@midscene/core 1.0.1-beta-20251107065041.0 → 1.0.1-beta-20251110055007.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/es/agent/task-builder.mjs +10 -12
- package/dist/es/agent/task-builder.mjs.map +1 -1
- package/dist/es/agent/utils.mjs +5 -3
- package/dist/es/agent/utils.mjs.map +1 -1
- package/dist/es/ai-model/common.mjs +1 -0
- package/dist/es/ai-model/common.mjs.map +1 -1
- package/dist/es/device/index.mjs.map +1 -1
- package/dist/es/service/index.mjs +1 -3
- package/dist/es/service/index.mjs.map +1 -1
- package/dist/es/utils.mjs +2 -2
- package/dist/lib/agent/task-builder.js +9 -11
- package/dist/lib/agent/task-builder.js.map +1 -1
- package/dist/lib/agent/utils.js +8 -3
- package/dist/lib/agent/utils.js.map +1 -1
- package/dist/lib/ai-model/common.js +1 -0
- package/dist/lib/ai-model/common.js.map +1 -1
- package/dist/lib/device/index.js.map +1 -1
- package/dist/lib/service/index.js +1 -3
- package/dist/lib/service/index.js.map +1 -1
- package/dist/lib/utils.js +2 -2
- package/dist/types/agent/utils.d.ts +1 -0
- package/dist/types/ai-model/common.d.ts +1 -1
- package/dist/types/device/index.d.ts +2 -2
- package/package.json +2 -2
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"service/index.mjs","sources":["webpack://@midscene/core/./src/service/index.ts"],"sourcesContent":["import { AIActionType, type AIArgs, expandSearchArea } from '@/ai-model/common';\nimport {\n AiExtractElementInfo,\n AiLocateElement,\n callAIWithObjectResponse,\n} from '@/ai-model/index';\nimport { AiLocateSection } from '@/ai-model/inspect';\nimport { elementDescriberInstruction } from '@/ai-model/prompt/describe';\nimport type {\n AIDescribeElementResponse,\n AIUsageInfo,\n DetailedLocateParam,\n LocateResultWithDump,\n PartialServiceDumpFromSDK,\n Rect,\n ServiceExtractOption,\n ServiceExtractParam,\n ServiceExtractResult,\n ServiceTaskInfo,\n UIContext,\n} from '@/types';\nimport { ServiceError } from '@/types';\nimport {\n type IModelConfig,\n MIDSCENE_FORCE_DEEP_THINK,\n globalConfigManager,\n} from '@midscene/shared/env';\nimport { compositeElementInfoImg, cropByRect } from '@midscene/shared/img';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport type { TMultimodalPrompt } from '../ai-model/common';\nimport { createServiceDump } from './utils';\n\nexport interface LocateOpts {\n context?: UIContext;\n}\n\nexport type AnyValue<T> = {\n [K in keyof T]: unknown extends T[K] ? any : T[K];\n};\n\ninterface ServiceOptions {\n taskInfo?: Omit<ServiceTaskInfo, 'durationMs'>;\n aiVendorFn?: typeof callAIWithObjectResponse;\n}\n\nconst debug = getDebug('ai:service');\nexport default class Service {\n contextRetrieverFn: () => Promise<UIContext> | UIContext;\n\n aiVendorFn: Exclude<ServiceOptions['aiVendorFn'], undefined> =\n callAIWithObjectResponse;\n\n taskInfo?: Omit<ServiceTaskInfo, 'durationMs'>;\n\n constructor(\n context: UIContext | (() => Promise<UIContext> | UIContext),\n opt?: ServiceOptions,\n ) {\n assert(context, 'context is required for Service');\n if (typeof context === 'function') {\n this.contextRetrieverFn = context;\n } else {\n this.contextRetrieverFn = () => Promise.resolve(context);\n }\n\n // just for unit test, aiVendorFn is callAIWithObjectResponse by default\n if (typeof opt?.aiVendorFn !== 'undefined') {\n this.aiVendorFn = opt.aiVendorFn;\n }\n if (typeof opt?.taskInfo !== 'undefined') {\n this.taskInfo = opt.taskInfo;\n }\n }\n\n async locate(\n query: DetailedLocateParam,\n opt: LocateOpts,\n modelConfig: IModelConfig,\n ): Promise<LocateResultWithDump> {\n const queryPrompt = typeof query === 'string' ? query : query.prompt;\n assert(queryPrompt, 'query is required for locate');\n\n assert(typeof query === 'object', 'query should be an object for locate');\n\n const globalDeepThinkSwitch = globalConfigManager.getEnvConfigInBoolean(\n MIDSCENE_FORCE_DEEP_THINK,\n );\n if (globalDeepThinkSwitch) {\n debug('globalDeepThinkSwitch', globalDeepThinkSwitch);\n }\n let searchAreaPrompt;\n if (query.deepThink || globalDeepThinkSwitch) {\n searchAreaPrompt = query.prompt;\n }\n\n const { vlMode } = modelConfig;\n\n if (searchAreaPrompt && !vlMode) {\n console.warn(\n 'The \"deepThink\" feature is not supported with multimodal LLM. Please config VL model for Midscene. https://midscenejs.com/choose-a-model',\n );\n searchAreaPrompt = undefined;\n }\n\n const context = opt?.context || (await this.contextRetrieverFn());\n\n let searchArea: Rect | undefined = undefined;\n let searchAreaRawResponse: string | undefined = undefined;\n let searchAreaUsage: AIUsageInfo | undefined = undefined;\n let searchAreaResponse:\n | Awaited<ReturnType<typeof AiLocateSection>>\n | undefined = undefined;\n if (searchAreaPrompt) {\n searchAreaResponse = await AiLocateSection({\n context,\n sectionDescription: searchAreaPrompt,\n modelConfig,\n });\n assert(\n searchAreaResponse.rect,\n `cannot find search area for \"${searchAreaPrompt}\"${\n searchAreaResponse.error ? `: ${searchAreaResponse.error}` : ''\n }`,\n );\n searchAreaRawResponse = searchAreaResponse.rawResponse;\n searchAreaUsage = searchAreaResponse.usage;\n searchArea = searchAreaResponse.rect;\n }\n\n const startTime = Date.now();\n const { parseResult, rect, rawResponse, usage } = await AiLocateElement({\n callAIFn: this.aiVendorFn,\n context,\n targetElementDescription: queryPrompt,\n searchConfig: searchAreaResponse,\n modelConfig,\n });\n\n const timeCost = Date.now() - startTime;\n const taskInfo: ServiceTaskInfo = {\n ...(this.taskInfo ? this.taskInfo : {}),\n durationMs: timeCost,\n rawResponse: JSON.stringify(rawResponse),\n formatResponse: JSON.stringify(parseResult),\n usage,\n searchArea,\n searchAreaRawResponse,\n searchAreaUsage,\n };\n\n let errorLog: string | undefined;\n if (parseResult.errors?.length) {\n errorLog = `AI model failed to locate: \\n${parseResult.errors.join('\\n')}`;\n }\n\n const dumpData: PartialServiceDumpFromSDK = {\n type: 'locate',\n userQuery: {\n element: queryPrompt,\n },\n matchedElement: [],\n matchedRect: rect,\n data: null,\n taskInfo,\n deepThink: !!searchArea,\n error: errorLog,\n };\n\n const elements = parseResult.elements || [];\n\n const dump = createServiceDump({\n ...dumpData,\n matchedElement: elements,\n });\n\n if (errorLog) {\n throw new ServiceError(errorLog, dump);\n }\n\n if (elements.length > 1) {\n throw new ServiceError(\n `locate: multiple elements found, length = ${elements.length}`,\n dump,\n );\n }\n\n if (elements.length === 1) {\n return {\n element: {\n id: elements[0]!.id,\n center: elements[0]!.center,\n rect: elements[0]!.rect,\n isOrderSensitive: elements[0]!.isOrderSensitive,\n },\n rect,\n dump,\n };\n }\n\n return {\n element: null,\n rect,\n dump,\n };\n }\n\n async extract<T>(\n dataDemand: ServiceExtractParam,\n modelConfig: IModelConfig,\n opt?: ServiceExtractOption,\n pageDescription?: string,\n multimodalPrompt?: TMultimodalPrompt,\n ): Promise<ServiceExtractResult<T>> {\n assert(\n typeof dataDemand === 'object' || typeof dataDemand === 'string',\n `dataDemand should be object or string, but get ${typeof dataDemand}`,\n );\n const context = await this.contextRetrieverFn();\n\n const startTime = Date.now();\n\n const { parseResult, usage } = await AiExtractElementInfo<T>({\n context,\n dataQuery: dataDemand,\n multimodalPrompt,\n extractOption: opt,\n modelConfig,\n pageDescription,\n });\n\n const timeCost = Date.now() - startTime;\n const taskInfo: ServiceTaskInfo = {\n ...(this.taskInfo ? this.taskInfo : {}),\n durationMs: timeCost,\n rawResponse: JSON.stringify(parseResult),\n };\n\n let errorLog: string | undefined;\n if (parseResult.errors?.length) {\n errorLog = `AI response error: \\n${parseResult.errors.join('\\n')}`;\n }\n\n const dumpData: PartialServiceDumpFromSDK = {\n type: 'extract',\n userQuery: {\n dataDemand,\n },\n matchedElement: [],\n data: null,\n taskInfo,\n error: errorLog,\n };\n\n const { data, thought } = parseResult || {};\n\n // 4\n const dump = createServiceDump({\n ...dumpData,\n data,\n });\n\n if (errorLog && !data) {\n throw new ServiceError(errorLog, dump);\n }\n\n return {\n data,\n thought,\n usage,\n dump,\n };\n }\n\n async describe(\n target: Rect | [number, number],\n modelConfig: IModelConfig,\n opt?: {\n deepThink?: boolean;\n },\n ): Promise<Pick<AIDescribeElementResponse, 'description'>> {\n assert(target, 'target is required for service.describe');\n const context = await this.contextRetrieverFn();\n const { screenshotBase64, size } = context;\n assert(screenshotBase64, 'screenshot is required for service.describe');\n // The result of the \"describe\" function will be used for positioning, so essentially it is a form of grounding.\n const { vlMode } = modelConfig;\n const systemPrompt = elementDescriberInstruction();\n\n // Convert [x,y] center point to Rect if needed\n const defaultRectSize = 30;\n const targetRect: Rect = Array.isArray(target)\n ? {\n left: Math.floor(target[0] - defaultRectSize / 2),\n top: Math.floor(target[1] - defaultRectSize / 2),\n width: defaultRectSize,\n height: defaultRectSize,\n }\n : target;\n\n let imagePayload = await compositeElementInfoImg({\n inputImgBase64: screenshotBase64,\n size,\n elementsPositionInfo: [\n {\n rect: targetRect,\n },\n ],\n borderThickness: 3,\n });\n\n if (opt?.deepThink) {\n const searchArea = expandSearchArea(targetRect, context.size, vlMode);\n debug('describe: set searchArea', searchArea);\n const croppedResult = await cropByRect(\n imagePayload,\n searchArea,\n vlMode === 'qwen-vl',\n );\n imagePayload = croppedResult.imageBase64;\n }\n\n const msgs: AIArgs = [\n { role: 'system', content: systemPrompt },\n {\n role: 'user',\n content: [\n {\n type: 'image_url',\n image_url: {\n url: imagePayload,\n detail: 'high',\n },\n },\n ],\n },\n ];\n\n const callAIFn = this\n .aiVendorFn as typeof callAIWithObjectResponse<AIDescribeElementResponse>;\n\n const res = await callAIFn(\n msgs,\n AIActionType.DESCRIBE_ELEMENT,\n modelConfig,\n );\n\n const { content } = res;\n assert(!content.error, `describe failed: ${content.error}`);\n assert(content.description, 'failed to describe the element');\n return content;\n }\n}\n"],"names":["debug","getDebug","Service","query","opt","modelConfig","_parseResult_errors","queryPrompt","assert","globalDeepThinkSwitch","globalConfigManager","MIDSCENE_FORCE_DEEP_THINK","searchAreaPrompt","vlMode","console","undefined","context","searchArea","searchAreaRawResponse","searchAreaUsage","searchAreaResponse","AiLocateSection","startTime","Date","parseResult","rect","rawResponse","usage","AiLocateElement","timeCost","taskInfo","JSON","errorLog","dumpData","elements","dump","createServiceDump","ServiceError","dataDemand","pageDescription","multimodalPrompt","AiExtractElementInfo","data","thought","target","screenshotBase64","size","systemPrompt","elementDescriberInstruction","defaultRectSize","targetRect","Array","Math","imagePayload","compositeElementInfoImg","expandSearchArea","croppedResult","cropByRect","msgs","callAIFn","res","AIActionType","content","callAIWithObjectResponse","Promise"],"mappings":";;;;;;;;;;;;;;;;;;;;AA8CA,MAAMA,QAAQC,SAAS;AACR,MAAMC;IA4BnB,MAAM,OACJC,KAA0B,EAC1BC,GAAe,EACfC,WAAyB,EACM;YAyE3BC;QAxEJ,MAAMC,cAAc,AAAiB,YAAjB,OAAOJ,QAAqBA,QAAQA,MAAM,MAAM;QACpEK,OAAOD,aAAa;QAEpBC,OAAO,AAAiB,YAAjB,OAAOL,OAAoB;QAElC,MAAMM,wBAAwBC,oBAAoB,qBAAqB,CACrEC;QAEF,IAAIF,uBACFT,MAAM,yBAAyBS;QAEjC,IAAIG;QACJ,IAAIT,MAAM,SAAS,IAAIM,uBACrBG,mBAAmBT,MAAM,MAAM;QAGjC,MAAM,EAAEU,MAAM,EAAE,GAAGR;QAEnB,IAAIO,oBAAoB,CAACC,QAAQ;YAC/BC,QAAQ,IAAI,CACV;YAEFF,mBAAmBG;QACrB;QAEA,MAAMC,UAAUZ,AAAAA,CAAAA,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,OAAO,AAAD,KAAM,MAAM,IAAI,CAAC,kBAAkB;QAE9D,IAAIa;QACJ,IAAIC;QACJ,IAAIC;QACJ,IAAIC;QAGJ,IAAIR,kBAAkB;YACpBQ,qBAAqB,MAAMC,gBAAgB;gBACzCL;gBACA,oBAAoBJ;gBACpBP;YACF;YACAG,OACEY,mBAAmB,IAAI,EACvB,CAAC,6BAA6B,EAAER,iBAAiB,CAAC,EAChDQ,mBAAmB,KAAK,GAAG,CAAC,EAAE,EAAEA,mBAAmB,KAAK,EAAE,GAAG,IAC7D;YAEJF,wBAAwBE,mBAAmB,WAAW;YACtDD,kBAAkBC,mBAAmB,KAAK;YAC1CH,aAAaG,mBAAmB,IAAI;QACtC;QAEA,MAAME,YAAYC,KAAK,GAAG;QAC1B,MAAM,EAAEC,WAAW,EAAEC,IAAI,EAAEC,WAAW,EAAEC,KAAK,EAAE,GAAG,MAAMC,gBAAgB;YACtE,UAAU,IAAI,CAAC,UAAU;YACzBZ;YACA,0BAA0BT;YAC1B,cAAca;YACdf;QACF;QAEA,MAAMwB,WAAWN,KAAK,GAAG,KAAKD;QAC9B,MAAMQ,WAA4B;YAChC,GAAI,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC,QAAQ,GAAG,CAAC,CAAC;YACtC,YAAYD;YACZ,aAAaE,KAAK,SAAS,CAACL;YAC5B,gBAAgBK,KAAK,SAAS,CAACP;YAC/BG;YACAV;YACAC;YACAC;QACF;QAEA,IAAIa;QACJ,IAAI,QAAA1B,CAAAA,sBAAAA,YAAY,MAAM,AAAD,IAAjBA,KAAAA,IAAAA,oBAAoB,MAAM,EAC5B0B,WAAW,CAAC,6BAA6B,EAAER,YAAY,MAAM,CAAC,IAAI,CAAC,OAAO;QAG5E,MAAMS,WAAsC;YAC1C,MAAM;YACN,WAAW;gBACT,SAAS1B;YACX;YACA,gBAAgB,EAAE;YAClB,aAAakB;YACb,MAAM;YACNK;YACA,WAAW,CAAC,CAACb;YACb,OAAOe;QACT;QAEA,MAAME,WAAWV,YAAY,QAAQ,IAAI,EAAE;QAE3C,MAAMW,OAAOC,kBAAkB;YAC7B,GAAGH,QAAQ;YACX,gBAAgBC;QAClB;QAEA,IAAIF,UACF,MAAM,IAAIK,aAAaL,UAAUG;QAGnC,IAAID,SAAS,MAAM,GAAG,GACpB,MAAM,IAAIG,aACR,CAAC,0CAA0C,EAAEH,SAAS,MAAM,EAAE,EAC9DC;QAIJ,IAAID,AAAoB,MAApBA,SAAS,MAAM,EACjB,OAAO;YACL,SAAS;gBACP,IAAIA,QAAQ,CAAC,EAAE,CAAE,EAAE;gBACnB,QAAQA,QAAQ,CAAC,EAAE,CAAE,MAAM;gBAC3B,MAAMA,QAAQ,CAAC,EAAE,CAAE,IAAI;gBACvB,kBAAkBA,QAAQ,CAAC,EAAE,CAAE,gBAAgB;YACjD;YACAT;YACAU;QACF;QAGF,OAAO;YACL,SAAS;YACTV;YACAU;QACF;IACF;IAEA,MAAM,QACJG,UAA+B,EAC/BjC,WAAyB,EACzBD,GAA0B,EAC1BmC,eAAwB,EACxBC,gBAAoC,EACF;YA0B9BlC;QAzBJE,OACE,AAAsB,YAAtB,OAAO8B,cAA2B,AAAsB,YAAtB,OAAOA,YACzC,CAAC,+CAA+C,EAAE,OAAOA,YAAY;QAEvE,MAAMtB,UAAU,MAAM,IAAI,CAAC,kBAAkB;QAE7C,MAAMM,YAAYC,KAAK,GAAG;QAE1B,MAAM,EAAEC,WAAW,EAAEG,KAAK,EAAE,GAAG,MAAMc,qBAAwB;YAC3DzB;YACA,WAAWsB;YACXE;YACA,eAAepC;YACfC;YACAkC;QACF;QAEA,MAAMV,WAAWN,KAAK,GAAG,KAAKD;QAC9B,MAAMQ,WAA4B;YAChC,GAAI,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC,QAAQ,GAAG,CAAC,CAAC;YACtC,YAAYD;YACZ,aAAaE,KAAK,SAAS,CAACP;QAC9B;QAEA,IAAIQ;QACJ,IAAI,QAAA1B,CAAAA,sBAAAA,YAAY,MAAM,AAAD,IAAjBA,KAAAA,IAAAA,oBAAoB,MAAM,EAC5B0B,WAAW,CAAC,qBAAqB,EAAER,YAAY,MAAM,CAAC,IAAI,CAAC,OAAO;QAGpE,MAAMS,WAAsC;YAC1C,MAAM;YACN,WAAW;gBACTK;YACF;YACA,gBAAgB,EAAE;YAClB,MAAM;YACNR;YACA,OAAOE;QACT;QAEA,MAAM,EAAEU,IAAI,EAAEC,OAAO,EAAE,GAAGnB,eAAe,CAAC;QAG1C,MAAMW,OAAOC,kBAAkB;YAC7B,GAAGH,QAAQ;YACXS;QACF;QAEA,IAAIV,YAAY,CAACU,MACf,MAAM,IAAIL,aAAaL,UAAUG;QAGnC,OAAO;YACLO;YACAC;YACAhB;YACAQ;QACF;IACF;IAEA,MAAM,SACJS,MAA+B,EAC/BvC,WAAyB,EACzBD,GAEC,EACwD;QACzDI,OAAOoC,QAAQ;QACf,MAAM5B,UAAU,MAAM,IAAI,CAAC,kBAAkB;QAC7C,MAAM,EAAE6B,gBAAgB,EAAEC,IAAI,EAAE,GAAG9B;QACnCR,OAAOqC,kBAAkB;QAEzB,MAAM,EAAEhC,MAAM,EAAE,GAAGR;QACnB,MAAM0C,eAAeC;QAGrB,MAAMC,kBAAkB;QACxB,MAAMC,aAAmBC,MAAM,OAAO,CAACP,UACnC;YACE,MAAMQ,KAAK,KAAK,CAACR,MAAM,CAAC,EAAE,GAAGK,kBAAkB;YAC/C,KAAKG,KAAK,KAAK,CAACR,MAAM,CAAC,EAAE,GAAGK,kBAAkB;YAC9C,OAAOA;YACP,QAAQA;QACV,IACAL;QAEJ,IAAIS,eAAe,MAAMC,wBAAwB;YAC/C,gBAAgBT;YAChBC;YACA,sBAAsB;gBACpB;oBACE,MAAMI;gBACR;aACD;YACD,iBAAiB;QACnB;QAEA,IAAI9C,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,SAAS,EAAE;YAClB,MAAMa,aAAasC,iBAAiBL,YAAYlC,QAAQ,IAAI,EAAEH;YAC9Db,MAAM,4BAA4BiB;YAClC,MAAMuC,gBAAgB,MAAMC,WAC1BJ,cACApC,YACAJ,AAAW,cAAXA;YAEFwC,eAAeG,cAAc,WAAW;QAC1C;QAEA,MAAME,OAAe;YACnB;gBAAE,MAAM;gBAAU,SAASX;YAAa;YACxC;gBACE,MAAM;gBACN,SAAS;oBACP;wBACE,MAAM;wBACN,WAAW;4BACT,KAAKM;4BACL,QAAQ;wBACV;oBACF;iBACD;YACH;SACD;QAED,MAAMM,WAAW,IAAI,CAClB,UAAU;QAEb,MAAMC,MAAM,MAAMD,SAChBD,MACAG,aAAa,gBAAgB,EAC7BxD;QAGF,MAAM,EAAEyD,OAAO,EAAE,GAAGF;QACpBpD,OAAO,CAACsD,QAAQ,KAAK,EAAE,CAAC,iBAAiB,EAAEA,QAAQ,KAAK,EAAE;QAC1DtD,OAAOsD,QAAQ,WAAW,EAAE;QAC5B,OAAOA;IACT;IAxSA,YACE9C,OAA2D,EAC3DZ,GAAoB,CACpB;QAVF;QAEA,qCACE2D;QAEF;QAMEvD,OAAOQ,SAAS;QAChB,IAAI,AAAmB,cAAnB,OAAOA,SACT,IAAI,CAAC,kBAAkB,GAAGA;aAE1B,IAAI,CAAC,kBAAkB,GAAG,IAAMgD,QAAQ,OAAO,CAAChD;QAIlD,IAAI,AAA2B,WAApBZ,CAAAA,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,UAAU,AAAD,GACvB,IAAI,CAAC,UAAU,GAAGA,IAAI,UAAU;QAElC,IAAI,AAAyB,WAAlBA,CAAAA,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,QAAQ,AAAD,GACrB,IAAI,CAAC,QAAQ,GAAGA,IAAI,QAAQ;IAEhC;AAuRF"}
|
|
1
|
+
{"version":3,"file":"service/index.mjs","sources":["webpack://@midscene/core/./src/service/index.ts"],"sourcesContent":["import { AIActionType, type AIArgs, expandSearchArea } from '@/ai-model/common';\nimport {\n AiExtractElementInfo,\n AiLocateElement,\n callAIWithObjectResponse,\n} from '@/ai-model/index';\nimport { AiLocateSection } from '@/ai-model/inspect';\nimport { elementDescriberInstruction } from '@/ai-model/prompt/describe';\nimport type {\n AIDescribeElementResponse,\n AIUsageInfo,\n DetailedLocateParam,\n LocateResultWithDump,\n PartialServiceDumpFromSDK,\n Rect,\n ServiceExtractOption,\n ServiceExtractParam,\n ServiceExtractResult,\n ServiceTaskInfo,\n UIContext,\n} from '@/types';\nimport { ServiceError } from '@/types';\nimport {\n type IModelConfig,\n MIDSCENE_FORCE_DEEP_THINK,\n globalConfigManager,\n} from '@midscene/shared/env';\nimport { compositeElementInfoImg, cropByRect } from '@midscene/shared/img';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport type { TMultimodalPrompt } from '../ai-model/common';\nimport { createServiceDump } from './utils';\n\nexport interface LocateOpts {\n context?: UIContext;\n}\n\nexport type AnyValue<T> = {\n [K in keyof T]: unknown extends T[K] ? any : T[K];\n};\n\ninterface ServiceOptions {\n taskInfo?: Omit<ServiceTaskInfo, 'durationMs'>;\n aiVendorFn?: typeof callAIWithObjectResponse;\n}\n\nconst debug = getDebug('ai:service');\nexport default class Service {\n contextRetrieverFn: () => Promise<UIContext> | UIContext;\n\n aiVendorFn: Exclude<ServiceOptions['aiVendorFn'], undefined> =\n callAIWithObjectResponse;\n\n taskInfo?: Omit<ServiceTaskInfo, 'durationMs'>;\n\n constructor(\n context: UIContext | (() => Promise<UIContext> | UIContext),\n opt?: ServiceOptions,\n ) {\n assert(context, 'context is required for Service');\n if (typeof context === 'function') {\n this.contextRetrieverFn = context;\n } else {\n this.contextRetrieverFn = () => Promise.resolve(context);\n }\n\n // just for unit test, aiVendorFn is callAIWithObjectResponse by default\n if (typeof opt?.aiVendorFn !== 'undefined') {\n this.aiVendorFn = opt.aiVendorFn;\n }\n if (typeof opt?.taskInfo !== 'undefined') {\n this.taskInfo = opt.taskInfo;\n }\n }\n\n async locate(\n query: DetailedLocateParam,\n opt: LocateOpts,\n modelConfig: IModelConfig,\n ): Promise<LocateResultWithDump> {\n const queryPrompt = typeof query === 'string' ? query : query.prompt;\n assert(queryPrompt, 'query is required for locate');\n\n assert(typeof query === 'object', 'query should be an object for locate');\n\n const globalDeepThinkSwitch = globalConfigManager.getEnvConfigInBoolean(\n MIDSCENE_FORCE_DEEP_THINK,\n );\n if (globalDeepThinkSwitch) {\n debug('globalDeepThinkSwitch', globalDeepThinkSwitch);\n }\n let searchAreaPrompt;\n if (query.deepThink || globalDeepThinkSwitch) {\n searchAreaPrompt = query.prompt;\n }\n\n const { vlMode } = modelConfig;\n\n if (searchAreaPrompt && !vlMode) {\n console.warn(\n 'The \"deepThink\" feature is not supported with multimodal LLM. Please config VL model for Midscene. https://midscenejs.com/choose-a-model',\n );\n searchAreaPrompt = undefined;\n }\n\n const context = opt?.context || (await this.contextRetrieverFn());\n\n let searchArea: Rect | undefined = undefined;\n let searchAreaRawResponse: string | undefined = undefined;\n let searchAreaUsage: AIUsageInfo | undefined = undefined;\n let searchAreaResponse:\n | Awaited<ReturnType<typeof AiLocateSection>>\n | undefined = undefined;\n if (searchAreaPrompt) {\n searchAreaResponse = await AiLocateSection({\n context,\n sectionDescription: searchAreaPrompt,\n modelConfig,\n });\n assert(\n searchAreaResponse.rect,\n `cannot find search area for \"${searchAreaPrompt}\"${\n searchAreaResponse.error ? `: ${searchAreaResponse.error}` : ''\n }`,\n );\n searchAreaRawResponse = searchAreaResponse.rawResponse;\n searchAreaUsage = searchAreaResponse.usage;\n searchArea = searchAreaResponse.rect;\n }\n\n const startTime = Date.now();\n const { parseResult, rect, rawResponse, usage } = await AiLocateElement({\n callAIFn: this.aiVendorFn,\n context,\n targetElementDescription: queryPrompt,\n searchConfig: searchAreaResponse,\n modelConfig,\n });\n\n const timeCost = Date.now() - startTime;\n const taskInfo: ServiceTaskInfo = {\n ...(this.taskInfo ? this.taskInfo : {}),\n durationMs: timeCost,\n rawResponse: JSON.stringify(rawResponse),\n formatResponse: JSON.stringify(parseResult),\n usage,\n searchArea,\n searchAreaRawResponse,\n searchAreaUsage,\n };\n\n let errorLog: string | undefined;\n if (parseResult.errors?.length) {\n errorLog = `AI model failed to locate: \\n${parseResult.errors.join('\\n')}`;\n }\n\n const dumpData: PartialServiceDumpFromSDK = {\n type: 'locate',\n userQuery: {\n element: queryPrompt,\n },\n matchedElement: [],\n matchedRect: rect,\n data: null,\n taskInfo,\n deepThink: !!searchArea,\n error: errorLog,\n };\n\n const elements = parseResult.elements || [];\n\n const dump = createServiceDump({\n ...dumpData,\n matchedElement: elements,\n });\n\n if (errorLog) {\n throw new ServiceError(errorLog, dump);\n }\n\n if (elements.length > 1) {\n throw new ServiceError(\n `locate: multiple elements found, length = ${elements.length}`,\n dump,\n );\n }\n\n if (elements.length === 1) {\n return {\n element: {\n center: elements[0]!.center,\n rect: elements[0]!.rect,\n },\n rect,\n dump,\n };\n }\n\n return {\n element: null,\n rect,\n dump,\n };\n }\n\n async extract<T>(\n dataDemand: ServiceExtractParam,\n modelConfig: IModelConfig,\n opt?: ServiceExtractOption,\n pageDescription?: string,\n multimodalPrompt?: TMultimodalPrompt,\n ): Promise<ServiceExtractResult<T>> {\n assert(\n typeof dataDemand === 'object' || typeof dataDemand === 'string',\n `dataDemand should be object or string, but get ${typeof dataDemand}`,\n );\n const context = await this.contextRetrieverFn();\n\n const startTime = Date.now();\n\n const { parseResult, usage } = await AiExtractElementInfo<T>({\n context,\n dataQuery: dataDemand,\n multimodalPrompt,\n extractOption: opt,\n modelConfig,\n pageDescription,\n });\n\n const timeCost = Date.now() - startTime;\n const taskInfo: ServiceTaskInfo = {\n ...(this.taskInfo ? this.taskInfo : {}),\n durationMs: timeCost,\n rawResponse: JSON.stringify(parseResult),\n };\n\n let errorLog: string | undefined;\n if (parseResult.errors?.length) {\n errorLog = `AI response error: \\n${parseResult.errors.join('\\n')}`;\n }\n\n const dumpData: PartialServiceDumpFromSDK = {\n type: 'extract',\n userQuery: {\n dataDemand,\n },\n matchedElement: [],\n data: null,\n taskInfo,\n error: errorLog,\n };\n\n const { data, thought } = parseResult || {};\n\n // 4\n const dump = createServiceDump({\n ...dumpData,\n data,\n });\n\n if (errorLog && !data) {\n throw new ServiceError(errorLog, dump);\n }\n\n return {\n data,\n thought,\n usage,\n dump,\n };\n }\n\n async describe(\n target: Rect | [number, number],\n modelConfig: IModelConfig,\n opt?: {\n deepThink?: boolean;\n },\n ): Promise<Pick<AIDescribeElementResponse, 'description'>> {\n assert(target, 'target is required for service.describe');\n const context = await this.contextRetrieverFn();\n const { screenshotBase64, size } = context;\n assert(screenshotBase64, 'screenshot is required for service.describe');\n // The result of the \"describe\" function will be used for positioning, so essentially it is a form of grounding.\n const { vlMode } = modelConfig;\n const systemPrompt = elementDescriberInstruction();\n\n // Convert [x,y] center point to Rect if needed\n const defaultRectSize = 30;\n const targetRect: Rect = Array.isArray(target)\n ? {\n left: Math.floor(target[0] - defaultRectSize / 2),\n top: Math.floor(target[1] - defaultRectSize / 2),\n width: defaultRectSize,\n height: defaultRectSize,\n }\n : target;\n\n let imagePayload = await compositeElementInfoImg({\n inputImgBase64: screenshotBase64,\n size,\n elementsPositionInfo: [\n {\n rect: targetRect,\n },\n ],\n borderThickness: 3,\n });\n\n if (opt?.deepThink) {\n const searchArea = expandSearchArea(targetRect, context.size, vlMode);\n debug('describe: set searchArea', searchArea);\n const croppedResult = await cropByRect(\n imagePayload,\n searchArea,\n vlMode === 'qwen-vl',\n );\n imagePayload = croppedResult.imageBase64;\n }\n\n const msgs: AIArgs = [\n { role: 'system', content: systemPrompt },\n {\n role: 'user',\n content: [\n {\n type: 'image_url',\n image_url: {\n url: imagePayload,\n detail: 'high',\n },\n },\n ],\n },\n ];\n\n const callAIFn = this\n .aiVendorFn as typeof callAIWithObjectResponse<AIDescribeElementResponse>;\n\n const res = await callAIFn(\n msgs,\n AIActionType.DESCRIBE_ELEMENT,\n modelConfig,\n );\n\n const { content } = res;\n assert(!content.error, `describe failed: ${content.error}`);\n assert(content.description, 'failed to describe the element');\n return content;\n }\n}\n"],"names":["debug","getDebug","Service","query","opt","modelConfig","_parseResult_errors","queryPrompt","assert","globalDeepThinkSwitch","globalConfigManager","MIDSCENE_FORCE_DEEP_THINK","searchAreaPrompt","vlMode","console","undefined","context","searchArea","searchAreaRawResponse","searchAreaUsage","searchAreaResponse","AiLocateSection","startTime","Date","parseResult","rect","rawResponse","usage","AiLocateElement","timeCost","taskInfo","JSON","errorLog","dumpData","elements","dump","createServiceDump","ServiceError","dataDemand","pageDescription","multimodalPrompt","AiExtractElementInfo","data","thought","target","screenshotBase64","size","systemPrompt","elementDescriberInstruction","defaultRectSize","targetRect","Array","Math","imagePayload","compositeElementInfoImg","expandSearchArea","croppedResult","cropByRect","msgs","callAIFn","res","AIActionType","content","callAIWithObjectResponse","Promise"],"mappings":";;;;;;;;;;;;;;;;;;;;AA8CA,MAAMA,QAAQC,SAAS;AACR,MAAMC;IA4BnB,MAAM,OACJC,KAA0B,EAC1BC,GAAe,EACfC,WAAyB,EACM;YAyE3BC;QAxEJ,MAAMC,cAAc,AAAiB,YAAjB,OAAOJ,QAAqBA,QAAQA,MAAM,MAAM;QACpEK,OAAOD,aAAa;QAEpBC,OAAO,AAAiB,YAAjB,OAAOL,OAAoB;QAElC,MAAMM,wBAAwBC,oBAAoB,qBAAqB,CACrEC;QAEF,IAAIF,uBACFT,MAAM,yBAAyBS;QAEjC,IAAIG;QACJ,IAAIT,MAAM,SAAS,IAAIM,uBACrBG,mBAAmBT,MAAM,MAAM;QAGjC,MAAM,EAAEU,MAAM,EAAE,GAAGR;QAEnB,IAAIO,oBAAoB,CAACC,QAAQ;YAC/BC,QAAQ,IAAI,CACV;YAEFF,mBAAmBG;QACrB;QAEA,MAAMC,UAAUZ,AAAAA,CAAAA,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,OAAO,AAAD,KAAM,MAAM,IAAI,CAAC,kBAAkB;QAE9D,IAAIa;QACJ,IAAIC;QACJ,IAAIC;QACJ,IAAIC;QAGJ,IAAIR,kBAAkB;YACpBQ,qBAAqB,MAAMC,gBAAgB;gBACzCL;gBACA,oBAAoBJ;gBACpBP;YACF;YACAG,OACEY,mBAAmB,IAAI,EACvB,CAAC,6BAA6B,EAAER,iBAAiB,CAAC,EAChDQ,mBAAmB,KAAK,GAAG,CAAC,EAAE,EAAEA,mBAAmB,KAAK,EAAE,GAAG,IAC7D;YAEJF,wBAAwBE,mBAAmB,WAAW;YACtDD,kBAAkBC,mBAAmB,KAAK;YAC1CH,aAAaG,mBAAmB,IAAI;QACtC;QAEA,MAAME,YAAYC,KAAK,GAAG;QAC1B,MAAM,EAAEC,WAAW,EAAEC,IAAI,EAAEC,WAAW,EAAEC,KAAK,EAAE,GAAG,MAAMC,gBAAgB;YACtE,UAAU,IAAI,CAAC,UAAU;YACzBZ;YACA,0BAA0BT;YAC1B,cAAca;YACdf;QACF;QAEA,MAAMwB,WAAWN,KAAK,GAAG,KAAKD;QAC9B,MAAMQ,WAA4B;YAChC,GAAI,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC,QAAQ,GAAG,CAAC,CAAC;YACtC,YAAYD;YACZ,aAAaE,KAAK,SAAS,CAACL;YAC5B,gBAAgBK,KAAK,SAAS,CAACP;YAC/BG;YACAV;YACAC;YACAC;QACF;QAEA,IAAIa;QACJ,IAAI,QAAA1B,CAAAA,sBAAAA,YAAY,MAAM,AAAD,IAAjBA,KAAAA,IAAAA,oBAAoB,MAAM,EAC5B0B,WAAW,CAAC,6BAA6B,EAAER,YAAY,MAAM,CAAC,IAAI,CAAC,OAAO;QAG5E,MAAMS,WAAsC;YAC1C,MAAM;YACN,WAAW;gBACT,SAAS1B;YACX;YACA,gBAAgB,EAAE;YAClB,aAAakB;YACb,MAAM;YACNK;YACA,WAAW,CAAC,CAACb;YACb,OAAOe;QACT;QAEA,MAAME,WAAWV,YAAY,QAAQ,IAAI,EAAE;QAE3C,MAAMW,OAAOC,kBAAkB;YAC7B,GAAGH,QAAQ;YACX,gBAAgBC;QAClB;QAEA,IAAIF,UACF,MAAM,IAAIK,aAAaL,UAAUG;QAGnC,IAAID,SAAS,MAAM,GAAG,GACpB,MAAM,IAAIG,aACR,CAAC,0CAA0C,EAAEH,SAAS,MAAM,EAAE,EAC9DC;QAIJ,IAAID,AAAoB,MAApBA,SAAS,MAAM,EACjB,OAAO;YACL,SAAS;gBACP,QAAQA,QAAQ,CAAC,EAAE,CAAE,MAAM;gBAC3B,MAAMA,QAAQ,CAAC,EAAE,CAAE,IAAI;YACzB;YACAT;YACAU;QACF;QAGF,OAAO;YACL,SAAS;YACTV;YACAU;QACF;IACF;IAEA,MAAM,QACJG,UAA+B,EAC/BjC,WAAyB,EACzBD,GAA0B,EAC1BmC,eAAwB,EACxBC,gBAAoC,EACF;YA0B9BlC;QAzBJE,OACE,AAAsB,YAAtB,OAAO8B,cAA2B,AAAsB,YAAtB,OAAOA,YACzC,CAAC,+CAA+C,EAAE,OAAOA,YAAY;QAEvE,MAAMtB,UAAU,MAAM,IAAI,CAAC,kBAAkB;QAE7C,MAAMM,YAAYC,KAAK,GAAG;QAE1B,MAAM,EAAEC,WAAW,EAAEG,KAAK,EAAE,GAAG,MAAMc,qBAAwB;YAC3DzB;YACA,WAAWsB;YACXE;YACA,eAAepC;YACfC;YACAkC;QACF;QAEA,MAAMV,WAAWN,KAAK,GAAG,KAAKD;QAC9B,MAAMQ,WAA4B;YAChC,GAAI,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC,QAAQ,GAAG,CAAC,CAAC;YACtC,YAAYD;YACZ,aAAaE,KAAK,SAAS,CAACP;QAC9B;QAEA,IAAIQ;QACJ,IAAI,QAAA1B,CAAAA,sBAAAA,YAAY,MAAM,AAAD,IAAjBA,KAAAA,IAAAA,oBAAoB,MAAM,EAC5B0B,WAAW,CAAC,qBAAqB,EAAER,YAAY,MAAM,CAAC,IAAI,CAAC,OAAO;QAGpE,MAAMS,WAAsC;YAC1C,MAAM;YACN,WAAW;gBACTK;YACF;YACA,gBAAgB,EAAE;YAClB,MAAM;YACNR;YACA,OAAOE;QACT;QAEA,MAAM,EAAEU,IAAI,EAAEC,OAAO,EAAE,GAAGnB,eAAe,CAAC;QAG1C,MAAMW,OAAOC,kBAAkB;YAC7B,GAAGH,QAAQ;YACXS;QACF;QAEA,IAAIV,YAAY,CAACU,MACf,MAAM,IAAIL,aAAaL,UAAUG;QAGnC,OAAO;YACLO;YACAC;YACAhB;YACAQ;QACF;IACF;IAEA,MAAM,SACJS,MAA+B,EAC/BvC,WAAyB,EACzBD,GAEC,EACwD;QACzDI,OAAOoC,QAAQ;QACf,MAAM5B,UAAU,MAAM,IAAI,CAAC,kBAAkB;QAC7C,MAAM,EAAE6B,gBAAgB,EAAEC,IAAI,EAAE,GAAG9B;QACnCR,OAAOqC,kBAAkB;QAEzB,MAAM,EAAEhC,MAAM,EAAE,GAAGR;QACnB,MAAM0C,eAAeC;QAGrB,MAAMC,kBAAkB;QACxB,MAAMC,aAAmBC,MAAM,OAAO,CAACP,UACnC;YACE,MAAMQ,KAAK,KAAK,CAACR,MAAM,CAAC,EAAE,GAAGK,kBAAkB;YAC/C,KAAKG,KAAK,KAAK,CAACR,MAAM,CAAC,EAAE,GAAGK,kBAAkB;YAC9C,OAAOA;YACP,QAAQA;QACV,IACAL;QAEJ,IAAIS,eAAe,MAAMC,wBAAwB;YAC/C,gBAAgBT;YAChBC;YACA,sBAAsB;gBACpB;oBACE,MAAMI;gBACR;aACD;YACD,iBAAiB;QACnB;QAEA,IAAI9C,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,SAAS,EAAE;YAClB,MAAMa,aAAasC,iBAAiBL,YAAYlC,QAAQ,IAAI,EAAEH;YAC9Db,MAAM,4BAA4BiB;YAClC,MAAMuC,gBAAgB,MAAMC,WAC1BJ,cACApC,YACAJ,AAAW,cAAXA;YAEFwC,eAAeG,cAAc,WAAW;QAC1C;QAEA,MAAME,OAAe;YACnB;gBAAE,MAAM;gBAAU,SAASX;YAAa;YACxC;gBACE,MAAM;gBACN,SAAS;oBACP;wBACE,MAAM;wBACN,WAAW;4BACT,KAAKM;4BACL,QAAQ;wBACV;oBACF;iBACD;YACH;SACD;QAED,MAAMM,WAAW,IAAI,CAClB,UAAU;QAEb,MAAMC,MAAM,MAAMD,SAChBD,MACAG,aAAa,gBAAgB,EAC7BxD;QAGF,MAAM,EAAEyD,OAAO,EAAE,GAAGF;QACpBpD,OAAO,CAACsD,QAAQ,KAAK,EAAE,CAAC,iBAAiB,EAAEA,QAAQ,KAAK,EAAE;QAC1DtD,OAAOsD,QAAQ,WAAW,EAAE;QAC5B,OAAOA;IACT;IAtSA,YACE9C,OAA2D,EAC3DZ,GAAoB,CACpB;QAVF;QAEA,qCACE2D;QAEF;QAMEvD,OAAOQ,SAAS;QAChB,IAAI,AAAmB,cAAnB,OAAOA,SACT,IAAI,CAAC,kBAAkB,GAAGA;aAE1B,IAAI,CAAC,kBAAkB,GAAG,IAAMgD,QAAQ,OAAO,CAAChD;QAIlD,IAAI,AAA2B,WAApBZ,CAAAA,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,UAAU,AAAD,GACvB,IAAI,CAAC,UAAU,GAAGA,IAAI,UAAU;QAElC,IAAI,AAAyB,WAAlBA,CAAAA,QAAAA,MAAAA,KAAAA,IAAAA,IAAK,QAAQ,AAAD,GACrB,IAAI,CAAC,QAAQ,GAAGA,IAAI,QAAQ;IAEhC;AAqRF"}
|