@midscene/core 0.28.11-beta-20250922065131.0 → 0.28.12-beta-20250923080328.0

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

@@ -15,18 +15,31 @@ const pointToBbox = (point, width, height)=>[
         Math.round(Math.min(point.x + bboxSize / 2, width)),
         Math.round(Math.min(point.y + bboxSize / 2, height))
     ];
-async function vlmPlanning(options) {
-    const { conversationHistory, userInstruction, size, modelConfig } = options;
+async function vlmPlanning(userInstruction, options) {
+    const { conversationHistory, context, modelConfig } = options;
     const { uiTarsModelVersion } = modelConfig;
     const systemPrompt = getUiTarsPlanningPrompt() + userInstruction;
+    const imagePayload = await resizeImageForUiTars(context.screenshotBase64, context.size, uiTarsModelVersion);
+    conversationHistory.append({
+        role: 'user',
+        content: [
+            {
+                type: 'image_url',
+                image_url: {
+                    url: imagePayload
+                }
+            }
+        ]
+    });
     const res = await callAIWithStringResponse([
         {
             role: 'user',
             content: systemPrompt
         },
-        ...conversationHistory
+        ...conversationHistory.snapshot()
     ], AIActionType.INSPECT_ELEMENT, modelConfig);
     const convertedText = convertBboxToCoordinates(res.content);
+    const { size } = context;
     const { parsed } = actionParser({
         prediction: convertedText,
         factor: [
@@ -41,6 +54,7 @@ async function vlmPlanning(options) {
     });
     debug('ui-tars modelVer', uiTarsModelVersion, ', parsed', JSON.stringify(parsed));
     const transformActions = [];
+    let shouldContinue = true;
     parsed.forEach((action)=>{
         if ('click' === action.action_type) {
             assert(action.action_inputs.start_box, 'start_box is required');
@@ -118,6 +132,7 @@ async function vlmPlanning(options) {
             },
             thought: action.thought || ''
         });
+        else if ('finished' === action.action_type) shouldContinue = false;
     });
     if (0 === transformActions.length) throw new Error(`No actions found, response: ${res.content}`, {
         cause: {
@@ -126,12 +141,17 @@ async function vlmPlanning(options) {
         }
     });
     debug('transformActions', JSON.stringify(transformActions, null, 2));
+    const log = getSummary(res.content);
+    conversationHistory.append({
+        role: 'assistant',
+        content: log
+    });
     return {
         actions: transformActions,
-        actionsFromModel: parsed,
-        action_summary: getSummary(res.content),
+        log,
         usage: res.usage,
-        rawResponse: JSON.stringify(res.content, void 0, 2)
+        rawResponse: JSON.stringify(res.content, void 0, 2),
+        more_actions_needed_by_instruction: shouldContinue
     };
 }
 function convertBboxToCoordinates(text) {

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 { callAIWithStringResponse } from './service-caller/index';\ntype ActionType =\n  | 'click'\n  | 'drag'\n  | 'type'\n  | 'hotkey'\n  | 'finished'\n  | 'scroll'\n  | 'wait';\n\nconst debug = getDebug('ui-tars-planning');\nconst bboxSize = 10;\nconst pointToBbox = (\n  point: { x: number; y: number },\n  width: number,\n  height: number,\n): [number, number, number, number] => {\n  return [\n    Math.round(Math.max(point.x - bboxSize / 2, 0)),\n    Math.round(Math.max(point.y - bboxSize / 2, 0)),\n    Math.round(Math.min(point.x + bboxSize / 2, width)),\n    Math.round(Math.min(point.y + bboxSize / 2, height)),\n  ];\n};\n\nexport async function 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 callAIWithStringResponse(\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","callAIWithStringResponse","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,yBAChB;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  PlanningAIResponse,\n  PlanningAction,\n  Size,\n  UIContext,\n} from '@/types';\nimport { type IModelConfig, UITarsModelVersion } from '@midscene/shared/env';\nimport { resizeImgBase64 } from '@midscene/shared/img';\nimport { getDebug } from '@midscene/shared/logger';\nimport { transformHotkeyInput } from '@midscene/shared/us-keyboard-layout';\nimport { assert } from '@midscene/shared/utils';\nimport { actionParser } from '@ui-tars/action-parser';\nimport { AIActionType } from './common';\nimport type { ConversationHistory } from './conversation-history';\nimport { getSummary, getUiTarsPlanningPrompt } from './prompt/ui-tars-planning';\nimport { callAIWithStringResponse } from './service-caller/index';\ntype ActionType =\n  | 'click'\n  | 'drag'\n  | 'type'\n  | 'hotkey'\n  | 'finished'\n  | 'scroll'\n  | 'wait';\n\nconst debug = getDebug('ui-tars-planning');\nconst bboxSize = 10;\nconst pointToBbox = (\n  point: { x: number; y: number },\n  width: number,\n  height: number,\n): [number, number, number, number] => {\n  return [\n    Math.round(Math.max(point.x - bboxSize / 2, 0)),\n    Math.round(Math.max(point.y - bboxSize / 2, 0)),\n    Math.round(Math.min(point.x + bboxSize / 2, width)),\n    Math.round(Math.min(point.y + bboxSize / 2, height)),\n  ];\n};\n\nexport async function vlmPlanning(\n  userInstruction: string,\n  options: {\n    conversationHistory: ConversationHistory;\n    context: UIContext;\n    modelConfig: IModelConfig;\n  },\n): Promise<PlanningAIResponse> {\n  const { conversationHistory, context, modelConfig } = options;\n  const { uiTarsModelVersion } = modelConfig;\n  const systemPrompt = getUiTarsPlanningPrompt() + userInstruction;\n\n  const imagePayload = await resizeImageForUiTars(\n    context.screenshotBase64,\n    context.size,\n    uiTarsModelVersion,\n  );\n\n  conversationHistory.append({\n    role: 'user',\n    content: [\n      {\n        type: 'image_url',\n        image_url: {\n          url: imagePayload,\n        },\n      },\n    ],\n  });\n\n  const res = await callAIWithStringResponse(\n    [\n      {\n        role: 'user',\n        content: systemPrompt,\n      },\n      ...conversationHistory.snapshot(),\n    ],\n    AIActionType.INSPECT_ELEMENT,\n    modelConfig,\n  );\n  const convertedText = convertBboxToCoordinates(res.content);\n\n  const { size } = context;\n  const { parsed } = actionParser({\n    prediction: convertedText,\n    factor: [1000, 1000],\n    screenContext: {\n      width: size.width,\n      height: size.height,\n    },\n    modelVer: uiTarsModelVersion,\n  });\n\n  debug(\n    'ui-tars modelVer',\n    uiTarsModelVersion,\n    ', parsed',\n    JSON.stringify(parsed),\n  );\n\n  const transformActions: PlanningAction[] = [];\n  let shouldContinue = true;\n  parsed.forEach((action) => {\n    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 === 'finished') {\n      shouldContinue = false;\n    }\n  });\n\n  if (transformActions.length === 0) {\n    throw new Error(`No actions found, response: ${res.content}`, {\n      cause: {\n        prediction: res.content,\n        parsed,\n      },\n    });\n  }\n\n  debug('transformActions', JSON.stringify(transformActions, null, 2));\n  const log = getSummary(res.content);\n\n  conversationHistory.append({\n    role: 'assistant',\n    content: log,\n  });\n\n  return {\n    actions: transformActions,\n    log,\n    usage: res.usage,\n    rawResponse: JSON.stringify(res.content, undefined, 2),\n    more_actions_needed_by_instruction: shouldContinue,\n  };\n}\n\n/**\n * Converts bounding box notation to coordinate points\n * @param text - The text containing bbox tags to be converted\n * @returns The text with bbox tags replaced by coordinate points\n */\nfunction convertBboxToCoordinates(text: string): string {\n  // Match the four numbers after <bbox>\n  const pattern = /<bbox>(\\d+)\\s+(\\d+)\\s+(\\d+)\\s+(\\d+)<\\/bbox>/g;\n\n  function replaceMatch(\n    match: string,\n    x1: string,\n    y1: string,\n    x2: string,\n    y2: string,\n  ): string {\n    // Convert strings to numbers and calculate center point\n    const x1Num = Number.parseInt(x1, 10);\n    const y1Num = Number.parseInt(y1, 10);\n    const x2Num = Number.parseInt(x2, 10);\n    const y2Num = Number.parseInt(y2, 10);\n\n    // Use Math.floor to truncate and calculate center point\n    const x = Math.floor((x1Num + x2Num) / 2);\n    const y = Math.floor((y1Num + y2Num) / 2);\n\n    // Return formatted coordinate string\n    return `(${x},${y})`;\n  }\n\n  // Remove [EOS] and replace <bbox> coordinates\n  const cleanedText = text.replace(/\\[EOS\\]/g, '');\n  return cleanedText.replace(pattern, replaceMatch).trim();\n}\n\nfunction getPoint(startBox: string, size: { width: number; height: number }) {\n  const [x, y] = JSON.parse(startBox);\n  return [x * size.width, y * size.height];\n}\n\ninterface BaseAction {\n  action_type: ActionType;\n  action_inputs: Record<string, any>;\n  reflection: string | null;\n  thought: string | null;\n}\n\ninterface ClickAction extends BaseAction {\n  action_type: 'click';\n  action_inputs: {\n    start_box: string; // JSON string of [x, y] coordinates\n  };\n}\n\ninterface DragAction extends BaseAction {\n  action_type: 'drag';\n  action_inputs: {\n    start_box: string; // JSON string of [x, y] coordinates\n    end_box: string; // JSON string of [x, y] coordinates\n  };\n}\n\ninterface WaitAction extends BaseAction {\n  action_type: 'wait';\n  action_inputs: {\n    time: string; // JSON string of [x, y] coordinates\n  };\n}\n\ninterface TypeAction extends BaseAction {\n  action_type: 'type';\n  action_inputs: {\n    content: string;\n  };\n}\n\ninterface HotkeyAction extends BaseAction {\n  action_type: 'hotkey';\n  action_inputs: {\n    key: string;\n  };\n}\n\ninterface ScrollAction extends BaseAction {\n  action_type: 'scroll';\n  action_inputs: {\n    direction: 'up' | 'down';\n  };\n}\n\ninterface FinishedAction extends BaseAction {\n  action_type: 'finished';\n  action_inputs: Record<string, never>;\n}\n\nexport type Action =\n  | ClickAction\n  | DragAction\n  | TypeAction\n  | HotkeyAction\n  | ScrollAction\n  | FinishedAction\n  | WaitAction;\n\nexport async function resizeImageForUiTars(\n  imageBase64: string,\n  size: Size,\n  uiTarsVersion: UITarsModelVersion | undefined,\n) {\n  if (uiTarsVersion === UITarsModelVersion.V1_5) {\n    debug('ui-tars-v1.5, will check image size', size);\n    const currentPixels = size.width * size.height;\n    const maxPixels = 16384 * 28 * 28; //\n    if (currentPixels > maxPixels) {\n      const resizeFactor = Math.sqrt(maxPixels / currentPixels);\n      const newWidth = Math.floor(size.width * resizeFactor);\n      const newHeight = Math.floor(size.height * resizeFactor);\n      debug(\n        'resize image for ui-tars, new width: %s, new height: %s',\n        newWidth,\n        newHeight,\n      );\n      const resizedImage = await resizeImgBase64(imageBase64, {\n        width: newWidth,\n        height: newHeight,\n      });\n      return resizedImage;\n    }\n  }\n  return imageBase64;\n}\n"],"names":["debug","getDebug","bboxSize","pointToBbox","point","width","height","Math","vlmPlanning","userInstruction","options","conversationHistory","context","modelConfig","uiTarsModelVersion","systemPrompt","getUiTarsPlanningPrompt","imagePayload","resizeImageForUiTars","res","callAIWithStringResponse","AIActionType","convertedText","convertBboxToCoordinates","size","parsed","actionParser","JSON","transformActions","shouldContinue","action","assert","getPoint","startPoint","endPoint","keys","transformHotkeyInput","console","Error","log","getSummary","undefined","text","pattern","replaceMatch","match","x1","y1","x2","y2","x1Num","Number","y1Num","x2Num","y2Num","x","y","cleanedText","startBox","imageBase64","uiTarsVersion","UITarsModelVersion","currentPixels","maxPixels","resizeFactor","newWidth","newHeight","resizedImage","resizeImgBase64"],"mappings":";;;;;;;;;AAyBA,MAAMA,QAAQC,SAAS;AACvB,MAAMC,WAAW;AACjB,MAAMC,cAAc,CAClBC,OACAC,OACAC,SAEO;QACLC,KAAK,KAAK,CAACA,KAAK,GAAG,CAACH,MAAM,CAAC,GAAGF,WAAW,GAAG;QAC5CK,KAAK,KAAK,CAACA,KAAK,GAAG,CAACH,MAAM,CAAC,GAAGF,WAAW,GAAG;QAC5CK,KAAK,KAAK,CAACA,KAAK,GAAG,CAACH,MAAM,CAAC,GAAGF,WAAW,GAAGG;QAC5CE,KAAK,KAAK,CAACA,KAAK,GAAG,CAACH,MAAM,CAAC,GAAGF,WAAW,GAAGI;KAC7C;AAGI,eAAeE,YACpBC,eAAuB,EACvBC,OAIC;IAED,MAAM,EAAEC,mBAAmB,EAAEC,OAAO,EAAEC,WAAW,EAAE,GAAGH;IACtD,MAAM,EAAEI,kBAAkB,EAAE,GAAGD;IAC/B,MAAME,eAAeC,4BAA4BP;IAEjD,MAAMQ,eAAe,MAAMC,qBACzBN,QAAQ,gBAAgB,EACxBA,QAAQ,IAAI,EACZE;IAGFH,oBAAoB,MAAM,CAAC;QACzB,MAAM;QACN,SAAS;YACP;gBACE,MAAM;gBACN,WAAW;oBACT,KAAKM;gBACP;YACF;SACD;IACH;IAEA,MAAME,MAAM,MAAMC,yBAChB;QACE;YACE,MAAM;YACN,SAASL;QACX;WACGJ,oBAAoB,QAAQ;KAChC,EACDU,aAAa,eAAe,EAC5BR;IAEF,MAAMS,gBAAgBC,yBAAyBJ,IAAI,OAAO;IAE1D,MAAM,EAAEK,IAAI,EAAE,GAAGZ;IACjB,MAAM,EAAEa,MAAM,EAAE,GAAGC,aAAa;QAC9B,YAAYJ;QACZ,QAAQ;YAAC;YAAM;SAAK;QACpB,eAAe;YACb,OAAOE,KAAK,KAAK;YACjB,QAAQA,KAAK,MAAM;QACrB;QACA,UAAUV;IACZ;IAEAd,MACE,oBACAc,oBACA,YACAa,KAAK,SAAS,CAACF;IAGjB,MAAMG,mBAAqC,EAAE;IAC7C,IAAIC,iBAAiB;IACrBJ,OAAO,OAAO,CAAC,CAACK;QACd,IAAIA,AAAuB,YAAvBA,OAAO,WAAW,EAAc;YAClCC,OAAOD,OAAO,aAAa,CAAC,SAAS,EAAE;YACvC,MAAM1B,QAAQ4B,SAASF,OAAO,aAAa,CAAC,SAAS,EAAEN;YACvDI,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO;oBACL,QAAQ;wBACN,QAAQE,OAAO,OAAO,IAAI;wBAC1B,MAAM3B,YACJ;4BAAE,GAAGC,KAAK,CAAC,EAAE;4BAAE,GAAGA,KAAK,CAAC,EAAE;wBAAC,GAC3BoB,KAAK,KAAK,EACVA,KAAK,MAAM;oBAEf;gBACF;YACF;QACF,OAAO,IAAIM,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,EAAEN;YAC5D,MAAMU,WAAWF,SAASF,OAAO,aAAa,CAAC,OAAO,EAAEN;YACxDI,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO;oBACL,MAAM;wBACJ,QAAQE,OAAO,OAAO,IAAI;wBAC1B,MAAM3B,YACJ;4BAAE,GAAG8B,UAAU,CAAC,EAAE;4BAAE,GAAGA,UAAU,CAAC,EAAE;wBAAC,GACrCT,KAAK,KAAK,EACVA,KAAK,MAAM;oBAEf;oBACA,IAAI;wBACF,QAAQM,OAAO,OAAO,IAAI;wBAC1B,MAAM3B,YACJ;4BAAE,GAAG+B,QAAQ,CAAC,EAAE;4BAAE,GAAGA,QAAQ,CAAC,EAAE;wBAAC,GACjCV,KAAK,KAAK,EACVA,KAAK,MAAM;oBAEf;gBACF;gBACA,SAASM,OAAO,OAAO,IAAI;YAC7B;QACF,OAAO,IAAIA,AAAuB,WAAvBA,OAAO,WAAW,EAC3BF,iBAAiB,IAAI,CAAC;YACpB,MAAM;YACN,OAAO;gBACL,OAAOE,OAAO,aAAa,CAAC,OAAO;YACrC;YACA,SAASA,OAAO,OAAO,IAAI;QAC7B;aACK,IAAIA,AAAuB,aAAvBA,OAAO,WAAW,EAC3BF,iBAAiB,IAAI,CAAC;YACpB,MAAM;YACN,OAAO;gBACL,WAAWE,OAAO,aAAa,CAAC,SAAS;YAC3C;YACA,SAASA,OAAO,OAAO,IAAI;QAC7B;aACK,IAAIA,AAAuB,eAAvBA,OAAO,WAAW,EAC3BF,iBAAiB,IAAI,CAAC;YACpB,MAAM;YACN,OAAO,CAAC;YACR,SAASE,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;YAE1DF,iBAAiB,IAAI,CAAC;gBACpB,MAAM;gBACN,OAAO;oBACL,SAASO;gBACX;gBACA,SAASL,OAAO,OAAO,IAAI;YAC7B;QACF,OAbEO,QAAQ,IAAI,CACV;aAaC,IAAIP,AAAuB,WAAvBA,OAAO,WAAW,EAC3BF,iBAAiB,IAAI,CAAC;YACpB,MAAM;YACN,OAAO;gBACL,QAAQ;YACV;YACA,SAASE,OAAO,OAAO,IAAI;QAC7B;aACK,IAAIA,AAAuB,eAAvBA,OAAO,WAAW,EAC3BD,iBAAiB;IAErB;IAEA,IAAID,AAA4B,MAA5BA,iBAAiB,MAAM,EACzB,MAAM,IAAIU,MAAM,CAAC,4BAA4B,EAAEnB,IAAI,OAAO,EAAE,EAAE;QAC5D,OAAO;YACL,YAAYA,IAAI,OAAO;YACvBM;QACF;IACF;IAGFzB,MAAM,oBAAoB2B,KAAK,SAAS,CAACC,kBAAkB,MAAM;IACjE,MAAMW,MAAMC,WAAWrB,IAAI,OAAO;IAElCR,oBAAoB,MAAM,CAAC;QACzB,MAAM;QACN,SAAS4B;IACX;IAEA,OAAO;QACL,SAASX;QACTW;QACA,OAAOpB,IAAI,KAAK;QAChB,aAAaQ,KAAK,SAAS,CAACR,IAAI,OAAO,EAAEsB,QAAW;QACpD,oCAAoCZ;IACtC;AACF;AAOA,SAASN,yBAAyBmB,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,IAAIhD,KAAK,KAAK,CAAE2C,AAAAA,CAAAA,QAAQG,KAAI,IAAK;QACvC,MAAMG,IAAIjD,KAAK,KAAK,CAAE6C,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,EAAElC,IAAuC;IACzE,MAAM,CAAC+B,GAAGC,EAAE,GAAG7B,KAAK,KAAK,CAAC+B;IAC1B,OAAO;QAACH,IAAI/B,KAAK,KAAK;QAAEgC,IAAIhC,KAAK,MAAM;KAAC;AAC1C;AAkEO,eAAeN,qBACpByC,WAAmB,EACnBnC,IAAU,EACVoC,aAA6C;IAE7C,IAAIA,kBAAkBC,mBAAmB,IAAI,EAAE;QAC7C7D,MAAM,uCAAuCwB;QAC7C,MAAMsC,gBAAgBtC,KAAK,KAAK,GAAGA,KAAK,MAAM;QAC9C,MAAMuC,YAAY;QAClB,IAAID,gBAAgBC,WAAW;YAC7B,MAAMC,eAAezD,KAAK,IAAI,CAACwD,YAAYD;YAC3C,MAAMG,WAAW1D,KAAK,KAAK,CAACiB,KAAK,KAAK,GAAGwC;YACzC,MAAME,YAAY3D,KAAK,KAAK,CAACiB,KAAK,MAAM,GAAGwC;YAC3ChE,MACE,2DACAiE,UACAC;YAEF,MAAMC,eAAe,MAAMC,gBAAgBT,aAAa;gBACtD,OAAOM;gBACP,QAAQC;YACV;YACA,OAAOC;QACT;IACF;IACA,OAAOR;AACT"}