@midscene/core 0.26.7-beta-20250821124744.0 → 0.26.7-beta-20250821132536.0

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

@@ -118,49 +118,6 @@ async function vlmPlanning(options) {
             },
             thought: action.thought || ''
         });
-        else if ('androidBackButton' === action.action_type) transformActions.push({
-            type: 'AndroidBackButton',
-            param: {},
-            thought: action.thought || ''
-        });
-        else if ('androidHomeButton' === action.action_type) transformActions.push({
-            type: 'AndroidHomeButton',
-            param: {},
-            thought: action.thought || ''
-        });
-        else if ('androidRecentAppsButton' === action.action_type) transformActions.push({
-            type: 'AndroidRecentAppsButton',
-            param: {}
-        });
-        else if ('androidLongPress' === action.action_type) {
-            assert(action.action_inputs.start_box, 'start_box is required for androidLongPress');
-            const start = getPoint(action.action_inputs.start_box, size);
-            transformActions.push({
-                type: 'AndroidLongPress',
-                param: {
-                    locate: {
-                        prompt: action.thought || '',
-                        bbox: pointToBbox({
-                            x: start[0],
-                            y: start[1]
-                        }, size.width, size.height)
-                    },
-                    duration: 1000
-                },
-                thought: action.thought || ''
-            });
-        } else if ('androidPull' === action.action_type) {
-            const pullDirection = action.action_inputs.direction || 'down';
-            transformActions.push({
-                type: 'AndroidPull',
-                param: {
-                    direction: pullDirection,
-                    distance: action.action_inputs.distance,
-                    duration: action.action_inputs.duration || 500
-                },
-                thought: action.thought || ''
-            });
-        }
     });
     if (0 === transformActions.length) throw new Error(`No actions found, response: ${res.content}`, {
         cause: {

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 {\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  | 'androidBackButton'\n  | 'androidHomeButton'\n  | 'androidRecentAppsButton'\n  | 'androidLongPress'\n  | 'androidPull';\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    } else if (action.action_type === 'androidBackButton') {\n      transformActions.push({\n        type: 'AndroidBackButton',\n        param: {},\n        thought: action.thought || '',\n      });\n    } else if (action.action_type === 'androidHomeButton') {\n      transformActions.push({\n        type: 'AndroidHomeButton',\n        param: {},\n        thought: action.thought || '',\n      });\n    } else if (action.action_type === 'androidRecentAppsButton') {\n      transformActions.push({\n        type: 'AndroidRecentAppsButton',\n        param: {},\n      });\n    } else if (action.action_type === 'androidLongPress') {\n      assert(\n        action.action_inputs.start_box,\n        'start_box is required for androidLongPress',\n      );\n      const start = getPoint(action.action_inputs.start_box, size);\n      transformActions.push({\n        type: 'AndroidLongPress',\n        param: {\n          locate: {\n            prompt: action.thought || '',\n            bbox: pointToBbox(\n              { x: start[0], y: start[1] },\n              size.width,\n              size.height,\n            ),\n          },\n          duration: 1000,\n        },\n        thought: action.thought || '',\n      });\n    } else if (action.action_type === 'androidPull') {\n      const pullDirection = action.action_inputs.direction || 'down';\n      transformActions.push({\n        type: 'AndroidPull',\n        param: {\n          direction: pullDirection as 'up' | 'down',\n          distance: (action.action_inputs as any).distance,\n          duration: (action.action_inputs as any).duration || 500,\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\ninterface AndroidLongPressAction extends BaseAction {\n  action_type: 'androidLongPress';\n  action_inputs: {\n    start_coords: [number, number]; // Coordinates for long press\n    duration?: number; // Duration in milliseconds\n  };\n}\n\nexport type Action =\n  | ClickAction\n  | DragAction\n  | TypeAction\n  | HotkeyAction\n  | ScrollAction\n  | FinishedAction\n  | WaitAction\n  | AndroidLongPressAction;\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","start","pullDirection","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":";;;;;;;;;AAmCA,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;aACK,IAAIA,AAAuB,wBAAvBA,OAAO,WAAW,EAC3BD,iBAAiB,IAAI,CAAC;YACpB,MAAM;YACN,OAAO,CAAC;YACR,SAASC,OAAO,OAAO,IAAI;QAC7B;aACK,IAAIA,AAAuB,wBAAvBA,OAAO,WAAW,EAC3BD,iBAAiB,IAAI,CAAC;YACpB,MAAM;YACN,OAAO,CAAC;YACR,SAASC,OAAO,OAAO,IAAI;QAC7B;aACK,IAAIA,AAAuB,8BAAvBA,OAAO,WAAW,EAC3BD,iBAAiB,IAAI,CAAC;YACpB,MAAM;YACN,OAAO,CAAC;QACV;aACK,IAAIC,AAAuB,uBAAvBA,OAAO,WAAW,EAAyB;YACpDC,OACED,OAAO,aAAa,CAAC,SAAS,EAC9B;YAEF,MAAMQ,QAAQN,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,GAAGiC,KAAK,CAAC,EAAE;4BAAE,GAAGA,KAAK,CAAC,EAAE;wBAAC,GAC3BxB,KAAK,KAAK,EACVA,KAAK,MAAM;oBAEf;oBACA,UAAU;gBACZ;gBACA,SAASgB,OAAO,OAAO,IAAI;YAC7B;QACF,OAAO,IAAIA,AAAuB,kBAAvBA,OAAO,WAAW,EAAoB;YAC/C,MAAMS,gBAAgBT,OAAO,aAAa,CAAC,SAAS,IAAI;YACxDD,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO;oBACL,WAAWU;oBACX,UAAWT,OAAO,aAAa,CAAS,QAAQ;oBAChD,UAAWA,OAAO,aAAa,CAAS,QAAQ,IAAI;gBACtD;gBACA,SAASA,OAAO,OAAO,IAAI;YAC7B;QACF;IACF;IAEA,IAAID,AAA4B,MAA5BA,iBAAiB,MAAM,EACzB,MAAM,IAAIW,MAAM,CAAC,4BAA4B,EAAEtB,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,gBAAgBgB,WAAWvB,IAAI,OAAO;QACtC,OAAOA,IAAI,KAAK;QAChB,aAAaU,KAAK,SAAS,CAACV,IAAI,OAAO,EAAES,QAAW;IACtD;AACF;AAOA,SAASL,yBAAyBoB,IAAY;IAE5C,MAAMC,UAAU;IAEhB,SAASC,aACPC,KAAa,EACbC,EAAU,EACVC,EAAU,EACVC,EAAU,EACVC,EAAU;QAGV,MAAMC,QAAQC,OAAO,QAAQ,CAACL,IAAI;QAClC,MAAMM,QAAQD,OAAO,QAAQ,CAACJ,IAAI;QAClC,MAAMM,QAAQF,OAAO,QAAQ,CAACH,IAAI;QAClC,MAAMM,QAAQH,OAAO,QAAQ,CAACF,IAAI;QAGlC,MAAMM,IAAI9C,KAAK,KAAK,CAAEyC,AAAAA,CAAAA,QAAQG,KAAI,IAAK;QACvC,MAAMG,IAAI/C,KAAK,KAAK,CAAE2C,AAAAA,CAAAA,QAAQE,KAAI,IAAK;QAGvC,OAAO,CAAC,CAAC,EAAEC,EAAE,CAAC,EAAEC,EAAE,CAAC,CAAC;IACtB;IAGA,MAAMC,cAAcf,KAAK,OAAO,CAAC,YAAY;IAC7C,OAAOe,YAAY,OAAO,CAACd,SAASC,cAAc,IAAI;AACxD;AAEA,SAASZ,SAAS0B,QAAgB,EAAE5C,IAAuC;IACzE,MAAM,CAACyC,GAAGC,EAAE,GAAG5B,KAAK,KAAK,CAAC8B;IAC1B,OAAO;QAACH,IAAIzC,KAAK,KAAK;QAAE0C,IAAI1C,KAAK,MAAM;KAAC;AAC1C;AA2EO,eAAe6C,qBACpBC,WAAmB,EACnB9C,IAAU,EACVC,gBAAmC;IAEnC,IACE8C,AAAmC,kBAAnCA,aAAa9C,qBACbS,mBAAmBT,sBAAsB+C,mBAAmB,IAAI,EAChE;QACA5D,MAAM,uCAAuCY;QAC7C,MAAMiD,gBAAgBjD,KAAK,KAAK,GAAGA,KAAK,MAAM;QAC9C,MAAMkD,YAAY;QAClB,IAAID,gBAAgBC,WAAW;YAC7B,MAAMC,eAAexD,KAAK,IAAI,CAACuD,YAAYD;YAC3C,MAAMG,WAAWzD,KAAK,KAAK,CAACK,KAAK,KAAK,GAAGmD;YACzC,MAAME,YAAY1D,KAAK,KAAK,CAACK,KAAK,MAAM,GAAGmD;YAC3C/D,MACE,2DACAgE,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"}