usage2","result","JSON","convertImageContent","imgBase64","mimeType","body","parseBase64","m","Array","_chunk_delta","anthropicUsage","e","newError","AIActionType","assertSchema","locatorSchema","planSchema","AIResponseFormat","callAIWithObjectResponse","response","jsonContent","safeParseJson","callAIWithStringResponse","msgs","extractJSONFromCodeBlock","jsonMatch","codeBlockMatch","jsonLikeMatch","preprocessDoubaoBboxJson","input","cleanJsonString","_cleanJsonString_match","jsonrepair","jsonString"],"mappings":";;;;;;;;;;;;;;;AAgCA,eAAeA,iBAAiB,EAC9BC,iBAAiB,EACjBC,WAAW,EAIZ;IAQC,MAAM,EACJC,UAAU,EACVC,SAAS,EACTC,SAAS,EACTC,aAAa,EACbC,YAAY,EACZC,iBAAiB,EACjBC,wBAAwB,EACxBC,cAAc,EACdC,gBAAgB,EAChBC,cAAc,EACdC,mBAAmB,EACnBC,qBAAqB,EACrBC,qBAAqB,EACrBC,gBAAgB,EAChBC,eAAe,EACfC,eAAe,EACfC,gBAAgB,EAChB,oBAAoBC,aAAa,EACjCC,MAAM,EACP,GAAGnB;IAEJ,IAAIoB;IAEJ,IAAIC;IACJ,MAAMC,aAAaC,SAAS;IAC5B,IAAIrB,WAAW;QACboB,WAAW,oBAAoBpB;QAC/BmB,aAAa,IAAIG,gBAAgBtB;IACnC,OAAO,IAAID,YAAY;QACrBqB,WAAW,qBAAqBrB;QAChCoB,aAAa,IAAII,gBAAgBxB;IACnC;IAEA,IAAIM,0BAEFa,SAAS,IAAIM,YAAY;QACvB,SAAStB;QACT,QAAQC;QACR,WAAWgB;QACX,GAAGf,iBAAiB;QACpB,yBAAyB;IAC3B;SACK,IAAIE,gBAAgB;QAGzB,IAAImB;QACJ,IAAIlB,kBAAkB;YACpBmB,OACE,CAACC,aACD;YAEF,MAAMC,aAAa,IAAIC;YAEvBJ,gBAAgBK,uBAAuBF,YAAYrB;YAEnDW,SAAS,IAAIM,YAAY;gBACvB,sBAAsBC;gBACtB,UAAUhB;gBACV,YAAYC;gBACZ,YAAYC;gBACZ,GAAGP,iBAAiB;gBACpB,GAAGQ,gBAAgB;YACrB;QACF,OAEEM,SAAS,IAAIM,YAAY;YACvB,QAAQhB;YACR,UAAUC;YACV,YAAYC;YACZ,YAAYC;YACZ,yBAAyB;YACzB,GAAGP,iBAAiB;YACpB,GAAGQ,gBAAgB;QACrB;IAEJ,OAAO,IAAI,CAACC,iBACVK,SAAS,IAAIa,SAAO;QAClB,SAAS7B;QACT,QAAQC;QACR,WAAWgB;QACX,GAAGf,iBAAiB;QACpB,gBAAgB;YACd,GAAIA,AAAAA,CAAAA,QAAAA,oBAAAA,KAAAA,IAAAA,kBAAmB,cAAc,AAAD,KAAK,CAAC,CAAC;YAC3C,CAAC4B,kBAAkB,EAAEnC,kBAAkB,QAAQ;QACjD;QACA,yBAAyB;IAC3B;IAGF,IACEqB,UACAe,oBAAoB,qBAAqB,CAACC,2BAC1C;QACA,IAAIP,aACF,MAAM,IAAIQ,MAAM;QAElBC,QAAQ,GAAG,CAAC;QACZ,MAAM,EAAEC,UAAU,EAAE,GAAG,MAAM,MAAM,CAAC;QACpCnB,SAASmB,WAAWnB;IACtB;IAEA,IAAI,AAAkB,WAAXA,QACT,OAAO;QACL,YAAYA,OAAO,IAAI,CAAC,WAAW;QACnC,OAAO;QACPjB;QACAc;QACAC;QACAC;IACF;IAIF,IAAIJ,iBACFK,SAAS,IAAIoB,UAAU;QACrB,QAAQxB;QACR,WAAWK;QACX,yBAAyB;IAC3B;IAGF,IAAI,AAAkB,WAAXD,UAA2BA,OAAe,QAAQ,EAC3D,OAAO;QACL,YAAaA,OAAe,QAAQ;QACpC,OAAO;QACPjB;QACAc;QACAC;QACAC;IACF;IAGF,MAAM,IAAIkB,MAAM;AAClB;AAEO,eAAeI,OACpBC,QAAsC,EACtC3C,iBAA+B,EAC/BC,WAAyB,EACzB2C,OAGC;IAED,MAAM,EACJC,UAAU,EACVC,KAAK,EACL1C,SAAS,EACTc,gBAAgB,EAChBC,aAAa,EACbC,MAAM,EACP,GAAG,MAAMrB,iBAAiB;QACzBC;QACAC;IACF;IAEA,MAAM8C,iBAAiBC,kBAAkB5C,WAAWJ;IAEpD,MAAMiD,YAAYb,oBAAoB,iBAAiB,CAACc;IACxD,MAAMC,YAAY3B,SAAS;IAC3B,MAAM4B,oBAAoB5B,SAAS;IACnC,MAAM6B,qBAAqB7B,SAAS;IAEpC,MAAM8B,YAAYC,KAAK,GAAG;IAE1B,MAAMC,cAAcZ,AAAAA,CAAAA,QAAAA,UAAAA,KAAAA,IAAAA,QAAS,MAAM,AAAD,KAAKA,CAAAA,QAAAA,UAAAA,KAAAA,IAAAA,QAAS,OAAO,AAAD;IACtD,IAAIa;IACJ,IAAIC,cAAc;IAClB,IAAIC;IACJ,IAAIC;IAEJ,MAAMC,eAAe;QACnB,aAAazC,AAAW,kBAAXA,SAA2B,MAAM;QAC9C,QAAQ,CAAC,CAACoC;QACV,YACE,AAAqB,YAArB,OAAOP,YACHA,YACAa,OAAO,QAAQ,CAACb,aAAa,QAAQ;QAC3C,GAAI7B,AAAW,cAAXA,SACA;YACE,2BAA2B;QAC7B,IACA,CAAC,CAAC;IACR;IAEA,IAAI;QACF,IAAI0B,AAAU,aAAVA,OAAoB;YACtBK,UACE,CAAC,QAAQ,EAAEK,cAAc,eAAe,GAAG,WAAW,EAAEpD,WAAW;YAGrE,IAAIoD,aAAa;gBACf,MAAMO,SAAU,MAAMlB,WAAW,MAAM,CACrC;oBACE,OAAOzC;oBACPuC;oBACA,iBAAiBI;oBACjB,GAAGc,YAAY;gBACjB,GACA;oBACE,QAAQ;gBACV;gBAKF,WAAW,MAAMG,SAASD,OAAQ;wBAChBE,uBAAAA,iBAAAA,gBAEbC,wBAAAA,kBAAAA,iBAoBCC,kBAAAA;oBAtBJ,MAAMV,UAAUQ,AAAAA,SAAAA,CAAAA,iBAAAA,MAAM,OAAO,AAAD,IAAZA,KAAAA,IAAAA,QAAAA,CAAAA,kBAAAA,cAAe,CAAC,EAAE,AAAD,IAAjBA,KAAAA,IAAAA,QAAAA,CAAAA,wBAAAA,gBAAoB,KAAK,AAAD,IAAxBA,KAAAA,IAAAA,sBAA2B,OAAO,AAAD,KAAK;oBACtD,MAAMG,oBACJ,AAAC,SAAAF,CAAAA,kBAAAA,MAAM,OAAO,AAAD,IAAZA,KAAAA,IAAAA,QAAAA,CAAAA,mBAAAA,eAAe,CAAC,EAAE,AAAD,IAAjBA,KAAAA,IAAAA,QAAAA,CAAAA,yBAAAA,iBAAoB,KAAK,AAAD,IAAxBA,KAAAA,IAAAA,uBAAmC,iBAAiB,AAAD,KAAK;oBAG3D,IAAIF,MAAM,KAAK,EACbL,QAAQK,MAAM,KAAK;oBAGrB,IAAIP,WAAWW,mBAAmB;wBAChCV,eAAeD;wBACf,MAAMY,YAAiC;4BACrCZ;4BACAW;4BACAV;4BACA,YAAY;4BACZ,OAAOY;wBACT;wBACA1B,QAAQ,OAAO,CAAEyB;oBACnB;oBAGA,IAAI,QAAAF,CAAAA,kBAAAA,MAAM,OAAO,AAAD,IAAZA,KAAAA,IAAAA,QAAAA,CAAAA,mBAAAA,eAAe,CAAC,EAAE,AAAD,IAAjBA,KAAAA,IAAAA,iBAAoB,aAAa,EAAE;wBACrCP,WAAWL,KAAK,GAAG,KAAKD;wBAGxB,IAAI,CAACK,OAAO;4BAEV,MAAMY,kBAAkBC,KAAK,GAAG,CAC9B,GACAA,KAAK,KAAK,CAACd,YAAY,MAAM,GAAG;4BAElCC,QAAQ;gCACN,eAAeY;gCACf,mBAAmBA;gCACnB,cAAcA,AAAkB,IAAlBA;4BAChB;wBACF;wBAGA,MAAME,aAAkC;4BACtC,SAAS;4BACTf;4BACA,mBAAmB;4BACnB,YAAY;4BACZ,OAAO;gCACL,eAAeC,MAAM,aAAa,IAAI;gCACtC,mBAAmBA,MAAM,iBAAiB,IAAI;gCAC9C,cAAcA,MAAM,YAAY,IAAI;gCACpC,WAAWC,YAAY;gCACvB,YAAYxD;gCACZ,mBAAmBc;gCACnB,QAAQjB,YAAY,MAAM;4BAC5B;wBACF;wBACA2C,QAAQ,OAAO,CAAE6B;wBACjB;oBACF;gBACF;gBACAhB,UAAUC;gBACVN,kBACE,CAAC,iBAAiB,EAAEhD,UAAU,QAAQ,EAAEgB,UAAU,UAAU,WAAW,EAAEwC,UAAU;YAEvF,OAAO;oBAUqGc,eAAyDC,gBAAwDC;gBAT3N,MAAMC,SAAS,MAAMhC,WAAW,MAAM,CAAC;oBACrC,OAAOzC;oBACPuC;oBACA,iBAAiBI;oBACjB,GAAGc,YAAY;gBACjB;gBACAD,WAAWL,KAAK,GAAG,KAAKD;gBAExBF,kBACE,CAAC,OAAO,EAAEhD,UAAU,QAAQ,EAAEgB,UAAU,UAAU,mBAAmB,EAAED,cAAc,iBAAiB,EAAEuD,AAAAA,SAAAA,CAAAA,gBAAAA,OAAO,KAAK,AAAD,IAAXA,KAAAA,IAAAA,cAAc,aAAa,AAAD,KAAK,GAAG,qBAAqB,EAAEC,AAAAA,SAAAA,CAAAA,iBAAAA,OAAO,KAAK,AAAD,IAAXA,KAAAA,IAAAA,eAAc,iBAAiB,AAAD,KAAK,GAAG,gBAAgB,EAAEC,AAAAA,SAAAA,CAAAA,iBAAAA,OAAO,KAAK,AAAD,IAAXA,KAAAA,IAAAA,eAAc,YAAY,AAAD,KAAK,GAAG,WAAW,EAAEhB,SAAS,aAAa,EAAEiB,OAAO,WAAW,IAAI,IAAI;gBAG3TxB,mBACE,CAAC,oBAAoB,EAAEyB,KAAK,SAAS,CAACD,OAAO,KAAK,GAAG;gBAGvDhD,OACEgD,OAAO,OAAO,EACd,CAAC,mCAAmC,EAAEC,KAAK,SAAS,CAACD,SAAS;gBAEhEpB,UAAUoB,OAAO,OAAO,CAAC,EAAE,CAAC,OAAO,CAAC,OAAO;gBAC3ClB,QAAQkB,OAAO,KAAK;YACtB;YAEA1B,UAAU,CAAC,UAAU,EAAEM,SAAS;YAChC5B,OAAO4B,SAAS;QAClB,OAAO,IAAIX,AAAU,gBAAVA,OAAuB;YAChC,MAAMiC,sBAAsB,CAACtB;gBAC3B,IAAIA,AAAiB,gBAAjBA,QAAQ,IAAI,EAAkB;oBAChC,MAAMuB,YAAYvB,QAAQ,SAAS,CAAC,GAAG;oBACvC5B,OAAOmD,WAAW;oBAClB,MAAM,EAAEC,QAAQ,EAAEC,IAAI,EAAE,GAAGC,YAAY1B,QAAQ,SAAS,CAAC,GAAG;oBAC5D,OAAO;wBACL,QAAQ;4BACN,MAAM;4BACN,YAAYwB;4BACZ,MAAMC;wBACR;wBACA,MAAM;oBACR;gBACF;gBACA,OAAOzB;YACT;YAEA,IAAID,aAAa;gBACf,MAAMO,SAAU,MAAMlB,WAAW,MAAM,CAAC;oBACtC,OAAOzC;oBACP,QAAQ;oBACR,UAAUuC,SAAS,GAAG,CAAC,CAACyC,IAAO;4BAC7B,MAAM;4BACN,SAASC,MAAM,OAAO,CAACD,EAAE,OAAO,IAC3BA,EAAE,OAAO,CAAS,GAAG,CAACL,uBACvBK,EAAE,OAAO;wBACf;oBACA,iBAAiBrC;oBACjB,GAAGc,YAAY;gBACjB;gBAEA,WAAW,MAAMG,SAASD,OAAQ;wBAChBuB;oBAAhB,MAAM7B,UAAU6B,AAAAA,SAAAA,CAAAA,eAAAA,MAAM,KAAK,AAAD,IAAVA,KAAAA,IAAAA,aAAa,IAAI,AAAD,KAAK;oBACrC,IAAI7B,SAAS;wBACXC,eAAeD;wBACf,MAAMY,YAAiC;4BACrCZ;4BACAC;4BACA,mBAAmB;4BACnB,YAAY;4BACZ,OAAOY;wBACT;wBACA1B,QAAQ,OAAO,CAAEyB;oBACnB;oBAGA,IAAIL,AAAe,mBAAfA,MAAM,IAAI,EAAqB;wBACjCJ,WAAWL,KAAK,GAAG,KAAKD;wBACxB,MAAMiC,iBAAiBvB,MAAM,KAAK;wBAGlC,MAAMS,aAAkC;4BACtC,SAAS;4BACTf;4BACA,mBAAmB;4BACnB,YAAY;4BACZ,OAAO6B,iBACH;gCACE,eAAeA,eAAe,YAAY,IAAI;gCAC9C,mBAAmBA,eAAe,aAAa,IAAI;gCACnD,cACGA,AAAAA,CAAAA,eAAe,YAAY,IAAI,KAC/BA,CAAAA,eAAe,aAAa,IAAI;gCACnC,WAAW3B,YAAY;gCACvB,YAAYxD;gCACZ,mBAAmBc;gCACnB,QAAQjB,YAAY,MAAM;4BAC5B,IACAqE;wBACN;wBACA1B,QAAQ,OAAO,CAAE6B;wBACjB;oBACF;gBACF;gBACAhB,UAAUC;YACZ,OAAO;gBACL,MAAMmB,SAAS,MAAMhC,WAAW,MAAM,CAAC;oBACrC,OAAOzC;oBACP,QAAQ;oBACR,UAAUuC,SAAS,GAAG,CAAC,CAACyC,IAAO;4BAC7B,MAAM;4BACN,SAASC,MAAM,OAAO,CAACD,EAAE,OAAO,IAC3BA,EAAE,OAAO,CAAS,GAAG,CAACL,uBACvBK,EAAE,OAAO;wBACf;oBACA,iBAAiBrC;oBACjB,GAAGc,YAAY;gBACjB;gBACAD,WAAWL,KAAK,GAAG,KAAKD;gBACxBG,UAAWoB,OAAe,OAAO,CAAC,EAAE,CAAC,IAAI;gBACzClB,QAAQkB,OAAO,KAAK;YACtB;YAEAhD,OAAO4B,SAAS;QAClB;QAEA,IAAID,eAAe,CAACG,OAAO;YAEzB,MAAMY,kBAAkBC,KAAK,GAAG,CAC9B,GACAA,KAAK,KAAK,CAAEf,AAAAA,CAAAA,WAAW,EAAC,EAAG,MAAM,GAAG;YAEtCE,QAAQ;gBACN,eAAeY;gBACf,mBAAmBA;gBACnB,cAAcA,AAAkB,IAAlBA;YAChB;QACF;QAEA,OAAO;YACL,SAASd,WAAW;YACpB,OAAOE,QACH;gBACE,eAAeA,MAAM,aAAa,IAAI;gBACtC,mBAAmBA,MAAM,iBAAiB,IAAI;gBAC9C,cAAcA,MAAM,YAAY,IAAI;gBACpC,WAAWC,YAAY;gBACvB,YAAYxD;gBACZ,mBAAmBc;gBACnB,QAAQjB,YAAY,MAAM;YAC5B,IACAqE;YACJ,YAAY,CAAC,CAACd;QAChB;IACF,EAAE,OAAOgC,GAAQ;QACfjD,QAAQ,KAAK,CAAC,kBAAkBiD;QAChC,MAAMC,WAAW,IAAInD,MACnB,CAAC,eAAe,EAAEkB,cAAc,eAAe,GAAG,kBAAkB,EAAEgC,EAAE,OAAO,CAAC,8DAA8D,CAAC,EAC/I;YACE,OAAOA;QACT;QAEF,MAAMC;IACR;AACF;AAEO,MAAMzC,oBAAoB,CAC/B5C,WACAJ;IAIA,IAAI+C;IAKJ,IAAI3C,UAAU,QAAQ,CAAC,UACrB,OAAQJ;QACN,KAAK0F,aAAa,MAAM;YACtB3C,iBAAiB4C;YACjB;QACF,KAAKD,aAAa,eAAe;YAC/B3C,iBAAiB6C;YACjB;QACF,KAAKF,aAAa,IAAI;YACpB3C,iBAAiB8C;YACjB;QACF,KAAKH,aAAa,YAAY;QAC9B,KAAKA,aAAa,gBAAgB;YAChC3C,iBAAiB;gBAAE,MAAM+C,iBAAiB,IAAI;YAAC;YAC/C;IACJ;IAIF,IAAI1F,AAAc,wBAAdA,WACF2C,iBAAiB;QAAE,MAAM+C,iBAAiB,IAAI;IAAC;IAGjD,OAAO/C;AACT;AAEO,eAAegD,yBACpBpD,QAAsC,EACtC3C,iBAA+B,EAC/BC,WAAyB;IAEzB,MAAM+F,WAAW,MAAMtD,OAAOC,UAAU3C,mBAAmBC;IAC3D4B,OAAOmE,UAAU;IACjB,MAAM5E,SAASnB,YAAY,MAAM;IACjC,MAAMgG,cAAcC,cAAcF,SAAS,OAAO,EAAE5E;IACpD,OAAO;QAAE,SAAS6E;QAAa,OAAOD,SAAS,KAAK;IAAC;AACvD;AAEO,eAAeG,yBACpBC,IAAY,EACZpG,iBAA+B,EAC/BC,WAAyB;IAEzB,MAAM,EAAEwD,OAAO,EAAEE,KAAK,EAAE,GAAG,MAAMjB,OAAO0D,MAAMpG,mBAAmBC;IACjE,OAAO;QAAEwD;QAASE;IAAM;AAC1B;AAEO,SAAS0C,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;AAEO,SAASR,cAAcQ,KAAa,EAAEtF,MAAgC;IAC3E,MAAMuF,kBAAkBN,yBAAyBK;IAEjD,IAAIC,QAAAA,kBAAAA,KAAAA,IAAAA,gBAAiB,KAAK,CAAC,oBAAoB;YACtCC;QAAP,OAAO,QAAAA,CAAAA,yBAAAA,gBACJ,KAAK,CAAC,kBAAiB,IADnBA,KAAAA,IAAAA,uBAEH,KAAK,CAAC,GACP,GAAG,CAAC9C;IACT;IACA,IAAI;QACF,OAAOgB,KAAK,KAAK,CAAC6B;IACpB,EAAE,OAAM,CAAC;IACT,IAAI;QACF,OAAO7B,KAAK,KAAK,CAAC+B,WAAWF;IAC/B,EAAE,OAAOnB,GAAG,CAAC;IAEb,IAAIpE,AAAW,oBAAXA,UAA8BA,AAAW,kBAAXA,QAA0B;QAC1D,MAAM0F,aAAaL,yBAAyBE;QAC5C,OAAO7B,KAAK,KAAK,CAAC+B,WAAWC;IAC/B;IACA,MAAMxE,MAAM,CAAC,+BAA+B,EAAEoE,OAAO;AACvD"}
+{"version":3,"file":"ai-model/service-caller/index.mjs","sources":["webpack://@midscene/core/./src/ai-model/service-caller/index.ts"],"sourcesContent":["import { AIResponseFormat, type AIUsageInfo } from '@/types';\nimport type { CodeGenerationChunk, StreamingCallback } from '@/types';\nimport { Anthropic } from '@anthropic-ai/sdk';\nimport {\n  DefaultAzureCredential,\n  getBearerTokenProvider,\n} from '@azure/identity';\nimport {\n  type IModelPreferences,\n  MIDSCENE_API_TYPE,\n  MIDSCENE_LANGSMITH_DEBUG,\n  OPENAI_MAX_TOKENS,\n  globalConfigManager,\n  uiTarsModelVersion,\n  vlLocateMode,\n} from '@midscene/shared/env';\nimport { parseBase64 } from '@midscene/shared/img';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport { ifInBrowser } from '@midscene/shared/utils';\nimport { HttpsProxyAgent } from 'https-proxy-agent';\nimport { jsonrepair } from 'jsonrepair';\nimport OpenAI, { AzureOpenAI } 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 { locatorSchema } from '../prompt/llm-locator';\nimport { planSchema } from '../prompt/llm-planning';\n\nasync function createChatClient({\n  AIActionTypeValue,\n  modelPreferences,\n}: {\n  AIActionTypeValue: AIActionType;\n  modelPreferences: IModelPreferences;\n}): Promise<{\n  completion: OpenAI.Chat.Completions;\n  style: 'openai' | 'anthropic';\n  modelName: string;\n  modelDescription: string;\n}> {\n  const {\n    socksProxy,\n    httpProxy,\n    modelName,\n    openaiBaseURL,\n    openaiApiKey,\n    openaiExtraConfig,\n    openaiUseAzureDeprecated,\n    useAzureOpenai,\n    azureOpenaiScope,\n    azureOpenaiKey,\n    azureOpenaiEndpoint,\n    azureOpenaiApiVersion,\n    azureOpenaiDeployment,\n    azureExtraConfig,\n    useAnthropicSdk,\n    anthropicApiKey,\n    modelDescription,\n  } = globalConfigManager.getModelConfigByIntent(modelPreferences.intent);\n\n  let openai: OpenAI | AzureOpenAI | undefined;\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  if (openaiUseAzureDeprecated) {\n    // this is deprecated\n    openai = new AzureOpenAI({\n      baseURL: openaiBaseURL,\n      apiKey: openaiApiKey,\n      httpAgent: proxyAgent,\n      ...openaiExtraConfig,\n      dangerouslyAllowBrowser: true,\n    }) as OpenAI;\n  } else if (useAzureOpenai) {\n    // https://learn.microsoft.com/en-us/azure/ai-services/openai/chatgpt-quickstart?tabs=bash%2Cjavascript-key%2Ctypescript-keyless%2Cpython&pivots=programming-language-javascript#rest-api\n    // keyless authentication\n    let tokenProvider: any = undefined;\n    if (azureOpenaiScope) {\n      assert(\n        !ifInBrowser,\n        'Azure OpenAI is not supported in browser with Midscene.',\n      );\n      const credential = new DefaultAzureCredential();\n\n      tokenProvider = getBearerTokenProvider(credential, azureOpenaiScope);\n\n      openai = new AzureOpenAI({\n        azureADTokenProvider: tokenProvider,\n        endpoint: azureOpenaiEndpoint,\n        apiVersion: azureOpenaiApiVersion,\n        deployment: azureOpenaiDeployment,\n        ...openaiExtraConfig,\n        ...azureExtraConfig,\n      });\n    } else {\n      // endpoint, apiKey, apiVersion, deployment\n      openai = new AzureOpenAI({\n        apiKey: azureOpenaiKey,\n        endpoint: azureOpenaiEndpoint,\n        apiVersion: azureOpenaiApiVersion,\n        deployment: azureOpenaiDeployment,\n        dangerouslyAllowBrowser: true,\n        ...openaiExtraConfig,\n        ...azureExtraConfig,\n      });\n    }\n  } else if (!useAnthropicSdk) {\n    openai = new OpenAI({\n      baseURL: openaiBaseURL,\n      apiKey: openaiApiKey,\n      httpAgent: proxyAgent,\n      ...openaiExtraConfig,\n      defaultHeaders: {\n        ...(openaiExtraConfig?.defaultHeaders || {}),\n        [MIDSCENE_API_TYPE]: AIActionTypeValue.toString(),\n      },\n      dangerouslyAllowBrowser: true,\n    });\n  }\n\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    const { wrapOpenAI } = await import('langsmith/wrappers');\n    openai = wrapOpenAI(openai);\n  }\n\n  if (typeof openai !== 'undefined') {\n    return {\n      completion: openai.chat.completions,\n      style: 'openai',\n      modelName,\n      modelDescription,\n    };\n  }\n\n  // Anthropic\n  if (useAnthropicSdk) {\n    openai = new Anthropic({\n      apiKey: anthropicApiKey,\n      httpAgent: proxyAgent,\n      dangerouslyAllowBrowser: true,\n    }) as any;\n  }\n\n  if (typeof openai !== 'undefined' && (openai as any).messages) {\n    return {\n      completion: (openai as any).messages,\n      style: 'anthropic',\n      modelName,\n      modelDescription,\n    };\n  }\n\n  throw new Error('Openai SDK or Anthropic SDK is not initialized');\n}\n\nexport async function call(\n  messages: ChatCompletionMessageParam[],\n  AIActionTypeValue: AIActionType,\n  modelPreferences: IModelPreferences,\n  options?: {\n    stream?: boolean;\n    onChunk?: StreamingCallback;\n  },\n): Promise<{ content: string; usage?: AIUsageInfo; isStreamed: boolean }> {\n  const { completion, style, modelName, modelDescription } =\n    await createChatClient({\n      AIActionTypeValue,\n      modelPreferences,\n    });\n\n  const responseFormat = getResponseFormat(modelName, AIActionTypeValue);\n\n  const maxTokens = 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 commonConfig = {\n    temperature: vlLocateMode(modelPreferences) === 'vlm-ui-tars' ? 0.0 : 0.1,\n    stream: !!isStreaming,\n    max_tokens:\n      typeof maxTokens === 'number'\n        ? maxTokens\n        : Number.parseInt(maxTokens || '2048', 10),\n    ...(vlLocateMode(modelPreferences) === 'qwen-vl' // qwen specific config\n      ? {\n          vl_high_resolution_images: true,\n        }\n      : {}),\n  };\n\n  try {\n    if (style === 'openai') {\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: {\n                prompt_tokens: usage.prompt_tokens ?? 0,\n                completion_tokens: usage.completion_tokens ?? 0,\n                total_tokens: usage.total_tokens ?? 0,\n                time_cost: timeCost ?? 0,\n                model_name: modelName,\n                model_description: modelDescription,\n                intent: modelPreferences.intent,\n              },\n            };\n            options.onChunk!(finalChunk);\n            break;\n          }\n        }\n        content = accumulated;\n        debugProfileStats(\n          `streaming model, ${modelName}, mode, ${vlLocateMode(modelPreferences) || '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, ${vlLocateMode(modelPreferences) || 'default'}, ui-tars-version, ${uiTarsModelVersion(modelPreferences)}, 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(\n          `model usage detail: ${JSON.stringify(result.usage)}`,\n        );\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    } else if (style === 'anthropic') {\n      const convertImageContent = (content: any) => {\n        if (content.type === 'image_url') {\n          const imgBase64 = content.image_url.url;\n          assert(imgBase64, 'image_url is required');\n          const { mimeType, body } = parseBase64(content.image_url.url);\n          return {\n            source: {\n              type: 'base64',\n              media_type: mimeType,\n              data: body,\n            },\n            type: 'image',\n          };\n        }\n        return content;\n      };\n\n      if (isStreaming) {\n        const stream = (await completion.create({\n          model: modelName,\n          system: 'You are a versatile professional in software UI automation',\n          messages: messages.map((m) => ({\n            role: 'user',\n            content: Array.isArray(m.content)\n              ? (m.content as any).map(convertImageContent)\n              : m.content,\n          })),\n          response_format: responseFormat,\n          ...commonConfig,\n        } as any)) as any;\n\n        for await (const chunk of stream) {\n          const content = chunk.delta?.text || '';\n          if (content) {\n            accumulated += content;\n            const chunkData: CodeGenerationChunk = {\n              content,\n              accumulated,\n              reasoning_content: '',\n              isComplete: false,\n              usage: undefined,\n            };\n            options.onChunk!(chunkData);\n          }\n\n          // Check if stream is complete\n          if (chunk.type === 'message_stop') {\n            timeCost = Date.now() - startTime;\n            const anthropicUsage = chunk.usage;\n\n            // Send final chunk\n            const finalChunk: CodeGenerationChunk = {\n              content: '',\n              accumulated,\n              reasoning_content: '',\n              isComplete: true,\n              usage: anthropicUsage\n                ? {\n                    prompt_tokens: anthropicUsage.input_tokens ?? 0,\n                    completion_tokens: anthropicUsage.output_tokens ?? 0,\n                    total_tokens:\n                      (anthropicUsage.input_tokens ?? 0) +\n                      (anthropicUsage.output_tokens ?? 0),\n                    time_cost: timeCost ?? 0,\n                    model_name: modelName,\n                    model_description: modelDescription,\n                    intent: modelPreferences.intent,\n                  }\n                : undefined,\n            };\n            options.onChunk!(finalChunk);\n            break;\n          }\n        }\n        content = accumulated;\n      } else {\n        const result = await completion.create({\n          model: modelName,\n          system: 'You are a versatile professional in software UI automation',\n          messages: messages.map((m) => ({\n            role: 'user',\n            content: Array.isArray(m.content)\n              ? (m.content as any).map(convertImageContent)\n              : m.content,\n          })),\n          response_format: responseFormat,\n          ...commonConfig,\n        } as any);\n        timeCost = Date.now() - startTime;\n        content = (result as any).content[0].text as string;\n        usage = result.usage;\n      }\n\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      };\n    }\n\n    return {\n      content: content || '',\n      usage: usage\n        ? {\n            prompt_tokens: usage.prompt_tokens ?? 0,\n            completion_tokens: usage.completion_tokens ?? 0,\n            total_tokens: usage.total_tokens ?? 0,\n            time_cost: timeCost ?? 0,\n            model_name: modelName,\n            model_description: modelDescription,\n            intent: modelPreferences.intent,\n          }\n        : undefined,\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.INSPECT_ELEMENT:\n        responseFormat = locatorSchema;\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    }\n  }\n\n  // gpt-4o-2024-05-13 only supports json_object response format\n  if (modelName === 'gpt-4o-2024-05-13') {\n    responseFormat = { type: AIResponseFormat.JSON };\n  }\n\n  return responseFormat;\n};\n\nexport async function callToGetJSONObject<T>(\n  messages: ChatCompletionMessageParam[],\n  AIActionTypeValue: AIActionType,\n  modelPreferences: IModelPreferences,\n): Promise<{ content: T; usage?: AIUsageInfo }> {\n  const response = await call(messages, AIActionTypeValue, modelPreferences);\n  assert(response, 'empty response');\n  const jsonContent = safeParseJson(response.content, modelPreferences);\n  return { content: jsonContent, usage: response.usage };\n}\n\nexport async function callAiFnWithStringResponse<T>(\n  msgs: AIArgs,\n  AIActionTypeValue: AIActionType,\n  modelPreferences: IModelPreferences,\n): Promise<{ content: string; usage?: AIUsageInfo }> {\n  const { content, usage } = await call(\n    msgs,\n    AIActionTypeValue,\n    modelPreferences,\n  );\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\nexport function safeParseJson(\n  input: string,\n  modelPreferences: IModelPreferences,\n) {\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  try {\n    return JSON.parse(cleanJsonString);\n  } catch {}\n  try {\n    return JSON.parse(jsonrepair(cleanJsonString));\n  } catch (e) {}\n\n  if (\n    vlLocateMode(modelPreferences) === 'doubao-vision' ||\n    vlLocateMode(modelPreferences) === 'vlm-ui-tars'\n  ) {\n    const jsonString = preprocessDoubaoBboxJson(cleanJsonString);\n    return JSON.parse(jsonrepair(jsonString));\n  }\n  throw Error(`failed to parse json response: ${input}`);\n}\n"],"names":["createChatClient","AIActionTypeValue","modelPreferences","socksProxy","httpProxy","modelName","openaiBaseURL","openaiApiKey","openaiExtraConfig","openaiUseAzureDeprecated","useAzureOpenai","azureOpenaiScope","azureOpenaiKey","azureOpenaiEndpoint","azureOpenaiApiVersion","azureOpenaiDeployment","azureExtraConfig","useAnthropicSdk","anthropicApiKey","modelDescription","globalConfigManager","openai","proxyAgent","debugProxy","getDebug","HttpsProxyAgent","SocksProxyAgent","AzureOpenAI","tokenProvider","assert","ifInBrowser","credential","DefaultAzureCredential","getBearerTokenProvider","OpenAI","MIDSCENE_API_TYPE","MIDSCENE_LANGSMITH_DEBUG","Error","console","wrapOpenAI","Anthropic","call","messages","options","completion","style","responseFormat","getResponseFormat","maxTokens","OPENAI_MAX_TOKENS","debugCall","debugProfileStats","debugProfileDetail","startTime","Date","isStreaming","content","accumulated","usage","timeCost","commonConfig","vlLocateMode","Number","stream","chunk","_chunk_choices__delta","_chunk_choices__delta1","_chunk_choices_2","reasoning_content","chunkData","undefined","estimatedTokens","Math","finalChunk","_result_usage","_result_usage1","_result_usage2","result","uiTarsModelVersion","JSON","convertImageContent","imgBase64","mimeType","body","parseBase64","m","Array","_chunk_delta","anthropicUsage","e","newError","AIActionType","assertSchema","locatorSchema","planSchema","AIResponseFormat","callToGetJSONObject","response","jsonContent","safeParseJson","callAiFnWithStringResponse","msgs","extractJSONFromCodeBlock","jsonMatch","codeBlockMatch","jsonLikeMatch","preprocessDoubaoBboxJson","input","cleanJsonString","_cleanJsonString_match","jsonrepair","jsonString"],"mappings":";;;;;;;;;;;;;;;AA+BA,eAAeA,iBAAiB,EAC9BC,iBAAiB,EACjBC,gBAAgB,EAIjB;IAMC,MAAM,EACJC,UAAU,EACVC,SAAS,EACTC,SAAS,EACTC,aAAa,EACbC,YAAY,EACZC,iBAAiB,EACjBC,wBAAwB,EACxBC,cAAc,EACdC,gBAAgB,EAChBC,cAAc,EACdC,mBAAmB,EACnBC,qBAAqB,EACrBC,qBAAqB,EACrBC,gBAAgB,EAChBC,eAAe,EACfC,eAAe,EACfC,gBAAgB,EACjB,GAAGC,oBAAoB,sBAAsB,CAAClB,iBAAiB,MAAM;IAEtE,IAAImB;IAEJ,IAAIC;IACJ,MAAMC,aAAaC,SAAS;IAC5B,IAAIpB,WAAW;QACbmB,WAAW,oBAAoBnB;QAC/BkB,aAAa,IAAIG,gBAAgBrB;IACnC,OAAO,IAAID,YAAY;QACrBoB,WAAW,qBAAqBpB;QAChCmB,aAAa,IAAII,gBAAgBvB;IACnC;IAEA,IAAIM,0BAEFY,SAAS,IAAIM,YAAY;QACvB,SAASrB;QACT,QAAQC;QACR,WAAWe;QACX,GAAGd,iBAAiB;QACpB,yBAAyB;IAC3B;SACK,IAAIE,gBAAgB;QAGzB,IAAIkB;QACJ,IAAIjB,kBAAkB;YACpBkB,OACE,CAACC,aACD;YAEF,MAAMC,aAAa,IAAIC;YAEvBJ,gBAAgBK,uBAAuBF,YAAYpB;YAEnDU,SAAS,IAAIM,YAAY;gBACvB,sBAAsBC;gBACtB,UAAUf;gBACV,YAAYC;gBACZ,YAAYC;gBACZ,GAAGP,iBAAiB;gBACpB,GAAGQ,gBAAgB;YACrB;QACF,OAEEK,SAAS,IAAIM,YAAY;YACvB,QAAQf;YACR,UAAUC;YACV,YAAYC;YACZ,YAAYC;YACZ,yBAAyB;YACzB,GAAGP,iBAAiB;YACpB,GAAGQ,gBAAgB;QACrB;IAEJ,OAAO,IAAI,CAACC,iBACVI,SAAS,IAAIa,SAAO;QAClB,SAAS5B;QACT,QAAQC;QACR,WAAWe;QACX,GAAGd,iBAAiB;QACpB,gBAAgB;YACd,GAAIA,AAAAA,CAAAA,QAAAA,oBAAAA,KAAAA,IAAAA,kBAAmB,cAAc,AAAD,KAAK,CAAC,CAAC;YAC3C,CAAC2B,kBAAkB,EAAElC,kBAAkB,QAAQ;QACjD;QACA,yBAAyB;IAC3B;IAGF,IACEoB,UACAD,oBAAoB,qBAAqB,CAACgB,2BAC1C;QACA,IAAIN,aACF,MAAM,IAAIO,MAAM;QAElBC,QAAQ,GAAG,CAAC;QACZ,MAAM,EAAEC,UAAU,EAAE,GAAG,MAAM,MAAM,CAAC;QACpClB,SAASkB,WAAWlB;IACtB;IAEA,IAAI,AAAkB,WAAXA,QACT,OAAO;QACL,YAAYA,OAAO,IAAI,CAAC,WAAW;QACnC,OAAO;QACPhB;QACAc;IACF;IAIF,IAAIF,iBACFI,SAAS,IAAImB,UAAU;QACrB,QAAQtB;QACR,WAAWI;QACX,yBAAyB;IAC3B;IAGF,IAAI,AAAkB,WAAXD,UAA2BA,OAAe,QAAQ,EAC3D,OAAO;QACL,YAAaA,OAAe,QAAQ;QACpC,OAAO;QACPhB;QACAc;IACF;IAGF,MAAM,IAAIkB,MAAM;AAClB;AAEO,eAAeI,KACpBC,QAAsC,EACtCzC,iBAA+B,EAC/BC,gBAAmC,EACnCyC,OAGC;IAED,MAAM,EAAEC,UAAU,EAAEC,KAAK,EAAExC,SAAS,EAAEc,gBAAgB,EAAE,GACtD,MAAMnB,iBAAiB;QACrBC;QACAC;IACF;IAEF,MAAM4C,iBAAiBC,kBAAkB1C,WAAWJ;IAEpD,MAAM+C,YAAY5B,oBAAoB,iBAAiB,CAAC6B;IACxD,MAAMC,YAAY1B,SAAS;IAC3B,MAAM2B,oBAAoB3B,SAAS;IACnC,MAAM4B,qBAAqB5B,SAAS;IAEpC,MAAM6B,YAAYC,KAAK,GAAG;IAE1B,MAAMC,cAAcZ,AAAAA,CAAAA,QAAAA,UAAAA,KAAAA,IAAAA,QAAS,MAAM,AAAD,KAAKA,CAAAA,QAAAA,UAAAA,KAAAA,IAAAA,QAAS,OAAO,AAAD;IACtD,IAAIa;IACJ,IAAIC,cAAc;IAClB,IAAIC;IACJ,IAAIC;IAEJ,MAAMC,eAAe;QACnB,aAAaC,AAAmC,kBAAnCA,aAAa3D,oBAAsC,MAAM;QACtE,QAAQ,CAAC,CAACqD;QACV,YACE,AAAqB,YAArB,OAAOP,YACHA,YACAc,OAAO,QAAQ,CAACd,aAAa,QAAQ;QAC3C,GAAIa,AAAmC,cAAnCA,aAAa3D,oBACb;YACE,2BAA2B;QAC7B,IACA,CAAC,CAAC;IACR;IAEA,IAAI;QACF,IAAI2C,AAAU,aAAVA,OAAoB;YACtBK,UACE,CAAC,QAAQ,EAAEK,cAAc,eAAe,GAAG,WAAW,EAAElD,WAAW;YAGrE,IAAIkD,aAAa;gBACf,MAAMQ,SAAU,MAAMnB,WAAW,MAAM,CACrC;oBACE,OAAOvC;oBACPqC;oBACA,iBAAiBI;oBACjB,GAAGc,YAAY;gBACjB,GACA;oBACE,QAAQ;gBACV;gBAKF,WAAW,MAAMI,SAASD,OAAQ;wBAChBE,uBAAAA,iBAAAA,gBAEbC,wBAAAA,kBAAAA,iBAoBCC,kBAAAA;oBAtBJ,MAAMX,UAAUS,AAAAA,SAAAA,CAAAA,iBAAAA,MAAM,OAAO,AAAD,IAAZA,KAAAA,IAAAA,QAAAA,CAAAA,kBAAAA,cAAe,CAAC,EAAE,AAAD,IAAjBA,KAAAA,IAAAA,QAAAA,CAAAA,wBAAAA,gBAAoB,KAAK,AAAD,IAAxBA,KAAAA,IAAAA,sBAA2B,OAAO,AAAD,KAAK;oBACtD,MAAMG,oBACJ,AAAC,SAAAF,CAAAA,kBAAAA,MAAM,OAAO,AAAD,IAAZA,KAAAA,IAAAA,QAAAA,CAAAA,mBAAAA,eAAe,CAAC,EAAE,AAAD,IAAjBA,KAAAA,IAAAA,QAAAA,CAAAA,yBAAAA,iBAAoB,KAAK,AAAD,IAAxBA,KAAAA,IAAAA,uBAAmC,iBAAiB,AAAD,KAAK;oBAG3D,IAAIF,MAAM,KAAK,EACbN,QAAQM,MAAM,KAAK;oBAGrB,IAAIR,WAAWY,mBAAmB;wBAChCX,eAAeD;wBACf,MAAMa,YAAiC;4BACrCb;4BACAY;4BACAX;4BACA,YAAY;4BACZ,OAAOa;wBACT;wBACA3B,QAAQ,OAAO,CAAE0B;oBACnB;oBAGA,IAAI,QAAAF,CAAAA,kBAAAA,MAAM,OAAO,AAAD,IAAZA,KAAAA,IAAAA,QAAAA,CAAAA,mBAAAA,eAAe,CAAC,EAAE,AAAD,IAAjBA,KAAAA,IAAAA,iBAAoB,aAAa,EAAE;wBACrCR,WAAWL,KAAK,GAAG,KAAKD;wBAGxB,IAAI,CAACK,OAAO;4BAEV,MAAMa,kBAAkBC,KAAK,GAAG,CAC9B,GACAA,KAAK,KAAK,CAACf,YAAY,MAAM,GAAG;4BAElCC,QAAQ;gCACN,eAAea;gCACf,mBAAmBA;gCACnB,cAAcA,AAAkB,IAAlBA;4BAChB;wBACF;wBAGA,MAAME,aAAkC;4BACtC,SAAS;4BACThB;4BACA,mBAAmB;4BACnB,YAAY;4BACZ,OAAO;gCACL,eAAeC,MAAM,aAAa,IAAI;gCACtC,mBAAmBA,MAAM,iBAAiB,IAAI;gCAC9C,cAAcA,MAAM,YAAY,IAAI;gCACpC,WAAWC,YAAY;gCACvB,YAAYtD;gCACZ,mBAAmBc;gCACnB,QAAQjB,iBAAiB,MAAM;4BACjC;wBACF;wBACAyC,QAAQ,OAAO,CAAE8B;wBACjB;oBACF;gBACF;gBACAjB,UAAUC;gBACVN,kBACE,CAAC,iBAAiB,EAAE9C,UAAU,QAAQ,EAAEwD,aAAa3D,qBAAqB,UAAU,WAAW,EAAEyD,UAAU;YAE/G,OAAO;oBAUoJe,eAAyDC,gBAAwDC;gBAT1Q,MAAMC,SAAS,MAAMjC,WAAW,MAAM,CAAC;oBACrC,OAAOvC;oBACPqC;oBACA,iBAAiBI;oBACjB,GAAGc,YAAY;gBACjB;gBACAD,WAAWL,KAAK,GAAG,KAAKD;gBAExBF,kBACE,CAAC,OAAO,EAAE9C,UAAU,QAAQ,EAAEwD,aAAa3D,qBAAqB,UAAU,mBAAmB,EAAE4E,mBAAmB5E,kBAAkB,iBAAiB,EAAEwE,AAAAA,SAAAA,CAAAA,gBAAAA,OAAO,KAAK,AAAD,IAAXA,KAAAA,IAAAA,cAAc,aAAa,AAAD,KAAK,GAAG,qBAAqB,EAAEC,AAAAA,SAAAA,CAAAA,iBAAAA,OAAO,KAAK,AAAD,IAAXA,KAAAA,IAAAA,eAAc,iBAAiB,AAAD,KAAK,GAAG,gBAAgB,EAAEC,AAAAA,SAAAA,CAAAA,iBAAAA,OAAO,KAAK,AAAD,IAAXA,KAAAA,IAAAA,eAAc,YAAY,AAAD,KAAK,GAAG,WAAW,EAAEjB,SAAS,aAAa,EAAEkB,OAAO,WAAW,IAAI,IAAI;gBAG1WzB,mBACE,CAAC,oBAAoB,EAAE2B,KAAK,SAAS,CAACF,OAAO,KAAK,GAAG;gBAGvDhD,OACEgD,OAAO,OAAO,EACd,CAAC,mCAAmC,EAAEE,KAAK,SAAS,CAACF,SAAS;gBAEhErB,UAAUqB,OAAO,OAAO,CAAC,EAAE,CAAC,OAAO,CAAC,OAAO;gBAC3CnB,QAAQmB,OAAO,KAAK;YACtB;YAEA3B,UAAU,CAAC,UAAU,EAAEM,SAAS;YAChC3B,OAAO2B,SAAS;QAClB,OAAO,IAAIX,AAAU,gBAAVA,OAAuB;YAChC,MAAMmC,sBAAsB,CAACxB;gBAC3B,IAAIA,AAAiB,gBAAjBA,QAAQ,IAAI,EAAkB;oBAChC,MAAMyB,YAAYzB,QAAQ,SAAS,CAAC,GAAG;oBACvC3B,OAAOoD,WAAW;oBAClB,MAAM,EAAEC,QAAQ,EAAEC,IAAI,EAAE,GAAGC,YAAY5B,QAAQ,SAAS,CAAC,GAAG;oBAC5D,OAAO;wBACL,QAAQ;4BACN,MAAM;4BACN,YAAY0B;4BACZ,MAAMC;wBACR;wBACA,MAAM;oBACR;gBACF;gBACA,OAAO3B;YACT;YAEA,IAAID,aAAa;gBACf,MAAMQ,SAAU,MAAMnB,WAAW,MAAM,CAAC;oBACtC,OAAOvC;oBACP,QAAQ;oBACR,UAAUqC,SAAS,GAAG,CAAC,CAAC2C,IAAO;4BAC7B,MAAM;4BACN,SAASC,MAAM,OAAO,CAACD,EAAE,OAAO,IAC3BA,EAAE,OAAO,CAAS,GAAG,CAACL,uBACvBK,EAAE,OAAO;wBACf;oBACA,iBAAiBvC;oBACjB,GAAGc,YAAY;gBACjB;gBAEA,WAAW,MAAMI,SAASD,OAAQ;wBAChBwB;oBAAhB,MAAM/B,UAAU+B,AAAAA,SAAAA,CAAAA,eAAAA,MAAM,KAAK,AAAD,IAAVA,KAAAA,IAAAA,aAAa,IAAI,AAAD,KAAK;oBACrC,IAAI/B,SAAS;wBACXC,eAAeD;wBACf,MAAMa,YAAiC;4BACrCb;4BACAC;4BACA,mBAAmB;4BACnB,YAAY;4BACZ,OAAOa;wBACT;wBACA3B,QAAQ,OAAO,CAAE0B;oBACnB;oBAGA,IAAIL,AAAe,mBAAfA,MAAM,IAAI,EAAqB;wBACjCL,WAAWL,KAAK,GAAG,KAAKD;wBACxB,MAAMmC,iBAAiBxB,MAAM,KAAK;wBAGlC,MAAMS,aAAkC;4BACtC,SAAS;4BACThB;4BACA,mBAAmB;4BACnB,YAAY;4BACZ,OAAO+B,iBACH;gCACE,eAAeA,eAAe,YAAY,IAAI;gCAC9C,mBAAmBA,eAAe,aAAa,IAAI;gCACnD,cACGA,AAAAA,CAAAA,eAAe,YAAY,IAAI,KAC/BA,CAAAA,eAAe,aAAa,IAAI;gCACnC,WAAW7B,YAAY;gCACvB,YAAYtD;gCACZ,mBAAmBc;gCACnB,QAAQjB,iBAAiB,MAAM;4BACjC,IACAoE;wBACN;wBACA3B,QAAQ,OAAO,CAAE8B;wBACjB;oBACF;gBACF;gBACAjB,UAAUC;YACZ,OAAO;gBACL,MAAMoB,SAAS,MAAMjC,WAAW,MAAM,CAAC;oBACrC,OAAOvC;oBACP,QAAQ;oBACR,UAAUqC,SAAS,GAAG,CAAC,CAAC2C,IAAO;4BAC7B,MAAM;4BACN,SAASC,MAAM,OAAO,CAACD,EAAE,OAAO,IAC3BA,EAAE,OAAO,CAAS,GAAG,CAACL,uBACvBK,EAAE,OAAO;wBACf;oBACA,iBAAiBvC;oBACjB,GAAGc,YAAY;gBACjB;gBACAD,WAAWL,KAAK,GAAG,KAAKD;gBACxBG,UAAWqB,OAAe,OAAO,CAAC,EAAE,CAAC,IAAI;gBACzCnB,QAAQmB,OAAO,KAAK;YACtB;YAEAhD,OAAO2B,SAAS;QAClB;QAEA,IAAID,eAAe,CAACG,OAAO;YAEzB,MAAMa,kBAAkBC,KAAK,GAAG,CAC9B,GACAA,KAAK,KAAK,CAAEhB,AAAAA,CAAAA,WAAW,EAAC,EAAG,MAAM,GAAG;YAEtCE,QAAQ;gBACN,eAAea;gBACf,mBAAmBA;gBACnB,cAAcA,AAAkB,IAAlBA;YAChB;QACF;QAEA,OAAO;YACL,SAASf,WAAW;YACpB,OAAOE,QACH;gBACE,eAAeA,MAAM,aAAa,IAAI;gBACtC,mBAAmBA,MAAM,iBAAiB,IAAI;gBAC9C,cAAcA,MAAM,YAAY,IAAI;gBACpC,WAAWC,YAAY;gBACvB,YAAYtD;gBACZ,mBAAmBc;gBACnB,QAAQjB,iBAAiB,MAAM;YACjC,IACAoE;YACJ,YAAY,CAAC,CAACf;QAChB;IACF,EAAE,OAAOkC,GAAQ;QACfnD,QAAQ,KAAK,CAAC,kBAAkBmD;QAChC,MAAMC,WAAW,IAAIrD,MACnB,CAAC,eAAe,EAAEkB,cAAc,eAAe,GAAG,kBAAkB,EAAEkC,EAAE,OAAO,CAAC,8DAA8D,CAAC,EAC/I;YACE,OAAOA;QACT;QAEF,MAAMC;IACR;AACF;AAEO,MAAM3C,oBAAoB,CAC/B1C,WACAJ;IAIA,IAAI6C;IAKJ,IAAIzC,UAAU,QAAQ,CAAC,UACrB,OAAQJ;QACN,KAAK0F,aAAa,MAAM;YACtB7C,iBAAiB8C;YACjB;QACF,KAAKD,aAAa,eAAe;YAC/B7C,iBAAiB+C;YACjB;QACF,KAAKF,aAAa,IAAI;YACpB7C,iBAAiBgD;YACjB;QACF,KAAKH,aAAa,YAAY;QAC9B,KAAKA,aAAa,gBAAgB;YAChC7C,iBAAiB;gBAAE,MAAMiD,iBAAiB,IAAI;YAAC;YAC/C;IACJ;IAIF,IAAI1F,AAAc,wBAAdA,WACFyC,iBAAiB;QAAE,MAAMiD,iBAAiB,IAAI;IAAC;IAGjD,OAAOjD;AACT;AAEO,eAAekD,oBACpBtD,QAAsC,EACtCzC,iBAA+B,EAC/BC,gBAAmC;IAEnC,MAAM+F,WAAW,MAAMxD,KAAKC,UAAUzC,mBAAmBC;IACzD2B,OAAOoE,UAAU;IACjB,MAAMC,cAAcC,cAAcF,SAAS,OAAO,EAAE/F;IACpD,OAAO;QAAE,SAASgG;QAAa,OAAOD,SAAS,KAAK;IAAC;AACvD;AAEO,eAAeG,2BACpBC,IAAY,EACZpG,iBAA+B,EAC/BC,gBAAmC;IAEnC,MAAM,EAAEsD,OAAO,EAAEE,KAAK,EAAE,GAAG,MAAMjB,KAC/B4D,MACApG,mBACAC;IAEF,OAAO;QAAEsD;QAASE;IAAM;AAC1B;AAEO,SAAS4C,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;AAEO,SAASR,cACdQ,KAAa,EACbzG,gBAAmC;IAEnC,MAAM0G,kBAAkBN,yBAAyBK;IAEjD,IAAIC,QAAAA,kBAAAA,KAAAA,IAAAA,gBAAiB,KAAK,CAAC,oBAAoB;YACtCC;QAAP,OAAO,QAAAA,CAAAA,yBAAAA,gBACJ,KAAK,CAAC,kBAAiB,IADnBA,KAAAA,IAAAA,uBAEH,KAAK,CAAC,GACP,GAAG,CAAC/C;IACT;IACA,IAAI;QACF,OAAOiB,KAAK,KAAK,CAAC6B;IACpB,EAAE,OAAM,CAAC;IACT,IAAI;QACF,OAAO7B,KAAK,KAAK,CAAC+B,WAAWF;IAC/B,EAAE,OAAOnB,GAAG,CAAC;IAEb,IACE5B,AAAmC,oBAAnCA,aAAa3D,qBACb2D,AAAmC,kBAAnCA,aAAa3D,mBACb;QACA,MAAM6G,aAAaL,yBAAyBE;QAC5C,OAAO7B,KAAK,KAAK,CAAC+B,WAAWC;IAC/B;IACA,MAAM1E,MAAM,CAAC,+BAA+B,EAAEsE,OAAO;AACvD"}

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

@@ -1,4 +1,4 @@
-import { UITarsModelVersion } from "@midscene/shared/env";
+import { UITarsModelVersion, uiTarsModelVersion, vlLocateMode } from "@midscene/shared/env";
 import { resizeImgBase64 } from "@midscene/shared/img";
 import { getDebug } from "@midscene/shared/logger";
 import { transformHotkeyInput } from "@midscene/shared/us-keyboard-layout";
@@ -6,7 +6,7 @@ import { assert } from "@midscene/shared/utils";
 import { actionParser } from "@ui-tars/action-parser";
 import { AIActionType } from "./common.mjs";
 import { getSummary, getUiTarsPlanningPrompt } from "./prompt/ui-tars-planning.mjs";
-import { callAI } from "./service-caller/index.mjs";
+import { call } from "./service-caller/index.mjs";
 const debug = getDebug('ui-tars-planning');
 const bboxSize = 10;
 const pointToBbox = (point, width, height)=>[
@@ -16,17 +16,17 @@ const pointToBbox = (point, width, height)=>[
         Math.round(Math.min(point.y + bboxSize / 2, height))
     ];
 async function vlmPlanning(options) {
-    const { conversationHistory, userInstruction, size, modelConfig } = options;
-    const { uiTarsModelVersion } = modelConfig;
+    const { conversationHistory, userInstruction, size, modelPreferences } = options;
     const systemPrompt = getUiTarsPlanningPrompt() + userInstruction;
-    const res = await callAI([
+    const res = await call([
         {
             role: 'user',
             content: systemPrompt
         },
         ...conversationHistory
-    ], AIActionType.INSPECT_ELEMENT, modelConfig);
+    ], AIActionType.INSPECT_ELEMENT, modelPreferences);
     const convertedText = convertBboxToCoordinates(res.content);
+    const modelVer = uiTarsModelVersion(modelPreferences);
     const { parsed } = actionParser({
         prediction: convertedText,
         factor: [
@@ -37,9 +37,9 @@ async function vlmPlanning(options) {
             width: size.width,
             height: size.height
         },
-        modelVer: uiTarsModelVersion
+        modelVer: modelVer || void 0
     });
-    debug('ui-tars modelVer', uiTarsModelVersion, ', parsed', JSON.stringify(parsed));
+    debug('ui-tars modelVer', modelVer, ', parsed', JSON.stringify(parsed));
     const transformActions = [];
     parsed.forEach((action)=>{
         if ('click' === action.action_type) {
@@ -155,8 +155,8 @@ function getPoint(startBox, size) {
         y * size.height
     ];
 }
-async function resizeImageForUiTars(imageBase64, size, uiTarsVersion) {
-    if (uiTarsVersion === UITarsModelVersion.V1_5) {
+async function resizeImageForUiTars(imageBase64, size, modelPreferences) {
+    if ('vlm-ui-tars' === vlLocateMode(modelPreferences) && uiTarsModelVersion(modelPreferences) === UITarsModelVersion.V1_5) {
         debug('ui-tars-v1.5, will check image size', size);
         const currentPixels = size.width * size.height;
         const maxPixels = 12845056;

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

@@ -1 +1 @@
-{"version":3,"file":"ai-model/ui-tars-planning.mjs","sources":["webpack://@midscene/core/./src/ai-model/ui-tars-planning.ts"],"sourcesContent":["import type {\n  AIUsageInfo,\n  MidsceneYamlFlowItem,\n  PlanningAction,\n  Size,\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 type { ChatCompletionMessageParam } from 'openai/resources/index';\nimport { AIActionType } from './common';\nimport { getSummary, getUiTarsPlanningPrompt } from './prompt/ui-tars-planning';\nimport { callAI } 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 vlmPlanning(options: {\n  userInstruction: string;\n  conversationHistory: ChatCompletionMessageParam[];\n  size: { width: number; height: number };\n  modelConfig: IModelConfig;\n}): Promise<{\n  actions: PlanningAction<any>[];\n  actionsFromModel: ReturnType<typeof actionParser>['parsed'];\n  action_summary: string;\n  yamlFlow?: MidsceneYamlFlowItem[];\n  usage?: AIUsageInfo;\n  rawResponse?: string;\n}> {\n  const { conversationHistory, userInstruction, size, modelConfig } = options;\n  const { uiTarsModelVersion } = modelConfig;\n  const systemPrompt = getUiTarsPlanningPrompt() + userInstruction;\n\n  const res = await callAI(\n    [\n      {\n        role: 'user',\n        content: systemPrompt,\n      },\n      ...conversationHistory,\n    ],\n    AIActionType.INSPECT_ELEMENT,\n    modelConfig,\n  );\n  const convertedText = convertBboxToCoordinates(res.content);\n\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  parsed.forEach((action) => {\n    if (action.action_type === '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 (action.action_type === '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 (action.action_type === 'type') {\n      transformActions.push({\n        type: 'Input',\n        param: {\n          value: action.action_inputs.content,\n        },\n        thought: action.thought || '',\n      });\n    } else if (action.action_type === 'scroll') {\n      transformActions.push({\n        type: 'Scroll',\n        param: {\n          direction: action.action_inputs.direction,\n        },\n        thought: action.thought || '',\n      });\n    } else if (action.action_type === 'finished') {\n      transformActions.push({\n        type: 'Finished',\n        param: {},\n        thought: action.thought || '',\n      });\n    } else if (action.action_type === '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 (action.action_type === '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: ${res.content}`, {\n      cause: {\n        prediction: res.content,\n        parsed,\n      },\n    });\n  }\n\n  debug('transformActions', JSON.stringify(transformActions, null, 2));\n\n  return {\n    actions: transformActions,\n    actionsFromModel: parsed,\n    action_summary: getSummary(res.content),\n    usage: res.usage,\n    rawResponse: JSON.stringify(res.content, undefined, 2),\n  };\n}\n\n/**\n * Converts bounding box notation to coordinate points\n * @param text - The text containing bbox tags to be converted\n * @returns The text with bbox tags replaced by coordinate points\n */\nfunction convertBboxToCoordinates(text: string): string {\n  // Match the four numbers after <bbox>\n  const pattern = /<bbox>(\\d+)\\s+(\\d+)\\s+(\\d+)\\s+(\\d+)<\\/bbox>/g;\n\n  function replaceMatch(\n    match: string,\n    x1: string,\n    y1: string,\n    x2: string,\n    y2: string,\n  ): string {\n    // Convert strings to numbers and calculate center point\n    const x1Num = Number.parseInt(x1, 10);\n    const y1Num = Number.parseInt(y1, 10);\n    const x2Num = Number.parseInt(x2, 10);\n    const y2Num = Number.parseInt(y2, 10);\n\n    // Use Math.floor to truncate and calculate center point\n    const x = Math.floor((x1Num + x2Num) / 2);\n    const y = Math.floor((y1Num + y2Num) / 2);\n\n    // Return formatted coordinate string\n    return `(${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","vlmPlanning","options","conversationHistory","userInstruction","size","modelConfig","uiTarsModelVersion","systemPrompt","getUiTarsPlanningPrompt","res","callAI","AIActionType","convertedText","convertBboxToCoordinates","parsed","actionParser","JSON","transformActions","action","assert","getPoint","startPoint","endPoint","keys","transformHotkeyInput","console","Error","getSummary","undefined","text","pattern","replaceMatch","match","x1","y1","x2","y2","x1Num","Number","y1Num","x2Num","y2Num","x","y","cleanedText","startBox","resizeImageForUiTars","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,YAAYC,OAKjC;IAQC,MAAM,EAAEC,mBAAmB,EAAEC,eAAe,EAAEC,IAAI,EAAEC,WAAW,EAAE,GAAGJ;IACpE,MAAM,EAAEK,kBAAkB,EAAE,GAAGD;IAC/B,MAAME,eAAeC,4BAA4BL;IAEjD,MAAMM,MAAM,MAAMC,OAChB;QACE;YACE,MAAM;YACN,SAASH;QACX;WACGL;KACJ,EACDS,aAAa,eAAe,EAC5BN;IAEF,MAAMO,gBAAgBC,yBAAyBJ,IAAI,OAAO;IAE1D,MAAM,EAAEK,MAAM,EAAE,GAAGC,aAAa;QAC9B,YAAYH;QACZ,QAAQ;YAAC;YAAM;SAAK;QACpB,eAAe;YACb,OAAOR,KAAK,KAAK;YACjB,QAAQA,KAAK,MAAM;QACrB;QACA,UAAUE;IACZ;IAEAd,MACE,oBACAc,oBACA,YACAU,KAAK,SAAS,CAACF;IAGjB,MAAMG,mBAAqC,EAAE;IAC7CH,OAAO,OAAO,CAAC,CAACI;QACd,IAAIA,AAAuB,YAAvBA,OAAO,WAAW,EAAc;YAClCC,OAAOD,OAAO,aAAa,CAAC,SAAS,EAAE;YACvC,MAAMtB,QAAQwB,SAASF,OAAO,aAAa,CAAC,SAAS,EAAEd;YACvDa,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO;oBACL,QAAQ;wBACN,QAAQC,OAAO,OAAO,IAAI;wBAC1B,MAAMvB,YACJ;4BAAE,GAAGC,KAAK,CAAC,EAAE;4BAAE,GAAGA,KAAK,CAAC,EAAE;wBAAC,GAC3BQ,KAAK,KAAK,EACVA,KAAK,MAAM;oBAEf;gBACF;YACF;QACF,OAAO,IAAIc,AAAuB,WAAvBA,OAAO,WAAW,EAAa;YACxCC,OAAOD,OAAO,aAAa,CAAC,SAAS,EAAE;YACvCC,OAAOD,OAAO,aAAa,CAAC,OAAO,EAAE;YACrC,MAAMG,aAAaD,SAASF,OAAO,aAAa,CAAC,SAAS,EAAEd;YAC5D,MAAMkB,WAAWF,SAASF,OAAO,aAAa,CAAC,OAAO,EAAEd;YACxDa,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO;oBACL,MAAM;wBACJ,QAAQC,OAAO,OAAO,IAAI;wBAC1B,MAAMvB,YACJ;4BAAE,GAAG0B,UAAU,CAAC,EAAE;4BAAE,GAAGA,UAAU,CAAC,EAAE;wBAAC,GACrCjB,KAAK,KAAK,EACVA,KAAK,MAAM;oBAEf;oBACA,IAAI;wBACF,QAAQc,OAAO,OAAO,IAAI;wBAC1B,MAAMvB,YACJ;4BAAE,GAAG2B,QAAQ,CAAC,EAAE;4BAAE,GAAGA,QAAQ,CAAC,EAAE;wBAAC,GACjClB,KAAK,KAAK,EACVA,KAAK,MAAM;oBAEf;gBACF;gBACA,SAASc,OAAO,OAAO,IAAI;YAC7B;QACF,OAAO,IAAIA,AAAuB,WAAvBA,OAAO,WAAW,EAC3BD,iBAAiB,IAAI,CAAC;YACpB,MAAM;YACN,OAAO;gBACL,OAAOC,OAAO,aAAa,CAAC,OAAO;YACrC;YACA,SAASA,OAAO,OAAO,IAAI;QAC7B;aACK,IAAIA,AAAuB,aAAvBA,OAAO,WAAW,EAC3BD,iBAAiB,IAAI,CAAC;YACpB,MAAM;YACN,OAAO;gBACL,WAAWC,OAAO,aAAa,CAAC,SAAS;YAC3C;YACA,SAASA,OAAO,OAAO,IAAI;QAC7B;aACK,IAAIA,AAAuB,eAAvBA,OAAO,WAAW,EAC3BD,iBAAiB,IAAI,CAAC;YACpB,MAAM;YACN,OAAO,CAAC;YACR,SAASC,OAAO,OAAO,IAAI;QAC7B;aACK,IAAIA,AAAuB,aAAvBA,OAAO,WAAW,EAC3B,IAAKA,OAAO,aAAa,CAAC,GAAG,EAItB;YACL,MAAMK,OAAOC,qBAAqBN,OAAO,aAAa,CAAC,GAAG;YAE1DD,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO;oBACL,SAASM;gBACX;gBACA,SAASL,OAAO,OAAO,IAAI;YAC7B;QACF,OAbEO,QAAQ,IAAI,CACV;aAaC,IAAIP,AAAuB,WAAvBA,OAAO,WAAW,EAC3BD,iBAAiB,IAAI,CAAC;YACpB,MAAM;YACN,OAAO;gBACL,QAAQ;YACV;YACA,SAASC,OAAO,OAAO,IAAI;QAC7B;IAEJ;IAEA,IAAID,AAA4B,MAA5BA,iBAAiB,MAAM,EACzB,MAAM,IAAIS,MAAM,CAAC,4BAA4B,EAAEjB,IAAI,OAAO,EAAE,EAAE;QAC5D,OAAO;YACL,YAAYA,IAAI,OAAO;YACvBK;QACF;IACF;IAGFtB,MAAM,oBAAoBwB,KAAK,SAAS,CAACC,kBAAkB,MAAM;IAEjE,OAAO;QACL,SAASA;QACT,kBAAkBH;QAClB,gBAAgBa,WAAWlB,IAAI,OAAO;QACtC,OAAOA,IAAI,KAAK;QAChB,aAAaO,KAAK,SAAS,CAACP,IAAI,OAAO,EAAEmB,QAAW;IACtD;AACF;AAOA,SAASf,yBAAyBgB,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,IAAI3C,KAAK,KAAK,CAAEsC,AAAAA,CAAAA,QAAQG,KAAI,IAAK;QACvC,MAAMG,IAAI5C,KAAK,KAAK,CAAEwC,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,SAASX,SAASyB,QAAgB,EAAEzC,IAAuC;IACzE,MAAM,CAACsC,GAAGC,EAAE,GAAG3B,KAAK,KAAK,CAAC6B;IAC1B,OAAO;QAACH,IAAItC,KAAK,KAAK;QAAEuC,IAAIvC,KAAK,MAAM;KAAC;AAC1C;AAkEO,eAAe0C,qBACpBC,WAAmB,EACnB3C,IAAU,EACV4C,aAA6C;IAE7C,IAAIA,kBAAkBC,mBAAmB,IAAI,EAAE;QAC7CzD,MAAM,uCAAuCY;QAC7C,MAAM8C,gBAAgB9C,KAAK,KAAK,GAAGA,KAAK,MAAM;QAC9C,MAAM+C,YAAY;QAClB,IAAID,gBAAgBC,WAAW;YAC7B,MAAMC,eAAerD,KAAK,IAAI,CAACoD,YAAYD;YAC3C,MAAMG,WAAWtD,KAAK,KAAK,CAACK,KAAK,KAAK,GAAGgD;YACzC,MAAME,YAAYvD,KAAK,KAAK,CAACK,KAAK,MAAM,GAAGgD;YAC3C5D,MACE,2DACA6D,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":["webpack://@midscene/core/./src/ai-model/ui-tars-planning.ts"],"sourcesContent":["import type {\n  AIUsageInfo,\n  MidsceneYamlFlowItem,\n  PlanningAction,\n  Size,\n} from '@/types';\nimport {\n  type IModelPreferences,\n  UITarsModelVersion,\n  uiTarsModelVersion,\n  vlLocateMode,\n} 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 type { ChatCompletionMessageParam } from 'openai/resources/index';\nimport { AIActionType } from './common';\nimport { getSummary, getUiTarsPlanningPrompt } from './prompt/ui-tars-planning';\nimport { call } 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 vlmPlanning(options: {\n  userInstruction: string;\n  conversationHistory: ChatCompletionMessageParam[];\n  size: { width: number; height: number };\n  modelPreferences: IModelPreferences;\n}): Promise<{\n  actions: PlanningAction<any>[];\n  actionsFromModel: ReturnType<typeof actionParser>['parsed'];\n  action_summary: string;\n  yamlFlow?: MidsceneYamlFlowItem[];\n  usage?: AIUsageInfo;\n  rawResponse?: string;\n}> {\n  const { conversationHistory, userInstruction, size, modelPreferences } =\n    options;\n  const systemPrompt = getUiTarsPlanningPrompt() + userInstruction;\n\n  const res = await call(\n    [\n      {\n        role: 'user',\n        content: systemPrompt,\n      },\n      ...conversationHistory,\n    ],\n    AIActionType.INSPECT_ELEMENT,\n    modelPreferences,\n  );\n  const convertedText = convertBboxToCoordinates(res.content);\n\n  const modelVer = uiTarsModelVersion(modelPreferences);\n\n  const { parsed } = actionParser({\n    prediction: convertedText,\n    factor: [1000, 1000],\n    screenContext: {\n      width: size.width,\n      height: size.height,\n    },\n    modelVer: modelVer || undefined,\n  });\n\n  debug('ui-tars modelVer', modelVer, ', parsed', JSON.stringify(parsed));\n\n  const transformActions: PlanningAction[] = [];\n  parsed.forEach((action) => {\n    if (action.action_type === '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 (action.action_type === '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 (action.action_type === 'type') {\n      transformActions.push({\n        type: 'Input',\n        param: {\n          value: action.action_inputs.content,\n        },\n        thought: action.thought || '',\n      });\n    } else if (action.action_type === 'scroll') {\n      transformActions.push({\n        type: 'Scroll',\n        param: {\n          direction: action.action_inputs.direction,\n        },\n        thought: action.thought || '',\n      });\n    } else if (action.action_type === 'finished') {\n      transformActions.push({\n        type: 'Finished',\n        param: {},\n        thought: action.thought || '',\n      });\n    } else if (action.action_type === '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 (action.action_type === '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: ${res.content}`, {\n      cause: {\n        prediction: res.content,\n        parsed,\n      },\n    });\n  }\n\n  debug('transformActions', JSON.stringify(transformActions, null, 2));\n\n  return {\n    actions: transformActions,\n    actionsFromModel: parsed,\n    action_summary: getSummary(res.content),\n    usage: res.usage,\n    rawResponse: JSON.stringify(res.content, undefined, 2),\n  };\n}\n\n/**\n * Converts bounding box notation to coordinate points\n * @param text - The text containing bbox tags to be converted\n * @returns The text with bbox tags replaced by coordinate points\n */\nfunction convertBboxToCoordinates(text: string): string {\n  // Match the four numbers after <bbox>\n  const pattern = /<bbox>(\\d+)\\s+(\\d+)\\s+(\\d+)\\s+(\\d+)<\\/bbox>/g;\n\n  function replaceMatch(\n    match: string,\n    x1: string,\n    y1: string,\n    x2: string,\n    y2: string,\n  ): string {\n    // Convert strings to numbers and calculate center point\n    const x1Num = Number.parseInt(x1, 10);\n    const y1Num = Number.parseInt(y1, 10);\n    const x2Num = Number.parseInt(x2, 10);\n    const y2Num = Number.parseInt(y2, 10);\n\n    // Use Math.floor to truncate and calculate center point\n    const x = Math.floor((x1Num + x2Num) / 2);\n    const y = Math.floor((y1Num + y2Num) / 2);\n\n    // Return formatted coordinate string\n    return `(${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  modelPreferences: IModelPreferences,\n) {\n  if (\n    vlLocateMode(modelPreferences) === 'vlm-ui-tars' &&\n    uiTarsModelVersion(modelPreferences) === UITarsModelVersion.V1_5\n  ) {\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","vlmPlanning","options","conversationHistory","userInstruction","size","modelPreferences","systemPrompt","getUiTarsPlanningPrompt","res","call","AIActionType","convertedText","convertBboxToCoordinates","modelVer","uiTarsModelVersion","parsed","actionParser","undefined","JSON","transformActions","action","assert","getPoint","startPoint","endPoint","keys","transformHotkeyInput","console","Error","getSummary","text","pattern","replaceMatch","match","x1","y1","x2","y2","x1Num","Number","y1Num","x2Num","y2Num","x","y","cleanedText","startBox","resizeImageForUiTars","imageBase64","vlLocateMode","UITarsModelVersion","currentPixels","maxPixels","resizeFactor","newWidth","newHeight","resizedImage","resizeImgBase64"],"mappings":";;;;;;;;;AA8BA,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,YAAYC,OAKjC;IAQC,MAAM,EAAEC,mBAAmB,EAAEC,eAAe,EAAEC,IAAI,EAAEC,gBAAgB,EAAE,GACpEJ;IACF,MAAMK,eAAeC,4BAA4BJ;IAEjD,MAAMK,MAAM,MAAMC,KAChB;QACE;YACE,MAAM;YACN,SAASH;QACX;WACGJ;KACJ,EACDQ,aAAa,eAAe,EAC5BL;IAEF,MAAMM,gBAAgBC,yBAAyBJ,IAAI,OAAO;IAE1D,MAAMK,WAAWC,mBAAmBT;IAEpC,MAAM,EAAEU,MAAM,EAAE,GAAGC,aAAa;QAC9B,YAAYL;QACZ,QAAQ;YAAC;YAAM;SAAK;QACpB,eAAe;YACb,OAAOP,KAAK,KAAK;YACjB,QAAQA,KAAK,MAAM;QACrB;QACA,UAAUS,YAAYI;IACxB;IAEAzB,MAAM,oBAAoBqB,UAAU,YAAYK,KAAK,SAAS,CAACH;IAE/D,MAAMI,mBAAqC,EAAE;IAC7CJ,OAAO,OAAO,CAAC,CAACK;QACd,IAAIA,AAAuB,YAAvBA,OAAO,WAAW,EAAc;YAClCC,OAAOD,OAAO,aAAa,CAAC,SAAS,EAAE;YACvC,MAAMxB,QAAQ0B,SAASF,OAAO,aAAa,CAAC,SAAS,EAAEhB;YACvDe,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO;oBACL,QAAQ;wBACN,QAAQC,OAAO,OAAO,IAAI;wBAC1B,MAAMzB,YACJ;4BAAE,GAAGC,KAAK,CAAC,EAAE;4BAAE,GAAGA,KAAK,CAAC,EAAE;wBAAC,GAC3BQ,KAAK,KAAK,EACVA,KAAK,MAAM;oBAEf;gBACF;YACF;QACF,OAAO,IAAIgB,AAAuB,WAAvBA,OAAO,WAAW,EAAa;YACxCC,OAAOD,OAAO,aAAa,CAAC,SAAS,EAAE;YACvCC,OAAOD,OAAO,aAAa,CAAC,OAAO,EAAE;YACrC,MAAMG,aAAaD,SAASF,OAAO,aAAa,CAAC,SAAS,EAAEhB;YAC5D,MAAMoB,WAAWF,SAASF,OAAO,aAAa,CAAC,OAAO,EAAEhB;YACxDe,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO;oBACL,MAAM;wBACJ,QAAQC,OAAO,OAAO,IAAI;wBAC1B,MAAMzB,YACJ;4BAAE,GAAG4B,UAAU,CAAC,EAAE;4BAAE,GAAGA,UAAU,CAAC,EAAE;wBAAC,GACrCnB,KAAK,KAAK,EACVA,KAAK,MAAM;oBAEf;oBACA,IAAI;wBACF,QAAQgB,OAAO,OAAO,IAAI;wBAC1B,MAAMzB,YACJ;4BAAE,GAAG6B,QAAQ,CAAC,EAAE;4BAAE,GAAGA,QAAQ,CAAC,EAAE;wBAAC,GACjCpB,KAAK,KAAK,EACVA,KAAK,MAAM;oBAEf;gBACF;gBACA,SAASgB,OAAO,OAAO,IAAI;YAC7B;QACF,OAAO,IAAIA,AAAuB,WAAvBA,OAAO,WAAW,EAC3BD,iBAAiB,IAAI,CAAC;YACpB,MAAM;YACN,OAAO;gBACL,OAAOC,OAAO,aAAa,CAAC,OAAO;YACrC;YACA,SAASA,OAAO,OAAO,IAAI;QAC7B;aACK,IAAIA,AAAuB,aAAvBA,OAAO,WAAW,EAC3BD,iBAAiB,IAAI,CAAC;YACpB,MAAM;YACN,OAAO;gBACL,WAAWC,OAAO,aAAa,CAAC,SAAS;YAC3C;YACA,SAASA,OAAO,OAAO,IAAI;QAC7B;aACK,IAAIA,AAAuB,eAAvBA,OAAO,WAAW,EAC3BD,iBAAiB,IAAI,CAAC;YACpB,MAAM;YACN,OAAO,CAAC;YACR,SAASC,OAAO,OAAO,IAAI;QAC7B;aACK,IAAIA,AAAuB,aAAvBA,OAAO,WAAW,EAC3B,IAAKA,OAAO,aAAa,CAAC,GAAG,EAItB;YACL,MAAMK,OAAOC,qBAAqBN,OAAO,aAAa,CAAC,GAAG;YAE1DD,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO;oBACL,SAASM;gBACX;gBACA,SAASL,OAAO,OAAO,IAAI;YAC7B;QACF,OAbEO,QAAQ,IAAI,CACV;aAaC,IAAIP,AAAuB,WAAvBA,OAAO,WAAW,EAC3BD,iBAAiB,IAAI,CAAC;YACpB,MAAM;YACN,OAAO;gBACL,QAAQ;YACV;YACA,SAASC,OAAO,OAAO,IAAI;QAC7B;IAEJ;IAEA,IAAID,AAA4B,MAA5BA,iBAAiB,MAAM,EACzB,MAAM,IAAIS,MAAM,CAAC,4BAA4B,EAAEpB,IAAI,OAAO,EAAE,EAAE;QAC5D,OAAO;YACL,YAAYA,IAAI,OAAO;YACvBO;QACF;IACF;IAGFvB,MAAM,oBAAoB0B,KAAK,SAAS,CAACC,kBAAkB,MAAM;IAEjE,OAAO;QACL,SAASA;QACT,kBAAkBJ;QAClB,gBAAgBc,WAAWrB,IAAI,OAAO;QACtC,OAAOA,IAAI,KAAK;QAChB,aAAaU,KAAK,SAAS,CAACV,IAAI,OAAO,EAAES,QAAW;IACtD;AACF;AAOA,SAASL,yBAAyBkB,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,IAAI5C,KAAK,KAAK,CAAEuC,AAAAA,CAAAA,QAAQG,KAAI,IAAK;QACvC,MAAMG,IAAI7C,KAAK,KAAK,CAAEyC,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,SAASV,SAASwB,QAAgB,EAAE1C,IAAuC;IACzE,MAAM,CAACuC,GAAGC,EAAE,GAAG1B,KAAK,KAAK,CAAC4B;IAC1B,OAAO;QAACH,IAAIvC,KAAK,KAAK;QAAEwC,IAAIxC,KAAK,MAAM;KAAC;AAC1C;AAkEO,eAAe2C,qBACpBC,WAAmB,EACnB5C,IAAU,EACVC,gBAAmC;IAEnC,IACE4C,AAAmC,kBAAnCA,aAAa5C,qBACbS,mBAAmBT,sBAAsB6C,mBAAmB,IAAI,EAChE;QACA1D,MAAM,uCAAuCY;QAC7C,MAAM+C,gBAAgB/C,KAAK,KAAK,GAAGA,KAAK,MAAM;QAC9C,MAAMgD,YAAY;QAClB,IAAID,gBAAgBC,WAAW;YAC7B,MAAMC,eAAetD,KAAK,IAAI,CAACqD,YAAYD;YAC3C,MAAMG,WAAWvD,KAAK,KAAK,CAACK,KAAK,KAAK,GAAGiD;YACzC,MAAME,YAAYxD,KAAK,KAAK,CAACK,KAAK,MAAM,GAAGiD;YAC3C7D,MACE,2DACA8D,UACAC;YAEF,MAAMC,eAAe,MAAMC,gBAAgBT,aAAa;gBACtD,OAAOM;gBACP,QAAQC;YACV;YACA,OAAOC;QACT;IACF;IACA,OAAOR;AACT"}