@midscene/core 0.26.7-beta-20250818035341.0 → 0.26.7-beta-20250820105545.0

package/dist/lib/ai-model/service-caller/index.js

@@ -50,14 +50,13 @@ var __webpack_exports__ = {};
 (()=>{
     __webpack_require__.r(__webpack_exports__);
     __webpack_require__.d(__webpack_exports__, {
-        preprocessDoubaoBboxJson: ()=>preprocessDoubaoBboxJson,
-        checkAIConfig: ()=>checkAIConfig,
-        callAiFnWithStringResponse: ()=>callAiFnWithStringResponse,
-        safeParseJson: ()=>safeParseJson,
         extractJSONFromCodeBlock: ()=>extractJSONFromCodeBlock,
+        preprocessDoubaoBboxJson: ()=>preprocessDoubaoBboxJson,
         callToGetJSONObject: ()=>callToGetJSONObject,
+        callAiFnWithStringResponse: ()=>callAiFnWithStringResponse,
+        getResponseFormat: ()=>getResponseFormat,
         call: ()=>call,
-        getModelName: ()=>getModelName
+        safeParseJson: ()=>safeParseJson
     });
     const external_types_js_namespaceObject = require("../../types.js");
     const sdk_namespaceObject = require("@anthropic-ai/sdk");
@@ -75,46 +74,9 @@ var __webpack_exports__ = {};
     const assertion_js_namespaceObject = require("../prompt/assertion.js");
     const llm_locator_js_namespaceObject = require("../prompt/llm-locator.js");
     const llm_planning_js_namespaceObject = require("../prompt/llm-planning.js");
-    function checkAIConfig() {
-        const openaiKey = (0, env_namespaceObject.getAIConfig)(env_namespaceObject.OPENAI_API_KEY);
-        const azureConfig = (0, env_namespaceObject.getAIConfig)(env_namespaceObject.MIDSCENE_USE_AZURE_OPENAI);
-        const anthropicKey = (0, env_namespaceObject.getAIConfig)(env_namespaceObject.ANTHROPIC_API_KEY);
-        const initConfigJson = (0, env_namespaceObject.getAIConfig)(env_namespaceObject.MIDSCENE_OPENAI_INIT_CONFIG_JSON);
-        if (openaiKey) return true;
-        if (azureConfig) return true;
-        if (anthropicKey) return true;
-        return Boolean(initConfigJson);
-    }
-    let debugConfigInitialized = false;
-    function initDebugConfig() {
-        if (debugConfigInitialized) return;
-        const shouldPrintTiming = (0, env_namespaceObject.getAIConfigInBoolean)(env_namespaceObject.MIDSCENE_DEBUG_AI_PROFILE);
-        let debugConfig = '';
-        if (shouldPrintTiming) {
-            console.warn('MIDSCENE_DEBUG_AI_PROFILE is deprecated, use DEBUG=midscene:ai:profile instead');
-            debugConfig = 'ai:profile';
-        }
-        const shouldPrintAIResponse = (0, env_namespaceObject.getAIConfigInBoolean)(env_namespaceObject.MIDSCENE_DEBUG_AI_RESPONSE);
-        if (shouldPrintAIResponse) {
-            console.warn('MIDSCENE_DEBUG_AI_RESPONSE is deprecated, use DEBUG=midscene:ai:response instead');
-            debugConfig = debugConfig ? 'ai:*' : 'ai:call';
-        }
-        if (debugConfig) (0, logger_namespaceObject.enableDebug)(debugConfig);
-        debugConfigInitialized = true;
-    }
-    const defaultModel = 'gpt-4o';
-    function getModelName() {
-        let modelName = defaultModel;
-        const nameInConfig = (0, env_namespaceObject.getAIConfig)(env_namespaceObject.MIDSCENE_MODEL_NAME);
-        if (nameInConfig) modelName = nameInConfig;
-        return modelName;
-    }
-    async function createChatClient({ AIActionTypeValue }) {
-        initDebugConfig();
+    async function createChatClient({ AIActionTypeValue, modelPreferences }) {
+        const { socksProxy, httpProxy, modelName, openaiBaseURL, openaiApiKey, openaiExtraConfig, openaiUseAzureDeprecated, useAzureOpenai, azureOpenaiScope, azureOpenaiKey, azureOpenaiEndpoint, azureOpenaiApiVersion, azureOpenaiDeployment, azureExtraConfig, useAnthropicSdk, anthropicApiKey } = (0, env_namespaceObject.decideModelConfig)(modelPreferences, true);
         let openai;
-        const extraConfig = (0, env_namespaceObject.getAIConfigInJson)(env_namespaceObject.MIDSCENE_OPENAI_INIT_CONFIG_JSON);
-        const socksProxy = (0, env_namespaceObject.getAIConfig)(env_namespaceObject.MIDSCENE_OPENAI_SOCKS_PROXY);
-        const httpProxy = (0, env_namespaceObject.getAIConfig)(env_namespaceObject.MIDSCENE_OPENAI_HTTP_PROXY);
         let proxyAgent;
         const debugProxy = (0, logger_namespaceObject.getDebug)('ai:call:proxy');
         if (httpProxy) {
@@ -124,56 +86,47 @@ var __webpack_exports__ = {};
             debugProxy('using socks proxy', socksProxy);
             proxyAgent = new external_socks_proxy_agent_namespaceObject.SocksProxyAgent(socksProxy);
         }
-        if ((0, env_namespaceObject.getAIConfig)(env_namespaceObject.OPENAI_USE_AZURE)) openai = new external_openai_namespaceObject.AzureOpenAI({
-            baseURL: (0, env_namespaceObject.getAIConfig)(env_namespaceObject.OPENAI_BASE_URL),
-            apiKey: (0, env_namespaceObject.getAIConfig)(env_namespaceObject.OPENAI_API_KEY),
+        if (openaiUseAzureDeprecated) openai = new external_openai_namespaceObject.AzureOpenAI({
+            baseURL: openaiBaseURL,
+            apiKey: openaiApiKey,
             httpAgent: proxyAgent,
-            ...extraConfig,
+            ...openaiExtraConfig,
             dangerouslyAllowBrowser: true
         });
-        else if ((0, env_namespaceObject.getAIConfig)(env_namespaceObject.MIDSCENE_USE_AZURE_OPENAI)) {
-            const extraAzureConfig = (0, env_namespaceObject.getAIConfigInJson)(env_namespaceObject.MIDSCENE_AZURE_OPENAI_INIT_CONFIG_JSON);
-            const scope = (0, env_namespaceObject.getAIConfig)(env_namespaceObject.MIDSCENE_AZURE_OPENAI_SCOPE);
+        else if (useAzureOpenai) {
             let tokenProvider;
-            if (scope) {
+            if (azureOpenaiScope) {
                 (0, utils_namespaceObject.assert)(!utils_namespaceObject.ifInBrowser, 'Azure OpenAI is not supported in browser with Midscene.');
                 const credential = new identity_namespaceObject.DefaultAzureCredential();
-                (0, utils_namespaceObject.assert)(scope, 'MIDSCENE_AZURE_OPENAI_SCOPE is required');
-                tokenProvider = (0, identity_namespaceObject.getBearerTokenProvider)(credential, scope);
+                tokenProvider = (0, identity_namespaceObject.getBearerTokenProvider)(credential, azureOpenaiScope);
                 openai = new external_openai_namespaceObject.AzureOpenAI({
                     azureADTokenProvider: tokenProvider,
-                    endpoint: (0, env_namespaceObject.getAIConfig)(env_namespaceObject.AZURE_OPENAI_ENDPOINT),
-                    apiVersion: (0, env_namespaceObject.getAIConfig)(env_namespaceObject.AZURE_OPENAI_API_VERSION),
-                    deployment: (0, env_namespaceObject.getAIConfig)(env_namespaceObject.AZURE_OPENAI_DEPLOYMENT),
-                    ...extraConfig,
-                    ...extraAzureConfig
+                    endpoint: azureOpenaiEndpoint,
+                    apiVersion: azureOpenaiApiVersion,
+                    deployment: azureOpenaiDeployment,
+                    ...openaiExtraConfig,
+                    ...azureExtraConfig
                 });
             } else openai = new external_openai_namespaceObject.AzureOpenAI({
-                apiKey: (0, env_namespaceObject.getAIConfig)(env_namespaceObject.AZURE_OPENAI_KEY),
-                endpoint: (0, env_namespaceObject.getAIConfig)(env_namespaceObject.AZURE_OPENAI_ENDPOINT),
-                apiVersion: (0, env_namespaceObject.getAIConfig)(env_namespaceObject.AZURE_OPENAI_API_VERSION),
-                deployment: (0, env_namespaceObject.getAIConfig)(env_namespaceObject.AZURE_OPENAI_DEPLOYMENT),
+                apiKey: azureOpenaiKey,
+                endpoint: azureOpenaiEndpoint,
+                apiVersion: azureOpenaiApiVersion,
+                deployment: azureOpenaiDeployment,
                 dangerouslyAllowBrowser: true,
-                ...extraConfig,
-                ...extraAzureConfig
-            });
-        } else if (!(0, env_namespaceObject.getAIConfig)(env_namespaceObject.MIDSCENE_USE_ANTHROPIC_SDK)) {
-            const baseURL = (0, env_namespaceObject.getAIConfig)(env_namespaceObject.OPENAI_BASE_URL);
-            if ('string' == typeof baseURL) {
-                if (!/^https?:\/\//.test(baseURL)) throw new Error(`OPENAI_BASE_URL must be a valid URL starting with http:// or https://, but got: ${baseURL}\nPlease check your config.`);
-            }
-            openai = new (external_openai_default())({
-                baseURL: (0, env_namespaceObject.getAIConfig)(env_namespaceObject.OPENAI_BASE_URL),
-                apiKey: (0, env_namespaceObject.getAIConfig)(env_namespaceObject.OPENAI_API_KEY),
-                httpAgent: proxyAgent,
-                ...extraConfig,
-                defaultHeaders: {
-                    ...(null == extraConfig ? void 0 : extraConfig.defaultHeaders) || {},
-                    [env_namespaceObject.MIDSCENE_API_TYPE]: AIActionTypeValue.toString()
-                },
-                dangerouslyAllowBrowser: true
+                ...openaiExtraConfig,
+                ...azureExtraConfig
             });
-        }
+        } else if (!useAnthropicSdk) openai = new (external_openai_default())({
+            baseURL: openaiBaseURL,
+            apiKey: openaiApiKey,
+            httpAgent: proxyAgent,
+            ...openaiExtraConfig,
+            defaultHeaders: {
+                ...(null == openaiExtraConfig ? void 0 : openaiExtraConfig.defaultHeaders) || {},
+                [env_namespaceObject.MIDSCENE_API_TYPE]: AIActionTypeValue.toString()
+            },
+            dangerouslyAllowBrowser: true
+        });
         if (openai && (0, env_namespaceObject.getAIConfigInBoolean)(env_namespaceObject.MIDSCENE_LANGSMITH_DEBUG)) {
             if (utils_namespaceObject.ifInBrowser) throw new Error('langsmith is not supported in browser');
             console.log('DEBUGGING MODE: langsmith wrapper enabled');
@@ -182,53 +135,51 @@ var __webpack_exports__ = {};
         }
         if (void 0 !== openai) return {
             completion: openai.chat.completions,
-            style: 'openai'
+            style: 'openai',
+            modelName
         };
-        if ((0, env_namespaceObject.getAIConfig)(env_namespaceObject.MIDSCENE_USE_ANTHROPIC_SDK)) {
-            const apiKey = (0, env_namespaceObject.getAIConfig)(env_namespaceObject.ANTHROPIC_API_KEY);
-            (0, utils_namespaceObject.assert)(apiKey, 'ANTHROPIC_API_KEY is required');
-            openai = new sdk_namespaceObject.Anthropic({
-                apiKey,
-                httpAgent: proxyAgent,
-                dangerouslyAllowBrowser: true
-            });
-        }
+        if (useAnthropicSdk) openai = new sdk_namespaceObject.Anthropic({
+            apiKey: anthropicApiKey,
+            httpAgent: proxyAgent,
+            dangerouslyAllowBrowser: true
+        });
         if (void 0 !== openai && openai.messages) return {
             completion: openai.messages,
-            style: 'anthropic'
+            style: 'anthropic',
+            modelName
         };
         throw new Error('Openai SDK or Anthropic SDK is not initialized');
     }
-    async function call(messages, AIActionTypeValue, responseFormat, options) {
-        (0, utils_namespaceObject.assert)(checkAIConfig(), 'Cannot find config for AI model service. If you are using a self-hosted model without validating the API key, please set `OPENAI_API_KEY` to any non-null value. https://midscenejs.com/model-provider.html');
-        const { completion, style } = await createChatClient({
-            AIActionTypeValue
+    async function call(messages, AIActionTypeValue, modelPreferences, options) {
+        const { completion, style, modelName } = await createChatClient({
+            AIActionTypeValue,
+            modelPreferences
         });
+        const responseFormat = getResponseFormat(modelName, AIActionTypeValue);
         const maxTokens = (0, env_namespaceObject.getAIConfig)(env_namespaceObject.OPENAI_MAX_TOKENS);
         const debugCall = (0, logger_namespaceObject.getDebug)('ai:call');
         const debugProfileStats = (0, logger_namespaceObject.getDebug)('ai:profile:stats');
         const debugProfileDetail = (0, logger_namespaceObject.getDebug)('ai:profile:detail');
         const startTime = Date.now();
-        const model = getModelName();
         const isStreaming = (null == options ? void 0 : options.stream) && (null == options ? void 0 : options.onChunk);
         let content;
         let accumulated = '';
         let usage;
         let timeCost;
         const commonConfig = {
-            temperature: 'vlm-ui-tars' === (0, env_namespaceObject.vlLocateMode)() ? 0.0 : 0.1,
+            temperature: 'vlm-ui-tars' === (0, env_namespaceObject.vlLocateMode)(modelPreferences) ? 0.0 : 0.1,
             stream: !!isStreaming,
             max_tokens: 'number' == typeof maxTokens ? maxTokens : Number.parseInt(maxTokens || '2048', 10),
-            ...'qwen-vl' === (0, env_namespaceObject.vlLocateMode)() ? {
+            ...'qwen-vl' === (0, env_namespaceObject.vlLocateMode)(modelPreferences) ? {
                 vl_high_resolution_images: true
             } : {}
         };
         try {
             if ('openai' === style) {
-                debugCall(`sending ${isStreaming ? 'streaming ' : ''}request to ${model}`);
+                debugCall(`sending ${isStreaming ? 'streaming ' : ''}request to ${modelName}`);
                 if (isStreaming) {
                     const stream = await completion.create({
-                        model,
+                        model: modelName,
                         messages,
                         response_format: responseFormat,
                         ...commonConfig
@@ -270,7 +221,8 @@ var __webpack_exports__ = {};
                                     prompt_tokens: usage.prompt_tokens ?? 0,
                                     completion_tokens: usage.completion_tokens ?? 0,
                                     total_tokens: usage.total_tokens ?? 0,
-                                    time_cost: timeCost ?? 0
+                                    time_cost: timeCost ?? 0,
+                                    model_name: modelName
                                 }
                             };
                             options.onChunk(finalChunk);
@@ -278,17 +230,17 @@ var __webpack_exports__ = {};
                         }
                     }
                     content = accumulated;
-                    debugProfileStats(`streaming model, ${model}, mode, ${(0, env_namespaceObject.vlLocateMode)() || 'default'}, cost-ms, ${timeCost}`);
+                    debugProfileStats(`streaming model, ${modelName}, mode, ${(0, env_namespaceObject.vlLocateMode)(modelPreferences) || 'default'}, cost-ms, ${timeCost}`);
                 } else {
                     var _result_usage, _result_usage1, _result_usage2;
                     const result = await completion.create({
-                        model,
+                        model: modelName,
                         messages,
                         response_format: responseFormat,
                         ...commonConfig
                     });
                     timeCost = Date.now() - startTime;
-                    debugProfileStats(`model, ${model}, mode, ${(0, env_namespaceObject.vlLocateMode)() || 'default'}, ui-tars-version, ${(0, env_namespaceObject.uiTarsModelVersion)()}, prompt-tokens, ${(null == (_result_usage = result.usage) ? void 0 : _result_usage.prompt_tokens) || ''}, completion-tokens, ${(null == (_result_usage1 = result.usage) ? void 0 : _result_usage1.completion_tokens) || ''}, total-tokens, ${(null == (_result_usage2 = result.usage) ? void 0 : _result_usage2.total_tokens) || ''}, cost-ms, ${timeCost}, requestId, ${result._request_id || ''}`);
+                    debugProfileStats(`model, ${modelName}, mode, ${(0, env_namespaceObject.vlLocateMode)(modelPreferences) || 'default'}, ui-tars-version, ${(0, env_namespaceObject.uiTarsModelVersion)(modelPreferences)}, prompt-tokens, ${(null == (_result_usage = result.usage) ? void 0 : _result_usage.prompt_tokens) || ''}, completion-tokens, ${(null == (_result_usage1 = result.usage) ? void 0 : _result_usage1.completion_tokens) || ''}, total-tokens, ${(null == (_result_usage2 = result.usage) ? void 0 : _result_usage2.total_tokens) || ''}, cost-ms, ${timeCost}, requestId, ${result._request_id || ''}`);
                     debugProfileDetail(`model usage detail: ${JSON.stringify(result.usage)}`);
                     (0, utils_namespaceObject.assert)(result.choices, `invalid response from LLM service: ${JSON.stringify(result)}`);
                     content = result.choices[0].message.content;
@@ -315,7 +267,7 @@ var __webpack_exports__ = {};
                 };
                 if (isStreaming) {
                     const stream = await completion.create({
-                        model,
+                        model: modelName,
                         system: 'You are a versatile professional in software UI automation',
                         messages: messages.map((m)=>({
                                 role: 'user',
@@ -350,7 +302,8 @@ var __webpack_exports__ = {};
                                     prompt_tokens: anthropicUsage.input_tokens ?? 0,
                                     completion_tokens: anthropicUsage.output_tokens ?? 0,
                                     total_tokens: (anthropicUsage.input_tokens ?? 0) + (anthropicUsage.output_tokens ?? 0),
-                                    time_cost: timeCost ?? 0
+                                    time_cost: timeCost ?? 0,
+                                    model_name: modelName
                                 } : void 0
                             };
                             options.onChunk(finalChunk);
@@ -360,7 +313,7 @@ var __webpack_exports__ = {};
                     content = accumulated;
                 } else {
                     const result = await completion.create({
-                        model,
+                        model: modelName,
                         system: 'You are a versatile professional in software UI automation',
                         messages: messages.map((m)=>({
                                 role: 'user',
@@ -389,7 +342,8 @@ var __webpack_exports__ = {};
                     prompt_tokens: usage.prompt_tokens ?? 0,
                     completion_tokens: usage.completion_tokens ?? 0,
                     total_tokens: usage.total_tokens ?? 0,
-                    time_cost: timeCost ?? 0
+                    time_cost: timeCost ?? 0,
+                    model_name: modelName
                 } : void 0,
                 isStreamed: !!isStreaming
             };
@@ -401,10 +355,9 @@ var __webpack_exports__ = {};
             throw newError;
         }
     }
-    async function callToGetJSONObject(messages, AIActionTypeValue) {
+    const getResponseFormat = (modelName, AIActionTypeValue)=>{
         let responseFormat;
-        const model = getModelName();
-        if (model.includes('gpt-4')) switch(AIActionTypeValue){
+        if (modelName.includes('gpt-4')) switch(AIActionTypeValue){
             case external_common_js_namespaceObject.AIActionType.ASSERT:
                 responseFormat = assertion_js_namespaceObject.assertSchema;
                 break;
@@ -421,19 +374,22 @@ var __webpack_exports__ = {};
                 };
                 break;
         }
-        if ('gpt-4o-2024-05-13' === model) responseFormat = {
+        if ('gpt-4o-2024-05-13' === modelName) responseFormat = {
             type: external_types_js_namespaceObject.AIResponseFormat.JSON
         };
-        const response = await call(messages, AIActionTypeValue, responseFormat);
+        return responseFormat;
+    };
+    async function callToGetJSONObject(messages, AIActionTypeValue, modelPreferences) {
+        const response = await call(messages, AIActionTypeValue, modelPreferences);
         (0, utils_namespaceObject.assert)(response, 'empty response');
-        const jsonContent = safeParseJson(response.content);
+        const jsonContent = safeParseJson(response.content, modelPreferences);
         return {
             content: jsonContent,
             usage: response.usage
         };
     }
-    async function callAiFnWithStringResponse(msgs, AIActionTypeValue) {
-        const { content, usage } = await call(msgs, AIActionTypeValue);
+    async function callAiFnWithStringResponse(msgs, AIActionTypeValue, modelPreferences) {
+        const { content, usage } = await call(msgs, AIActionTypeValue, modelPreferences);
         return {
             content,
             usage
@@ -454,7 +410,7 @@ var __webpack_exports__ = {};
         if (input.includes('bbox')) while(/\d+\s+\d+/.test(input))input = input.replace(/(\d+)\s+(\d+)/g, '$1,$2');
         return input;
     }
-    function safeParseJson(input) {
+    function safeParseJson(input, modelPreferences) {
         const cleanJsonString = extractJSONFromCodeBlock(input);
         if (null == cleanJsonString ? void 0 : cleanJsonString.match(/\((\d+),(\d+)\)/)) {
             var _cleanJsonString_match;
@@ -466,7 +422,7 @@ var __webpack_exports__ = {};
         try {
             return JSON.parse((0, external_jsonrepair_namespaceObject.jsonrepair)(cleanJsonString));
         } catch (e) {}
-        if ('doubao-vision' === (0, env_namespaceObject.vlLocateMode)() || 'vlm-ui-tars' === (0, env_namespaceObject.vlLocateMode)()) {
+        if ('doubao-vision' === (0, env_namespaceObject.vlLocateMode)(modelPreferences) || 'vlm-ui-tars' === (0, env_namespaceObject.vlLocateMode)(modelPreferences)) {
             const jsonString = preprocessDoubaoBboxJson(cleanJsonString);
             return JSON.parse((0, external_jsonrepair_namespaceObject.jsonrepair)(jsonString));
         }
@@ -476,18 +432,16 @@ var __webpack_exports__ = {};
 exports.call = __webpack_exports__.call;
 exports.callAiFnWithStringResponse = __webpack_exports__.callAiFnWithStringResponse;
 exports.callToGetJSONObject = __webpack_exports__.callToGetJSONObject;
-exports.checkAIConfig = __webpack_exports__.checkAIConfig;
 exports.extractJSONFromCodeBlock = __webpack_exports__.extractJSONFromCodeBlock;
-exports.getModelName = __webpack_exports__.getModelName;
+exports.getResponseFormat = __webpack_exports__.getResponseFormat;
 exports.preprocessDoubaoBboxJson = __webpack_exports__.preprocessDoubaoBboxJson;
 exports.safeParseJson = __webpack_exports__.safeParseJson;
 for(var __webpack_i__ in __webpack_exports__)if (-1 === [
     "call",
     "callAiFnWithStringResponse",
     "callToGetJSONObject",
-    "checkAIConfig",
     "extractJSONFromCodeBlock",
-    "getModelName",
+    "getResponseFormat",
     "preprocessDoubaoBboxJson",
     "safeParseJson"
 ].indexOf(__webpack_i__)) exports[__webpack_i__] = __webpack_exports__[__webpack_i__];

package/dist/lib/ai-model/service-caller/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"ai-model/service-caller/index.js","sources":["webpack://@midscene/core/webpack/runtime/compat_get_default_export","webpack://@midscene/core/webpack/runtime/define_property_getters","webpack://@midscene/core/webpack/runtime/has_own_property","webpack://@midscene/core/webpack/runtime/make_namespace_object","webpack://@midscene/core/./src/ai-model/service-caller/index.ts"],"sourcesContent":["// getDefaultExport function for compatibility with non-ESM modules\n__webpack_require__.n = (module) => {\n\tvar getter = module && module.__esModule ?\n\t\t() => (module['default']) :\n\t\t() => (module);\n\t__webpack_require__.d(getter, { a: getter });\n\treturn getter;\n};\n","__webpack_require__.d = (exports, definition) => {\n\tfor(var key in definition) {\n        if(__webpack_require__.o(definition, key) && !__webpack_require__.o(exports, key)) {\n            Object.defineProperty(exports, key, { enumerable: true, get: definition[key] });\n        }\n    }\n};","__webpack_require__.o = (obj, prop) => (Object.prototype.hasOwnProperty.call(obj, prop))","// define __esModule on exports\n__webpack_require__.r = (exports) => {\n\tif(typeof Symbol !== 'undefined' && Symbol.toStringTag) {\n\t\tObject.defineProperty(exports, Symbol.toStringTag, { value: 'Module' });\n\t}\n\tObject.defineProperty(exports, '__esModule', { value: true });\n};","import { AIResponseFormat, type AIUsageInfo } from '@/types';\nimport type { CodeGenerationChunk, StreamingCallback } from '@/types';\nimport { Anthropic } from '@anthropic-ai/sdk';\nimport {\n  DefaultAzureCredential,\n  getBearerTokenProvider,\n} from '@azure/identity';\nimport {\n  ANTHROPIC_API_KEY,\n  AZURE_OPENAI_API_VERSION,\n  AZURE_OPENAI_DEPLOYMENT,\n  AZURE_OPENAI_ENDPOINT,\n  AZURE_OPENAI_KEY,\n  MIDSCENE_API_TYPE,\n  MIDSCENE_AZURE_OPENAI_INIT_CONFIG_JSON,\n  MIDSCENE_AZURE_OPENAI_SCOPE,\n  MIDSCENE_DEBUG_AI_PROFILE,\n  MIDSCENE_DEBUG_AI_RESPONSE,\n  MIDSCENE_LANGSMITH_DEBUG,\n  MIDSCENE_MODEL_NAME,\n  MIDSCENE_OPENAI_HTTP_PROXY,\n  MIDSCENE_OPENAI_INIT_CONFIG_JSON,\n  MIDSCENE_OPENAI_SOCKS_PROXY,\n  MIDSCENE_USE_ANTHROPIC_SDK,\n  MIDSCENE_USE_AZURE_OPENAI,\n  OPENAI_API_KEY,\n  OPENAI_BASE_URL,\n  OPENAI_MAX_TOKENS,\n  OPENAI_USE_AZURE,\n  getAIConfig,\n  getAIConfigInBoolean,\n  getAIConfigInJson,\n  uiTarsModelVersion,\n  vlLocateMode,\n} from '@midscene/shared/env';\nimport { parseBase64 } from '@midscene/shared/img';\nimport { enableDebug, getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport { ifInBrowser } from '@midscene/shared/utils';\nimport { HttpsProxyAgent } from 'https-proxy-agent';\nimport { jsonrepair } from 'jsonrepair';\nimport OpenAI, { AzureOpenAI } from 'openai';\nimport type { ChatCompletionMessageParam } from 'openai/resources/index';\nimport type { Stream } from 'openai/streaming';\nimport { SocksProxyAgent } from 'socks-proxy-agent';\nimport { AIActionType, type AIArgs } from '../common';\nimport { assertSchema } from '../prompt/assertion';\nimport { locatorSchema } from '../prompt/llm-locator';\nimport { planSchema } from '../prompt/llm-planning';\n\nexport function checkAIConfig() {\n  const openaiKey = getAIConfig(OPENAI_API_KEY);\n  const azureConfig = getAIConfig(MIDSCENE_USE_AZURE_OPENAI);\n  const anthropicKey = getAIConfig(ANTHROPIC_API_KEY);\n  const initConfigJson = getAIConfig(MIDSCENE_OPENAI_INIT_CONFIG_JSON);\n\n  if (openaiKey) return true;\n  if (azureConfig) return true;\n  if (anthropicKey) return true;\n\n  return Boolean(initConfigJson);\n}\n\n// if debug config is initialized\nlet debugConfigInitialized = false;\n\nfunction initDebugConfig() {\n  // if debug config is initialized, return\n  if (debugConfigInitialized) return;\n\n  const shouldPrintTiming = getAIConfigInBoolean(MIDSCENE_DEBUG_AI_PROFILE);\n  let debugConfig = '';\n  if (shouldPrintTiming) {\n    console.warn(\n      'MIDSCENE_DEBUG_AI_PROFILE is deprecated, use DEBUG=midscene:ai:profile instead',\n    );\n    debugConfig = 'ai:profile';\n  }\n  const shouldPrintAIResponse = getAIConfigInBoolean(\n    MIDSCENE_DEBUG_AI_RESPONSE,\n  );\n  if (shouldPrintAIResponse) {\n    console.warn(\n      'MIDSCENE_DEBUG_AI_RESPONSE is deprecated, use DEBUG=midscene:ai:response instead',\n    );\n    if (debugConfig) {\n      debugConfig = 'ai:*';\n    } else {\n      debugConfig = 'ai:call';\n    }\n  }\n  if (debugConfig) {\n    enableDebug(debugConfig);\n  }\n\n  // mark as initialized\n  debugConfigInitialized = true;\n}\n\n// default model\nconst defaultModel = 'gpt-4o';\nexport function getModelName() {\n  let modelName = defaultModel;\n  const nameInConfig = getAIConfig(MIDSCENE_MODEL_NAME);\n  if (nameInConfig) {\n    modelName = nameInConfig;\n  }\n  return modelName;\n}\n\nasync function createChatClient({\n  AIActionTypeValue,\n}: {\n  AIActionTypeValue: AIActionType;\n}): Promise<{\n  completion: OpenAI.Chat.Completions;\n  style: 'openai' | 'anthropic';\n}> {\n  initDebugConfig();\n  let openai: OpenAI | AzureOpenAI | undefined;\n  const extraConfig = getAIConfigInJson(MIDSCENE_OPENAI_INIT_CONFIG_JSON);\n\n  const socksProxy = getAIConfig(MIDSCENE_OPENAI_SOCKS_PROXY);\n  const httpProxy = getAIConfig(MIDSCENE_OPENAI_HTTP_PROXY);\n\n  let proxyAgent = undefined;\n  const debugProxy = getDebug('ai:call:proxy');\n  if (httpProxy) {\n    debugProxy('using http proxy', httpProxy);\n    proxyAgent = new HttpsProxyAgent(httpProxy);\n  } else if (socksProxy) {\n    debugProxy('using socks proxy', socksProxy);\n    proxyAgent = new SocksProxyAgent(socksProxy);\n  }\n\n  if (getAIConfig(OPENAI_USE_AZURE)) {\n    // this is deprecated\n    openai = new AzureOpenAI({\n      baseURL: getAIConfig(OPENAI_BASE_URL),\n      apiKey: getAIConfig(OPENAI_API_KEY),\n      httpAgent: proxyAgent,\n      ...extraConfig,\n      dangerouslyAllowBrowser: true,\n    }) as OpenAI;\n  } else if (getAIConfig(MIDSCENE_USE_AZURE_OPENAI)) {\n    const extraAzureConfig = getAIConfigInJson(\n      MIDSCENE_AZURE_OPENAI_INIT_CONFIG_JSON,\n    );\n\n    // https://learn.microsoft.com/en-us/azure/ai-services/openai/chatgpt-quickstart?tabs=bash%2Cjavascript-key%2Ctypescript-keyless%2Cpython&pivots=programming-language-javascript#rest-api\n    // keyless authentication\n    const scope = getAIConfig(MIDSCENE_AZURE_OPENAI_SCOPE);\n    let tokenProvider: any = undefined;\n    if (scope) {\n      assert(\n        !ifInBrowser,\n        'Azure OpenAI is not supported in browser with Midscene.',\n      );\n      const credential = new DefaultAzureCredential();\n\n      assert(scope, 'MIDSCENE_AZURE_OPENAI_SCOPE is required');\n      tokenProvider = getBearerTokenProvider(credential, scope);\n\n      openai = new AzureOpenAI({\n        azureADTokenProvider: tokenProvider,\n        endpoint: getAIConfig(AZURE_OPENAI_ENDPOINT),\n        apiVersion: getAIConfig(AZURE_OPENAI_API_VERSION),\n        deployment: getAIConfig(AZURE_OPENAI_DEPLOYMENT),\n        ...extraConfig,\n        ...extraAzureConfig,\n      });\n    } else {\n      // endpoint, apiKey, apiVersion, deployment\n      openai = new AzureOpenAI({\n        apiKey: getAIConfig(AZURE_OPENAI_KEY),\n        endpoint: getAIConfig(AZURE_OPENAI_ENDPOINT),\n        apiVersion: getAIConfig(AZURE_OPENAI_API_VERSION),\n        deployment: getAIConfig(AZURE_OPENAI_DEPLOYMENT),\n        dangerouslyAllowBrowser: true,\n        ...extraConfig,\n        ...extraAzureConfig,\n      });\n    }\n  } else if (!getAIConfig(MIDSCENE_USE_ANTHROPIC_SDK)) {\n    const baseURL = getAIConfig(OPENAI_BASE_URL);\n    if (typeof baseURL === 'string') {\n      if (!/^https?:\\/\\//.test(baseURL)) {\n        throw new Error(\n          `OPENAI_BASE_URL must be a valid URL starting with http:// or https://, but got: ${baseURL}\\nPlease check your config.`,\n        );\n      }\n    }\n\n    openai = new OpenAI({\n      baseURL: getAIConfig(OPENAI_BASE_URL),\n      apiKey: getAIConfig(OPENAI_API_KEY),\n      httpAgent: proxyAgent,\n      ...extraConfig,\n      defaultHeaders: {\n        ...(extraConfig?.defaultHeaders || {}),\n        [MIDSCENE_API_TYPE]: AIActionTypeValue.toString(),\n      },\n      dangerouslyAllowBrowser: true,\n    });\n  }\n\n  if (openai && getAIConfigInBoolean(MIDSCENE_LANGSMITH_DEBUG)) {\n    if (ifInBrowser) {\n      throw new Error('langsmith is not supported in browser');\n    }\n    console.log('DEBUGGING MODE: langsmith wrapper enabled');\n    const { wrapOpenAI } = await import('langsmith/wrappers');\n    openai = wrapOpenAI(openai);\n  }\n\n  if (typeof openai !== 'undefined') {\n    return {\n      completion: openai.chat.completions,\n      style: 'openai',\n    };\n  }\n\n  // Anthropic\n  if (getAIConfig(MIDSCENE_USE_ANTHROPIC_SDK)) {\n    const apiKey = getAIConfig(ANTHROPIC_API_KEY);\n    assert(apiKey, 'ANTHROPIC_API_KEY is required');\n    openai = new Anthropic({\n      apiKey,\n      httpAgent: proxyAgent,\n      dangerouslyAllowBrowser: true,\n    }) as any;\n  }\n\n  if (typeof openai !== 'undefined' && (openai as any).messages) {\n    return {\n      completion: (openai as any).messages,\n      style: 'anthropic',\n    };\n  }\n\n  throw new Error('Openai SDK or Anthropic SDK is not initialized');\n}\n\nexport async function call(\n  messages: ChatCompletionMessageParam[],\n  AIActionTypeValue: AIActionType,\n  responseFormat?:\n    | OpenAI.ChatCompletionCreateParams['response_format']\n    | OpenAI.ResponseFormatJSONObject,\n  options?: {\n    stream?: boolean;\n    onChunk?: StreamingCallback;\n  },\n): Promise<{ content: string; usage?: AIUsageInfo; isStreamed: boolean }> {\n  assert(\n    checkAIConfig(),\n    'Cannot find config for AI model service. If you are using a self-hosted model without validating the API key, please set `OPENAI_API_KEY` to any non-null value. https://midscenejs.com/model-provider.html',\n  );\n\n  const { completion, style } = await createChatClient({\n    AIActionTypeValue,\n  });\n\n  const maxTokens = getAIConfig(OPENAI_MAX_TOKENS);\n  const debugCall = getDebug('ai:call');\n  const debugProfileStats = getDebug('ai:profile:stats');\n  const debugProfileDetail = getDebug('ai:profile:detail');\n\n  const startTime = Date.now();\n  const model = getModelName();\n  const isStreaming = options?.stream && options?.onChunk;\n  let content: string | undefined;\n  let accumulated = '';\n  let usage: OpenAI.CompletionUsage | undefined;\n  let timeCost: number | undefined;\n\n  const commonConfig = {\n    temperature: vlLocateMode() === 'vlm-ui-tars' ? 0.0 : 0.1,\n    stream: !!isStreaming,\n    max_tokens:\n      typeof maxTokens === 'number'\n        ? maxTokens\n        : Number.parseInt(maxTokens || '2048', 10),\n    ...(vlLocateMode() === 'qwen-vl' // qwen specific config\n      ? {\n          vl_high_resolution_images: true,\n        }\n      : {}),\n  };\n\n  try {\n    if (style === 'openai') {\n      debugCall(\n        `sending ${isStreaming ? 'streaming ' : ''}request to ${model}`,\n      );\n\n      if (isStreaming) {\n        const stream = (await completion.create(\n          {\n            model,\n            messages,\n            response_format: responseFormat,\n            ...commonConfig,\n          },\n          {\n            stream: true,\n          },\n        )) as Stream<OpenAI.Chat.Completions.ChatCompletionChunk> & {\n          _request_id?: string | null;\n        };\n\n        for await (const chunk of stream) {\n          const content = chunk.choices?.[0]?.delta?.content || '';\n          const reasoning_content =\n            (chunk.choices?.[0]?.delta as any)?.reasoning_content || '';\n\n          // Check for usage info in any chunk (OpenAI provides usage in separate chunks)\n          if (chunk.usage) {\n            usage = chunk.usage;\n          }\n\n          if (content || reasoning_content) {\n            accumulated += content;\n            const chunkData: CodeGenerationChunk = {\n              content,\n              reasoning_content,\n              accumulated,\n              isComplete: false,\n              usage: undefined,\n            };\n            options.onChunk!(chunkData);\n          }\n\n          // Check if stream is complete\n          if (chunk.choices?.[0]?.finish_reason) {\n            timeCost = Date.now() - startTime;\n\n            // If usage is not available from the stream, provide a basic usage info\n            if (!usage) {\n              // Estimate token counts based on content length (rough approximation)\n              const estimatedTokens = Math.max(\n                1,\n                Math.floor(accumulated.length / 4),\n              );\n              usage = {\n                prompt_tokens: estimatedTokens,\n                completion_tokens: estimatedTokens,\n                total_tokens: estimatedTokens * 2,\n              };\n            }\n\n            // Send final chunk\n            const finalChunk: CodeGenerationChunk = {\n              content: '',\n              accumulated,\n              reasoning_content: '',\n              isComplete: true,\n              usage: {\n                prompt_tokens: usage.prompt_tokens ?? 0,\n                completion_tokens: usage.completion_tokens ?? 0,\n                total_tokens: usage.total_tokens ?? 0,\n                time_cost: timeCost ?? 0,\n              },\n            };\n            options.onChunk!(finalChunk);\n            break;\n          }\n        }\n        content = accumulated;\n        debugProfileStats(\n          `streaming model, ${model}, mode, ${vlLocateMode() || 'default'}, cost-ms, ${timeCost}`,\n        );\n      } else {\n        const result = await completion.create({\n          model,\n          messages,\n          response_format: responseFormat,\n          ...commonConfig,\n        } as any);\n        timeCost = Date.now() - startTime;\n\n        debugProfileStats(\n          `model, ${model}, mode, ${vlLocateMode() || 'default'}, ui-tars-version, ${uiTarsModelVersion()}, prompt-tokens, ${result.usage?.prompt_tokens || ''}, completion-tokens, ${result.usage?.completion_tokens || ''}, total-tokens, ${result.usage?.total_tokens || ''}, cost-ms, ${timeCost}, requestId, ${result._request_id || ''}`,\n        );\n\n        debugProfileDetail(\n          `model usage detail: ${JSON.stringify(result.usage)}`,\n        );\n\n        assert(\n          result.choices,\n          `invalid response from LLM service: ${JSON.stringify(result)}`,\n        );\n        content = result.choices[0].message.content!;\n        usage = result.usage;\n      }\n\n      debugCall(`response: ${content}`);\n      assert(content, 'empty content');\n    } else if (style === 'anthropic') {\n      const convertImageContent = (content: any) => {\n        if (content.type === 'image_url') {\n          const imgBase64 = content.image_url.url;\n          assert(imgBase64, 'image_url is required');\n          const { mimeType, body } = parseBase64(content.image_url.url);\n          return {\n            source: {\n              type: 'base64',\n              media_type: mimeType,\n              data: body,\n            },\n            type: 'image',\n          };\n        }\n        return content;\n      };\n\n      if (isStreaming) {\n        const stream = (await completion.create({\n          model,\n          system: 'You are a versatile professional in software UI automation',\n          messages: messages.map((m) => ({\n            role: 'user',\n            content: Array.isArray(m.content)\n              ? (m.content as any).map(convertImageContent)\n              : m.content,\n          })),\n          response_format: responseFormat,\n          ...commonConfig,\n        } as any)) as any;\n\n        for await (const chunk of stream) {\n          const content = chunk.delta?.text || '';\n          if (content) {\n            accumulated += content;\n            const chunkData: CodeGenerationChunk = {\n              content,\n              accumulated,\n              reasoning_content: '',\n              isComplete: false,\n              usage: undefined,\n            };\n            options.onChunk!(chunkData);\n          }\n\n          // Check if stream is complete\n          if (chunk.type === 'message_stop') {\n            timeCost = Date.now() - startTime;\n            const anthropicUsage = chunk.usage;\n\n            // Send final chunk\n            const finalChunk: CodeGenerationChunk = {\n              content: '',\n              accumulated,\n              reasoning_content: '',\n              isComplete: true,\n              usage: anthropicUsage\n                ? {\n                    prompt_tokens: anthropicUsage.input_tokens ?? 0,\n                    completion_tokens: anthropicUsage.output_tokens ?? 0,\n                    total_tokens:\n                      (anthropicUsage.input_tokens ?? 0) +\n                      (anthropicUsage.output_tokens ?? 0),\n                    time_cost: timeCost ?? 0,\n                  }\n                : undefined,\n            };\n            options.onChunk!(finalChunk);\n            break;\n          }\n        }\n        content = accumulated;\n      } else {\n        const result = await completion.create({\n          model,\n          system: 'You are a versatile professional in software UI automation',\n          messages: messages.map((m) => ({\n            role: 'user',\n            content: Array.isArray(m.content)\n              ? (m.content as any).map(convertImageContent)\n              : m.content,\n          })),\n          response_format: responseFormat,\n          ...commonConfig,\n        } as any);\n        timeCost = Date.now() - startTime;\n        content = (result as any).content[0].text as string;\n        usage = result.usage;\n      }\n\n      assert(content, 'empty content');\n    }\n    // Ensure we always have usage info for streaming responses\n    if (isStreaming && !usage) {\n      // Estimate token counts based on content length (rough approximation)\n      const estimatedTokens = Math.max(\n        1,\n        Math.floor((content || '').length / 4),\n      );\n      usage = {\n        prompt_tokens: estimatedTokens,\n        completion_tokens: estimatedTokens,\n        total_tokens: estimatedTokens * 2,\n      };\n    }\n\n    return {\n      content: content || '',\n      usage: usage\n        ? {\n            prompt_tokens: usage.prompt_tokens ?? 0,\n            completion_tokens: usage.completion_tokens ?? 0,\n            total_tokens: usage.total_tokens ?? 0,\n            time_cost: timeCost ?? 0,\n          }\n        : undefined,\n      isStreamed: !!isStreaming,\n    };\n  } catch (e: any) {\n    console.error(' call AI error', e);\n    const newError = new Error(\n      `failed to call ${isStreaming ? 'streaming ' : ''}AI model service: ${e.message}. Trouble shooting: https://midscenejs.com/model-provider.html`,\n      {\n        cause: e,\n      },\n    );\n    throw newError;\n  }\n}\n\nexport async function callToGetJSONObject<T>(\n  messages: ChatCompletionMessageParam[],\n  AIActionTypeValue: AIActionType,\n): Promise<{ content: T; usage?: AIUsageInfo }> {\n  let responseFormat:\n    | OpenAI.ChatCompletionCreateParams['response_format']\n    | OpenAI.ResponseFormatJSONObject\n    | undefined;\n\n  const model = getModelName();\n\n  if (model.includes('gpt-4')) {\n    switch (AIActionTypeValue) {\n      case AIActionType.ASSERT:\n        responseFormat = assertSchema;\n        break;\n      case AIActionType.INSPECT_ELEMENT:\n        responseFormat = locatorSchema;\n        break;\n      case AIActionType.PLAN:\n        responseFormat = planSchema;\n        break;\n      case AIActionType.EXTRACT_DATA:\n      case AIActionType.DESCRIBE_ELEMENT:\n        responseFormat = { type: AIResponseFormat.JSON };\n        break;\n    }\n  }\n\n  // gpt-4o-2024-05-13 only supports json_object response format\n  if (model === 'gpt-4o-2024-05-13') {\n    responseFormat = { type: AIResponseFormat.JSON };\n  }\n\n  const response = await call(messages, AIActionTypeValue, responseFormat);\n  assert(response, 'empty response');\n  const jsonContent = safeParseJson(response.content);\n  return { content: jsonContent, usage: response.usage };\n}\n\nexport async function callAiFnWithStringResponse<T>(\n  msgs: AIArgs,\n  AIActionTypeValue: AIActionType,\n): Promise<{ content: string; usage?: AIUsageInfo }> {\n  const { content, usage } = await call(msgs, AIActionTypeValue);\n  return { content, usage };\n}\n\nexport function extractJSONFromCodeBlock(response: string) {\n  try {\n    // First, try to match a JSON object directly in the response\n    const jsonMatch = response.match(/^\\s*(\\{[\\s\\S]*\\})\\s*$/);\n    if (jsonMatch) {\n      return jsonMatch[1];\n    }\n\n    // If no direct JSON object is found, try to extract JSON from a code block\n    const codeBlockMatch = response.match(\n      /```(?:json)?\\s*(\\{[\\s\\S]*?\\})\\s*```/,\n    );\n    if (codeBlockMatch) {\n      return codeBlockMatch[1];\n    }\n\n    // If no code block is found, try to find a JSON-like structure in the text\n    const jsonLikeMatch = response.match(/\\{[\\s\\S]*\\}/);\n    if (jsonLikeMatch) {\n      return jsonLikeMatch[0];\n    }\n  } catch {}\n  // If no JSON-like structure is found, return the original response\n  return response;\n}\n\nexport function preprocessDoubaoBboxJson(input: string) {\n  if (input.includes('bbox')) {\n    // when its values like 940 445 969 490, replace all /\\d+\\s+\\d+/g with /$1,$2/g\n    while (/\\d+\\s+\\d+/.test(input)) {\n      input = input.replace(/(\\d+)\\s+(\\d+)/g, '$1,$2');\n    }\n  }\n  return input;\n}\n\nexport function safeParseJson(input: string) {\n  const cleanJsonString = extractJSONFromCodeBlock(input);\n  // match the point\n  if (cleanJsonString?.match(/\\((\\d+),(\\d+)\\)/)) {\n    return cleanJsonString\n      .match(/\\((\\d+),(\\d+)\\)/)\n      ?.slice(1)\n      .map(Number);\n  }\n  try {\n    return JSON.parse(cleanJsonString);\n  } catch {}\n  try {\n    return JSON.parse(jsonrepair(cleanJsonString));\n  } catch (e) {}\n\n  if (vlLocateMode() === 'doubao-vision' || vlLocateMode() === 'vlm-ui-tars') {\n    const jsonString = preprocessDoubaoBboxJson(cleanJsonString);\n    return JSON.parse(jsonrepair(jsonString));\n  }\n  throw Error(`failed to parse json response: ${input}`);\n}\n"],"names":["__webpack_require__","module","getter","definition","key","Object","obj","prop","Symbol","checkAIConfig","openaiKey","getAIConfig","OPENAI_API_KEY","azureConfig","MIDSCENE_USE_AZURE_OPENAI","anthropicKey","ANTHROPIC_API_KEY","initConfigJson","MIDSCENE_OPENAI_INIT_CONFIG_JSON","Boolean","debugConfigInitialized","initDebugConfig","shouldPrintTiming","getAIConfigInBoolean","MIDSCENE_DEBUG_AI_PROFILE","debugConfig","console","shouldPrintAIResponse","MIDSCENE_DEBUG_AI_RESPONSE","enableDebug","defaultModel","getModelName","modelName","nameInConfig","MIDSCENE_MODEL_NAME","createChatClient","AIActionTypeValue","openai","extraConfig","getAIConfigInJson","socksProxy","MIDSCENE_OPENAI_SOCKS_PROXY","httpProxy","MIDSCENE_OPENAI_HTTP_PROXY","proxyAgent","debugProxy","getDebug","HttpsProxyAgent","SocksProxyAgent","OPENAI_USE_AZURE","AzureOpenAI","OPENAI_BASE_URL","extraAzureConfig","MIDSCENE_AZURE_OPENAI_INIT_CONFIG_JSON","scope","MIDSCENE_AZURE_OPENAI_SCOPE","tokenProvider","assert","ifInBrowser","credential","DefaultAzureCredential","getBearerTokenProvider","AZURE_OPENAI_ENDPOINT","AZURE_OPENAI_API_VERSION","AZURE_OPENAI_DEPLOYMENT","AZURE_OPENAI_KEY","MIDSCENE_USE_ANTHROPIC_SDK","baseURL","Error","OpenAI","MIDSCENE_API_TYPE","MIDSCENE_LANGSMITH_DEBUG","wrapOpenAI","apiKey","Anthropic","call","messages","responseFormat","options","completion","style","maxTokens","OPENAI_MAX_TOKENS","debugCall","debugProfileStats","debugProfileDetail","startTime","Date","model","isStreaming","content","accumulated","usage","timeCost","commonConfig","vlLocateMode","Number","stream","chunk","_chunk_choices__delta","_chunk_choices__delta1","_chunk_choices_2","reasoning_content","chunkData","undefined","estimatedTokens","Math","finalChunk","_result_usage","_result_usage1","_result_usage2","result","uiTarsModelVersion","JSON","convertImageContent","imgBase64","mimeType","body","parseBase64","m","Array","_chunk_delta","anthropicUsage","e","newError","callToGetJSONObject","AIActionType","assertSchema","locatorSchema","planSchema","AIResponseFormat","response","jsonContent","safeParseJson","callAiFnWithStringResponse","msgs","extractJSONFromCodeBlock","jsonMatch","codeBlockMatch","jsonLikeMatch","preprocessDoubaoBboxJson","input","cleanJsonString","_cleanJsonString_match","jsonrepair","jsonString"],"mappings":";;;;;;;;;;;;;;;;;;;IACAA,oBAAoB,CAAC,GAAG,CAACC;QACxB,IAAIC,SAASD,UAAUA,OAAO,UAAU,GACvC,IAAOA,MAAM,CAAC,UAAU,GACxB,IAAOA;QACRD,oBAAoB,CAAC,CAACE,QAAQ;YAAE,GAAGA;QAAO;QAC1C,OAAOA;IACR;;;ICPAF,oBAAoB,CAAC,GAAG,CAAC,UAASG;QACjC,IAAI,IAAIC,OAAOD,WACR,IAAGH,oBAAoB,CAAC,CAACG,YAAYC,QAAQ,CAACJ,oBAAoB,CAAC,CAAC,UAASI,MACzEC,OAAO,cAAc,CAAC,UAASD,KAAK;YAAE,YAAY;YAAM,KAAKD,UAAU,CAACC,IAAI;QAAC;IAGzF;;;ICNAJ,oBAAoB,CAAC,GAAG,CAACM,KAAKC,OAAUF,OAAO,SAAS,CAAC,cAAc,CAAC,IAAI,CAACC,KAAKC;;;ICClFP,oBAAoB,CAAC,GAAG,CAAC;QACxB,IAAG,AAAkB,eAAlB,OAAOQ,UAA0BA,OAAO,WAAW,EACrDH,OAAO,cAAc,CAAC,UAASG,OAAO,WAAW,EAAE;YAAE,OAAO;QAAS;QAEtEH,OAAO,cAAc,CAAC,UAAS,cAAc;YAAE,OAAO;QAAK;IAC5D;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;IC4CO,SAASI;QACd,MAAMC,YAAYC,AAAAA,IAAAA,oBAAAA,WAAAA,AAAAA,EAAYC,oBAAAA,cAAcA;QAC5C,MAAMC,cAAcF,AAAAA,IAAAA,oBAAAA,WAAAA,AAAAA,EAAYG,oBAAAA,yBAAyBA;QACzD,MAAMC,eAAeJ,AAAAA,IAAAA,oBAAAA,WAAAA,AAAAA,EAAYK,oBAAAA,iBAAiBA;QAClD,MAAMC,iBAAiBN,AAAAA,IAAAA,oBAAAA,WAAAA,AAAAA,EAAYO,oBAAAA,gCAAgCA;QAEnE,IAAIR,WAAW,OAAO;QACtB,IAAIG,aAAa,OAAO;QACxB,IAAIE,cAAc,OAAO;QAEzB,OAAOI,QAAQF;IACjB;IAGA,IAAIG,yBAAyB;IAE7B,SAASC;QAEP,IAAID,wBAAwB;QAE5B,MAAME,oBAAoBC,AAAAA,IAAAA,oBAAAA,oBAAAA,AAAAA,EAAqBC,oBAAAA,yBAAyBA;QACxE,IAAIC,cAAc;QAClB,IAAIH,mBAAmB;YACrBI,QAAQ,IAAI,CACV;YAEFD,cAAc;QAChB;QACA,MAAME,wBAAwBJ,AAAAA,IAAAA,oBAAAA,oBAAAA,AAAAA,EAC5BK,oBAAAA,0BAA0BA;QAE5B,IAAID,uBAAuB;YACzBD,QAAQ,IAAI,CACV;YAGAD,cADEA,cACY,SAEA;QAElB;QACA,IAAIA,aACFI,AAAAA,IAAAA,uBAAAA,WAAAA,AAAAA,EAAYJ;QAIdL,yBAAyB;IAC3B;IAGA,MAAMU,eAAe;IACd,SAASC;QACd,IAAIC,YAAYF;QAChB,MAAMG,eAAetB,AAAAA,IAAAA,oBAAAA,WAAAA,AAAAA,EAAYuB,oBAAAA,mBAAmBA;QACpD,IAAID,cACFD,YAAYC;QAEd,OAAOD;IACT;IAEA,eAAeG,iBAAiB,EAC9BC,iBAAiB,EAGlB;QAICf;QACA,IAAIgB;QACJ,MAAMC,cAAcC,AAAAA,IAAAA,oBAAAA,iBAAAA,AAAAA,EAAkBrB,oBAAAA,gCAAgCA;QAEtE,MAAMsB,aAAa7B,AAAAA,IAAAA,oBAAAA,WAAAA,AAAAA,EAAY8B,oBAAAA,2BAA2BA;QAC1D,MAAMC,YAAY/B,AAAAA,IAAAA,oBAAAA,WAAAA,AAAAA,EAAYgC,oBAAAA,0BAA0BA;QAExD,IAAIC;QACJ,MAAMC,aAAaC,AAAAA,IAAAA,uBAAAA,QAAAA,AAAAA,EAAS;QAC5B,IAAIJ,WAAW;YACbG,WAAW,oBAAoBH;YAC/BE,aAAa,IAAIG,2CAAAA,eAAeA,CAACL;QACnC,OAAO,IAAIF,YAAY;YACrBK,WAAW,qBAAqBL;YAChCI,aAAa,IAAII,2CAAAA,eAAeA,CAACR;QACnC;QAEA,IAAI7B,AAAAA,IAAAA,oBAAAA,WAAAA,AAAAA,EAAYsC,oBAAAA,gBAAgBA,GAE9BZ,SAAS,IAAIa,gCAAAA,WAAWA,CAAC;YACvB,SAASvC,AAAAA,IAAAA,oBAAAA,WAAAA,AAAAA,EAAYwC,oBAAAA,eAAeA;YACpC,QAAQxC,AAAAA,IAAAA,oBAAAA,WAAAA,AAAAA,EAAYC,oBAAAA,cAAcA;YAClC,WAAWgC;YACX,GAAGN,WAAW;YACd,yBAAyB;QAC3B;aACK,IAAI3B,AAAAA,IAAAA,oBAAAA,WAAAA,AAAAA,EAAYG,oBAAAA,yBAAyBA,GAAG;YACjD,MAAMsC,mBAAmBb,AAAAA,IAAAA,oBAAAA,iBAAAA,AAAAA,EACvBc,oBAAAA,sCAAsCA;YAKxC,MAAMC,QAAQ3C,AAAAA,IAAAA,oBAAAA,WAAAA,AAAAA,EAAY4C,oBAAAA,2BAA2BA;YACrD,IAAIC;YACJ,IAAIF,OAAO;gBACTG,IAAAA,sBAAAA,MAAAA,AAAAA,EACE,CAACC,sBAAAA,WAAWA,EACZ;gBAEF,MAAMC,aAAa,IAAIC,yBAAAA,sBAAsBA;gBAE7CH,IAAAA,sBAAAA,MAAAA,AAAAA,EAAOH,OAAO;gBACdE,gBAAgBK,AAAAA,IAAAA,yBAAAA,sBAAAA,AAAAA,EAAuBF,YAAYL;gBAEnDjB,SAAS,IAAIa,gCAAAA,WAAWA,CAAC;oBACvB,sBAAsBM;oBACtB,UAAU7C,AAAAA,IAAAA,oBAAAA,WAAAA,AAAAA,EAAYmD,oBAAAA,qBAAqBA;oBAC3C,YAAYnD,AAAAA,IAAAA,oBAAAA,WAAAA,AAAAA,EAAYoD,oBAAAA,wBAAwBA;oBAChD,YAAYpD,AAAAA,IAAAA,oBAAAA,WAAAA,AAAAA,EAAYqD,oBAAAA,uBAAuBA;oBAC/C,GAAG1B,WAAW;oBACd,GAAGc,gBAAgB;gBACrB;YACF,OAEEf,SAAS,IAAIa,gCAAAA,WAAWA,CAAC;gBACvB,QAAQvC,AAAAA,IAAAA,oBAAAA,WAAAA,AAAAA,EAAYsD,oBAAAA,gBAAgBA;gBACpC,UAAUtD,AAAAA,IAAAA,oBAAAA,WAAAA,AAAAA,EAAYmD,oBAAAA,qBAAqBA;gBAC3C,YAAYnD,AAAAA,IAAAA,oBAAAA,WAAAA,AAAAA,EAAYoD,oBAAAA,wBAAwBA;gBAChD,YAAYpD,AAAAA,IAAAA,oBAAAA,WAAAA,AAAAA,EAAYqD,oBAAAA,uBAAuBA;gBAC/C,yBAAyB;gBACzB,GAAG1B,WAAW;gBACd,GAAGc,gBAAgB;YACrB;QAEJ,OAAO,IAAI,CAACzC,AAAAA,IAAAA,oBAAAA,WAAAA,AAAAA,EAAYuD,oBAAAA,0BAA0BA,GAAG;YACnD,MAAMC,UAAUxD,AAAAA,IAAAA,oBAAAA,WAAAA,AAAAA,EAAYwC,oBAAAA,eAAeA;YAC3C,IAAI,AAAmB,YAAnB,OAAOgB,SACT;gBAAA,IAAI,CAAC,eAAe,IAAI,CAACA,UACvB,MAAM,IAAIC,MACR,CAAC,gFAAgF,EAAED,QAAQ,2BAA2B,CAAC;YAE3H;YAGF9B,SAAS,IAAIgC,CAAAA,yBAAAA,EAAO;gBAClB,SAAS1D,AAAAA,IAAAA,oBAAAA,WAAAA,AAAAA,EAAYwC,oBAAAA,eAAeA;gBACpC,QAAQxC,AAAAA,IAAAA,oBAAAA,WAAAA,AAAAA,EAAYC,oBAAAA,cAAcA;gBAClC,WAAWgC;gBACX,GAAGN,WAAW;gBACd,gBAAgB;oBACd,GAAIA,AAAAA,CAAAA,QAAAA,cAAAA,KAAAA,IAAAA,YAAa,cAAc,AAAD,KAAK,CAAC,CAAC;oBACrC,CAACgC,oBAAAA,iBAAiBA,CAAC,EAAElC,kBAAkB,QAAQ;gBACjD;gBACA,yBAAyB;YAC3B;QACF;QAEA,IAAIC,UAAUd,AAAAA,IAAAA,oBAAAA,oBAAAA,AAAAA,EAAqBgD,oBAAAA,wBAAwBA,GAAG;YAC5D,IAAIb,sBAAAA,WAAWA,EACb,MAAM,IAAIU,MAAM;YAElB1C,QAAQ,GAAG,CAAC;YACZ,MAAM,EAAE8C,UAAU,EAAE,GAAG,MAAM;YAC7BnC,SAASmC,WAAWnC;QACtB;QAEA,IAAI,AAAkB,WAAXA,QACT,OAAO;YACL,YAAYA,OAAO,IAAI,CAAC,WAAW;YACnC,OAAO;QACT;QAIF,IAAI1B,AAAAA,IAAAA,oBAAAA,WAAAA,AAAAA,EAAYuD,oBAAAA,0BAA0BA,GAAG;YAC3C,MAAMO,SAAS9D,AAAAA,IAAAA,oBAAAA,WAAAA,AAAAA,EAAYK,oBAAAA,iBAAiBA;YAC5CyC,IAAAA,sBAAAA,MAAAA,AAAAA,EAAOgB,QAAQ;YACfpC,SAAS,IAAIqC,oBAAAA,SAASA,CAAC;gBACrBD;gBACA,WAAW7B;gBACX,yBAAyB;YAC3B;QACF;QAEA,IAAI,AAAkB,WAAXP,UAA2BA,OAAe,QAAQ,EAC3D,OAAO;YACL,YAAaA,OAAe,QAAQ;YACpC,OAAO;QACT;QAGF,MAAM,IAAI+B,MAAM;IAClB;IAEO,eAAeO,KACpBC,QAAsC,EACtCxC,iBAA+B,EAC/ByC,cAEmC,EACnCC,OAGC;QAEDrB,IAAAA,sBAAAA,MAAAA,AAAAA,EACEhD,iBACA;QAGF,MAAM,EAAEsE,UAAU,EAAEC,KAAK,EAAE,GAAG,MAAM7C,iBAAiB;YACnDC;QACF;QAEA,MAAM6C,YAAYtE,AAAAA,IAAAA,oBAAAA,WAAAA,AAAAA,EAAYuE,oBAAAA,iBAAiBA;QAC/C,MAAMC,YAAYrC,AAAAA,IAAAA,uBAAAA,QAAAA,AAAAA,EAAS;QAC3B,MAAMsC,oBAAoBtC,AAAAA,IAAAA,uBAAAA,QAAAA,AAAAA,EAAS;QACnC,MAAMuC,qBAAqBvC,AAAAA,IAAAA,uBAAAA,QAAAA,AAAAA,EAAS;QAEpC,MAAMwC,YAAYC,KAAK,GAAG;QAC1B,MAAMC,QAAQzD;QACd,MAAM0D,cAAcX,AAAAA,CAAAA,QAAAA,UAAAA,KAAAA,IAAAA,QAAS,MAAM,AAAD,KAAKA,CAAAA,QAAAA,UAAAA,KAAAA,IAAAA,QAAS,OAAO,AAAD;QACtD,IAAIY;QACJ,IAAIC,cAAc;QAClB,IAAIC;QACJ,IAAIC;QAEJ,MAAMC,eAAe;YACnB,aAAaC,AAAmB,kBAAnBA,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA,MAAmC,MAAM;YACtD,QAAQ,CAAC,CAACN;YACV,YACE,AAAqB,YAArB,OAAOR,YACHA,YACAe,OAAO,QAAQ,CAACf,aAAa,QAAQ;YAC3C,GAAIc,AAAmB,cAAnBA,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA,MACA;gBACE,2BAA2B;YAC7B,IACA,CAAC,CAAC;QACR;QAEA,IAAI;YACF,IAAIf,AAAU,aAAVA,OAAoB;gBACtBG,UACE,CAAC,QAAQ,EAAEM,cAAc,eAAe,GAAG,WAAW,EAAED,OAAO;gBAGjE,IAAIC,aAAa;oBACf,MAAMQ,SAAU,MAAMlB,WAAW,MAAM,CACrC;wBACES;wBACAZ;wBACA,iBAAiBC;wBACjB,GAAGiB,YAAY;oBACjB,GACA;wBACE,QAAQ;oBACV;oBAKF,WAAW,MAAMI,SAASD,OAAQ;4BAChBE,uBAAAA,iBAAAA,gBAEbC,wBAAAA,kBAAAA,iBAoBCC,kBAAAA;wBAtBJ,MAAMX,UAAUS,AAAAA,SAAAA,CAAAA,iBAAAA,MAAM,OAAO,AAAD,IAAZA,KAAAA,IAAAA,QAAAA,CAAAA,kBAAAA,cAAe,CAAC,EAAE,AAAD,IAAjBA,KAAAA,IAAAA,QAAAA,CAAAA,wBAAAA,gBAAoB,KAAK,AAAD,IAAxBA,KAAAA,IAAAA,sBAA2B,OAAO,AAAD,KAAK;wBACtD,MAAMG,oBACJ,AAAC,SAAAF,CAAAA,kBAAAA,MAAM,OAAO,AAAD,IAAZA,KAAAA,IAAAA,QAAAA,CAAAA,mBAAAA,eAAe,CAAC,EAAE,AAAD,IAAjBA,KAAAA,IAAAA,QAAAA,CAAAA,yBAAAA,iBAAoB,KAAK,AAAD,IAAxBA,KAAAA,IAAAA,uBAAmC,iBAAiB,AAAD,KAAK;wBAG3D,IAAIF,MAAM,KAAK,EACbN,QAAQM,MAAM,KAAK;wBAGrB,IAAIR,WAAWY,mBAAmB;4BAChCX,eAAeD;4BACf,MAAMa,YAAiC;gCACrCb;gCACAY;gCACAX;gCACA,YAAY;gCACZ,OAAOa;4BACT;4BACA1B,QAAQ,OAAO,CAAEyB;wBACnB;wBAGA,IAAI,QAAAF,CAAAA,kBAAAA,MAAM,OAAO,AAAD,IAAZA,KAAAA,IAAAA,QAAAA,CAAAA,mBAAAA,eAAe,CAAC,EAAE,AAAD,IAAjBA,KAAAA,IAAAA,iBAAoB,aAAa,EAAE;4BACrCR,WAAWN,KAAK,GAAG,KAAKD;4BAGxB,IAAI,CAACM,OAAO;gCAEV,MAAMa,kBAAkBC,KAAK,GAAG,CAC9B,GACAA,KAAK,KAAK,CAACf,YAAY,MAAM,GAAG;gCAElCC,QAAQ;oCACN,eAAea;oCACf,mBAAmBA;oCACnB,cAAcA,AAAkB,IAAlBA;gCAChB;4BACF;4BAGA,MAAME,aAAkC;gCACtC,SAAS;gCACThB;gCACA,mBAAmB;gCACnB,YAAY;gCACZ,OAAO;oCACL,eAAeC,MAAM,aAAa,IAAI;oCACtC,mBAAmBA,MAAM,iBAAiB,IAAI;oCAC9C,cAAcA,MAAM,YAAY,IAAI;oCACpC,WAAWC,YAAY;gCACzB;4BACF;4BACAf,QAAQ,OAAO,CAAE6B;4BACjB;wBACF;oBACF;oBACAjB,UAAUC;oBACVP,kBACE,CAAC,iBAAiB,EAAEI,MAAM,QAAQ,EAAEO,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA,OAAkB,UAAU,WAAW,EAAEF,UAAU;gBAE3F,OAAO;wBAUgHe,eAAyDC,gBAAwDC;oBATtO,MAAMC,SAAS,MAAMhC,WAAW,MAAM,CAAC;wBACrCS;wBACAZ;wBACA,iBAAiBC;wBACjB,GAAGiB,YAAY;oBACjB;oBACAD,WAAWN,KAAK,GAAG,KAAKD;oBAExBF,kBACE,CAAC,OAAO,EAAEI,MAAM,QAAQ,EAAEO,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA,OAAkB,UAAU,mBAAmB,EAAEiB,AAAAA,IAAAA,oBAAAA,kBAAAA,AAAAA,IAAqB,iBAAiB,EAAEJ,AAAAA,SAAAA,CAAAA,gBAAAA,OAAO,KAAK,AAAD,IAAXA,KAAAA,IAAAA,cAAc,aAAa,AAAD,KAAK,GAAG,qBAAqB,EAAEC,AAAAA,SAAAA,CAAAA,iBAAAA,OAAO,KAAK,AAAD,IAAXA,KAAAA,IAAAA,eAAc,iBAAiB,AAAD,KAAK,GAAG,gBAAgB,EAAEC,AAAAA,SAAAA,CAAAA,iBAAAA,OAAO,KAAK,AAAD,IAAXA,KAAAA,IAAAA,eAAc,YAAY,AAAD,KAAK,GAAG,WAAW,EAAEjB,SAAS,aAAa,EAAEkB,OAAO,WAAW,IAAI,IAAI;oBAGtU1B,mBACE,CAAC,oBAAoB,EAAE4B,KAAK,SAAS,CAACF,OAAO,KAAK,GAAG;oBAGvDtD,IAAAA,sBAAAA,MAAAA,AAAAA,EACEsD,OAAO,OAAO,EACd,CAAC,mCAAmC,EAAEE,KAAK,SAAS,CAACF,SAAS;oBAEhErB,UAAUqB,OAAO,OAAO,CAAC,EAAE,CAAC,OAAO,CAAC,OAAO;oBAC3CnB,QAAQmB,OAAO,KAAK;gBACtB;gBAEA5B,UAAU,CAAC,UAAU,EAAEO,SAAS;gBAChCjC,IAAAA,sBAAAA,MAAAA,AAAAA,EAAOiC,SAAS;YAClB,OAAO,IAAIV,AAAU,gBAAVA,OAAuB;gBAChC,MAAMkC,sBAAsB,CAACxB;oBAC3B,IAAIA,AAAiB,gBAAjBA,QAAQ,IAAI,EAAkB;wBAChC,MAAMyB,YAAYzB,QAAQ,SAAS,CAAC,GAAG;wBACvCjC,IAAAA,sBAAAA,MAAAA,AAAAA,EAAO0D,WAAW;wBAClB,MAAM,EAAEC,QAAQ,EAAEC,IAAI,EAAE,GAAGC,AAAAA,IAAAA,oBAAAA,WAAAA,AAAAA,EAAY5B,QAAQ,SAAS,CAAC,GAAG;wBAC5D,OAAO;4BACL,QAAQ;gCACN,MAAM;gCACN,YAAY0B;gCACZ,MAAMC;4BACR;4BACA,MAAM;wBACR;oBACF;oBACA,OAAO3B;gBACT;gBAEA,IAAID,aAAa;oBACf,MAAMQ,SAAU,MAAMlB,WAAW,MAAM,CAAC;wBACtCS;wBACA,QAAQ;wBACR,UAAUZ,SAAS,GAAG,CAAC,CAAC2C,IAAO;gCAC7B,MAAM;gCACN,SAASC,MAAM,OAAO,CAACD,EAAE,OAAO,IAC3BA,EAAE,OAAO,CAAS,GAAG,CAACL,uBACvBK,EAAE,OAAO;4BACf;wBACA,iBAAiB1C;wBACjB,GAAGiB,YAAY;oBACjB;oBAEA,WAAW,MAAMI,SAASD,OAAQ;4BAChBwB;wBAAhB,MAAM/B,UAAU+B,AAAAA,SAAAA,CAAAA,eAAAA,MAAM,KAAK,AAAD,IAAVA,KAAAA,IAAAA,aAAa,IAAI,AAAD,KAAK;wBACrC,IAAI/B,SAAS;4BACXC,eAAeD;4BACf,MAAMa,YAAiC;gCACrCb;gCACAC;gCACA,mBAAmB;gCACnB,YAAY;gCACZ,OAAOa;4BACT;4BACA1B,QAAQ,OAAO,CAAEyB;wBACnB;wBAGA,IAAIL,AAAe,mBAAfA,MAAM,IAAI,EAAqB;4BACjCL,WAAWN,KAAK,GAAG,KAAKD;4BACxB,MAAMoC,iBAAiBxB,MAAM,KAAK;4BAGlC,MAAMS,aAAkC;gCACtC,SAAS;gCACThB;gCACA,mBAAmB;gCACnB,YAAY;gCACZ,OAAO+B,iBACH;oCACE,eAAeA,eAAe,YAAY,IAAI;oCAC9C,mBAAmBA,eAAe,aAAa,IAAI;oCACnD,cACGA,AAAAA,CAAAA,eAAe,YAAY,IAAI,KAC/BA,CAAAA,eAAe,aAAa,IAAI;oCACnC,WAAW7B,YAAY;gCACzB,IACAW;4BACN;4BACA1B,QAAQ,OAAO,CAAE6B;4BACjB;wBACF;oBACF;oBACAjB,UAAUC;gBACZ,OAAO;oBACL,MAAMoB,SAAS,MAAMhC,WAAW,MAAM,CAAC;wBACrCS;wBACA,QAAQ;wBACR,UAAUZ,SAAS,GAAG,CAAC,CAAC2C,IAAO;gCAC7B,MAAM;gCACN,SAASC,MAAM,OAAO,CAACD,EAAE,OAAO,IAC3BA,EAAE,OAAO,CAAS,GAAG,CAACL,uBACvBK,EAAE,OAAO;4BACf;wBACA,iBAAiB1C;wBACjB,GAAGiB,YAAY;oBACjB;oBACAD,WAAWN,KAAK,GAAG,KAAKD;oBACxBI,UAAWqB,OAAe,OAAO,CAAC,EAAE,CAAC,IAAI;oBACzCnB,QAAQmB,OAAO,KAAK;gBACtB;gBAEAtD,IAAAA,sBAAAA,MAAAA,AAAAA,EAAOiC,SAAS;YAClB;YAEA,IAAID,eAAe,CAACG,OAAO;gBAEzB,MAAMa,kBAAkBC,KAAK,GAAG,CAC9B,GACAA,KAAK,KAAK,CAAEhB,AAAAA,CAAAA,WAAW,EAAC,EAAG,MAAM,GAAG;gBAEtCE,QAAQ;oBACN,eAAea;oBACf,mBAAmBA;oBACnB,cAAcA,AAAkB,IAAlBA;gBAChB;YACF;YAEA,OAAO;gBACL,SAASf,WAAW;gBACpB,OAAOE,QACH;oBACE,eAAeA,MAAM,aAAa,IAAI;oBACtC,mBAAmBA,MAAM,iBAAiB,IAAI;oBAC9C,cAAcA,MAAM,YAAY,IAAI;oBACpC,WAAWC,YAAY;gBACzB,IACAW;gBACJ,YAAY,CAAC,CAACf;YAChB;QACF,EAAE,OAAOkC,GAAQ;YACfjG,QAAQ,KAAK,CAAC,kBAAkBiG;YAChC,MAAMC,WAAW,IAAIxD,MACnB,CAAC,eAAe,EAAEqB,cAAc,eAAe,GAAG,kBAAkB,EAAEkC,EAAE,OAAO,CAAC,8DAA8D,CAAC,EAC/I;gBACE,OAAOA;YACT;YAEF,MAAMC;QACR;IACF;IAEO,eAAeC,oBACpBjD,QAAsC,EACtCxC,iBAA+B;QAE/B,IAAIyC;QAKJ,MAAMW,QAAQzD;QAEd,IAAIyD,MAAM,QAAQ,CAAC,UACjB,OAAQpD;YACN,KAAK0F,mCAAAA,YAAAA,CAAAA,MAAmB;gBACtBjD,iBAAiBkD,6BAAAA,YAAYA;gBAC7B;YACF,KAAKD,mCAAAA,YAAAA,CAAAA,eAA4B;gBAC/BjD,iBAAiBmD,+BAAAA,aAAaA;gBAC9B;YACF,KAAKF,mCAAAA,YAAAA,CAAAA,IAAiB;gBACpBjD,iBAAiBoD,gCAAAA,UAAUA;gBAC3B;YACF,KAAKH,mCAAAA,YAAAA,CAAAA,YAAyB;YAC9B,KAAKA,mCAAAA,YAAAA,CAAAA,gBAA6B;gBAChCjD,iBAAiB;oBAAE,MAAMqD,kCAAAA,gBAAAA,CAAAA,IAAqB;gBAAC;gBAC/C;QACJ;QAIF,IAAI1C,AAAU,wBAAVA,OACFX,iBAAiB;YAAE,MAAMqD,kCAAAA,gBAAAA,CAAAA,IAAqB;QAAC;QAGjD,MAAMC,WAAW,MAAMxD,KAAKC,UAAUxC,mBAAmByC;QACzDpB,IAAAA,sBAAAA,MAAAA,AAAAA,EAAO0E,UAAU;QACjB,MAAMC,cAAcC,cAAcF,SAAS,OAAO;QAClD,OAAO;YAAE,SAASC;YAAa,OAAOD,SAAS,KAAK;QAAC;IACvD;IAEO,eAAeG,2BACpBC,IAAY,EACZnG,iBAA+B;QAE/B,MAAM,EAAEsD,OAAO,EAAEE,KAAK,EAAE,GAAG,MAAMjB,KAAK4D,MAAMnG;QAC5C,OAAO;YAAEsD;YAASE;QAAM;IAC1B;IAEO,SAAS4C,yBAAyBL,QAAgB;QACvD,IAAI;YAEF,MAAMM,YAAYN,SAAS,KAAK,CAAC;YACjC,IAAIM,WACF,OAAOA,SAAS,CAAC,EAAE;YAIrB,MAAMC,iBAAiBP,SAAS,KAAK,CACnC;YAEF,IAAIO,gBACF,OAAOA,cAAc,CAAC,EAAE;YAI1B,MAAMC,gBAAgBR,SAAS,KAAK,CAAC;YACrC,IAAIQ,eACF,OAAOA,aAAa,CAAC,EAAE;QAE3B,EAAE,OAAM,CAAC;QAET,OAAOR;IACT;IAEO,SAASS,yBAAyBC,KAAa;QACpD,IAAIA,MAAM,QAAQ,CAAC,SAEjB,MAAO,YAAY,IAAI,CAACA,OACtBA,QAAQA,MAAM,OAAO,CAAC,kBAAkB;QAG5C,OAAOA;IACT;IAEO,SAASR,cAAcQ,KAAa;QACzC,MAAMC,kBAAkBN,yBAAyBK;QAEjD,IAAIC,QAAAA,kBAAAA,KAAAA,IAAAA,gBAAiB,KAAK,CAAC,oBAAoB;gBACtCC;YAAP,OAAO,QAAAA,CAAAA,yBAAAA,gBACJ,KAAK,CAAC,kBAAiB,IADnBA,KAAAA,IAAAA,uBAEH,KAAK,CAAC,GACP,GAAG,CAAC/C;QACT;QACA,IAAI;YACF,OAAOiB,KAAK,KAAK,CAAC6B;QACpB,EAAE,OAAM,CAAC;QACT,IAAI;YACF,OAAO7B,KAAK,KAAK,CAAC+B,AAAAA,IAAAA,oCAAAA,UAAAA,AAAAA,EAAWF;QAC/B,EAAE,OAAOnB,GAAG,CAAC;QAEb,IAAI5B,AAAmB,oBAAnBA,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA,OAAsCA,AAAmB,kBAAnBA,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA,KAAkC;YAC1E,MAAMkD,aAAaL,yBAAyBE;YAC5C,OAAO7B,KAAK,KAAK,CAAC+B,AAAAA,IAAAA,oCAAAA,UAAAA,AAAAA,EAAWC;QAC/B;QACA,MAAM7E,MAAM,CAAC,+BAA+B,EAAEyE,OAAO;IACvD"}
+{"version":3,"file":"ai-model/service-caller/index.js","sources":["webpack://@midscene/core/webpack/runtime/compat_get_default_export","webpack://@midscene/core/webpack/runtime/define_property_getters","webpack://@midscene/core/webpack/runtime/has_own_property","webpack://@midscene/core/webpack/runtime/make_namespace_object","webpack://@midscene/core/./src/ai-model/service-caller/index.ts"],"sourcesContent":["// getDefaultExport function for compatibility with non-ESM modules\n__webpack_require__.n = (module) => {\n\tvar getter = module && module.__esModule ?\n\t\t() => (module['default']) :\n\t\t() => (module);\n\t__webpack_require__.d(getter, { a: getter });\n\treturn getter;\n};\n","__webpack_require__.d = (exports, definition) => {\n\tfor(var key in definition) {\n        if(__webpack_require__.o(definition, key) && !__webpack_require__.o(exports, key)) {\n            Object.defineProperty(exports, key, { enumerable: true, get: definition[key] });\n        }\n    }\n};","__webpack_require__.o = (obj, prop) => (Object.prototype.hasOwnProperty.call(obj, prop))","// define __esModule on exports\n__webpack_require__.r = (exports) => {\n\tif(typeof Symbol !== 'undefined' && Symbol.toStringTag) {\n\t\tObject.defineProperty(exports, Symbol.toStringTag, { value: 'Module' });\n\t}\n\tObject.defineProperty(exports, '__esModule', { value: true });\n};","import { AIResponseFormat, type AIUsageInfo } from '@/types';\nimport type { CodeGenerationChunk, StreamingCallback } from '@/types';\nimport { Anthropic } from '@anthropic-ai/sdk';\nimport {\n  DefaultAzureCredential,\n  getBearerTokenProvider,\n} from '@azure/identity';\nimport {\n  type IModelPreferences,\n  MIDSCENE_API_TYPE,\n  MIDSCENE_LANGSMITH_DEBUG,\n  OPENAI_MAX_TOKENS,\n  decideModelConfig,\n  getAIConfig,\n  getAIConfigInBoolean,\n  uiTarsModelVersion,\n  vlLocateMode,\n} from '@midscene/shared/env';\nimport { parseBase64 } from '@midscene/shared/img';\nimport { getDebug } from '@midscene/shared/logger';\nimport { assert } from '@midscene/shared/utils';\nimport { ifInBrowser } from '@midscene/shared/utils';\nimport { HttpsProxyAgent } from 'https-proxy-agent';\nimport { jsonrepair } from 'jsonrepair';\nimport OpenAI, { AzureOpenAI } from 'openai';\nimport type { ChatCompletionMessageParam } from 'openai/resources/index';\nimport type { Stream } from 'openai/streaming';\nimport { SocksProxyAgent } from 'socks-proxy-agent';\nimport { AIActionType, type AIArgs } from '../common';\nimport { assertSchema } from '../prompt/assertion';\nimport { locatorSchema } from '../prompt/llm-locator';\nimport { planSchema } from '../prompt/llm-planning';\n\nasync function createChatClient({\n  AIActionTypeValue,\n  modelPreferences,\n}: {\n  AIActionTypeValue: AIActionType;\n  modelPreferences: IModelPreferences;\n}): Promise<{\n  completion: OpenAI.Chat.Completions;\n  style: 'openai' | 'anthropic';\n  modelName: string;\n}> {\n  const {\n    socksProxy,\n    httpProxy,\n    modelName,\n    openaiBaseURL,\n    openaiApiKey,\n    openaiExtraConfig,\n    openaiUseAzureDeprecated,\n    useAzureOpenai,\n    azureOpenaiScope,\n    azureOpenaiKey,\n    azureOpenaiEndpoint,\n    azureOpenaiApiVersion,\n    azureOpenaiDeployment,\n    azureExtraConfig,\n    useAnthropicSdk,\n    anthropicApiKey,\n  } = decideModelConfig(modelPreferences, true);\n\n  let openai: OpenAI | AzureOpenAI | undefined;\n\n  let proxyAgent = undefined;\n  const debugProxy = getDebug('ai:call:proxy');\n  if (httpProxy) {\n    debugProxy('using http proxy', httpProxy);\n    proxyAgent = new HttpsProxyAgent(httpProxy);\n  } else if (socksProxy) {\n    debugProxy('using socks proxy', socksProxy);\n    proxyAgent = new SocksProxyAgent(socksProxy);\n  }\n\n  if (openaiUseAzureDeprecated) {\n    // this is deprecated\n    openai = new AzureOpenAI({\n      baseURL: openaiBaseURL,\n      apiKey: openaiApiKey,\n      httpAgent: proxyAgent,\n      ...openaiExtraConfig,\n      dangerouslyAllowBrowser: true,\n    }) as OpenAI;\n  } else if (useAzureOpenai) {\n    // https://learn.microsoft.com/en-us/azure/ai-services/openai/chatgpt-quickstart?tabs=bash%2Cjavascript-key%2Ctypescript-keyless%2Cpython&pivots=programming-language-javascript#rest-api\n    // keyless authentication\n    let tokenProvider: any = undefined;\n    if (azureOpenaiScope) {\n      assert(\n        !ifInBrowser,\n        'Azure OpenAI is not supported in browser with Midscene.',\n      );\n      const credential = new DefaultAzureCredential();\n\n      tokenProvider = getBearerTokenProvider(credential, azureOpenaiScope);\n\n      openai = new AzureOpenAI({\n        azureADTokenProvider: tokenProvider,\n        endpoint: azureOpenaiEndpoint,\n        apiVersion: azureOpenaiApiVersion,\n        deployment: azureOpenaiDeployment,\n        ...openaiExtraConfig,\n        ...azureExtraConfig,\n      });\n    } else {\n      // endpoint, apiKey, apiVersion, deployment\n      openai = new AzureOpenAI({\n        apiKey: azureOpenaiKey,\n        endpoint: azureOpenaiEndpoint,\n        apiVersion: azureOpenaiApiVersion,\n        deployment: azureOpenaiDeployment,\n        dangerouslyAllowBrowser: true,\n        ...openaiExtraConfig,\n        ...azureExtraConfig,\n      });\n    }\n  } else if (!useAnthropicSdk) {\n    openai = new OpenAI({\n      baseURL: openaiBaseURL,\n      apiKey: openaiApiKey,\n      httpAgent: proxyAgent,\n      ...openaiExtraConfig,\n      defaultHeaders: {\n        ...(openaiExtraConfig?.defaultHeaders || {}),\n        [MIDSCENE_API_TYPE]: AIActionTypeValue.toString(),\n      },\n      dangerouslyAllowBrowser: true,\n    });\n  }\n\n  if (openai && getAIConfigInBoolean(MIDSCENE_LANGSMITH_DEBUG)) {\n    if (ifInBrowser) {\n      throw new Error('langsmith is not supported in browser');\n    }\n    console.log('DEBUGGING MODE: langsmith wrapper enabled');\n    const { wrapOpenAI } = await import('langsmith/wrappers');\n    openai = wrapOpenAI(openai);\n  }\n\n  if (typeof openai !== 'undefined') {\n    return {\n      completion: openai.chat.completions,\n      style: 'openai',\n      modelName,\n    };\n  }\n\n  // Anthropic\n  if (useAnthropicSdk) {\n    openai = new Anthropic({\n      apiKey: anthropicApiKey,\n      httpAgent: proxyAgent,\n      dangerouslyAllowBrowser: true,\n    }) as any;\n  }\n\n  if (typeof openai !== 'undefined' && (openai as any).messages) {\n    return {\n      completion: (openai as any).messages,\n      style: 'anthropic',\n      modelName,\n    };\n  }\n\n  throw new Error('Openai SDK or Anthropic SDK is not initialized');\n}\n\nexport async function call(\n  messages: ChatCompletionMessageParam[],\n  AIActionTypeValue: AIActionType,\n  modelPreferences: IModelPreferences,\n  options?: {\n    stream?: boolean;\n    onChunk?: StreamingCallback;\n  },\n): Promise<{ content: string; usage?: AIUsageInfo; isStreamed: boolean }> {\n  const { completion, style, modelName } = await createChatClient({\n    AIActionTypeValue,\n    modelPreferences,\n  });\n\n  const responseFormat = getResponseFormat(modelName, AIActionTypeValue);\n\n  const maxTokens = getAIConfig(OPENAI_MAX_TOKENS);\n  const debugCall = getDebug('ai:call');\n  const debugProfileStats = getDebug('ai:profile:stats');\n  const debugProfileDetail = getDebug('ai:profile:detail');\n\n  const startTime = Date.now();\n\n  const isStreaming = options?.stream && options?.onChunk;\n  let content: string | undefined;\n  let accumulated = '';\n  let usage: OpenAI.CompletionUsage | undefined;\n  let timeCost: number | undefined;\n\n  const commonConfig = {\n    temperature: vlLocateMode(modelPreferences) === 'vlm-ui-tars' ? 0.0 : 0.1,\n    stream: !!isStreaming,\n    max_tokens:\n      typeof maxTokens === 'number'\n        ? maxTokens\n        : Number.parseInt(maxTokens || '2048', 10),\n    ...(vlLocateMode(modelPreferences) === 'qwen-vl' // qwen specific config\n      ? {\n          vl_high_resolution_images: true,\n        }\n      : {}),\n  };\n\n  try {\n    if (style === 'openai') {\n      debugCall(\n        `sending ${isStreaming ? 'streaming ' : ''}request to ${modelName}`,\n      );\n\n      if (isStreaming) {\n        const stream = (await completion.create(\n          {\n            model: modelName,\n            messages,\n            response_format: responseFormat,\n            ...commonConfig,\n          },\n          {\n            stream: true,\n          },\n        )) as Stream<OpenAI.Chat.Completions.ChatCompletionChunk> & {\n          _request_id?: string | null;\n        };\n\n        for await (const chunk of stream) {\n          const content = chunk.choices?.[0]?.delta?.content || '';\n          const reasoning_content =\n            (chunk.choices?.[0]?.delta as any)?.reasoning_content || '';\n\n          // Check for usage info in any chunk (OpenAI provides usage in separate chunks)\n          if (chunk.usage) {\n            usage = chunk.usage;\n          }\n\n          if (content || reasoning_content) {\n            accumulated += content;\n            const chunkData: CodeGenerationChunk = {\n              content,\n              reasoning_content,\n              accumulated,\n              isComplete: false,\n              usage: undefined,\n            };\n            options.onChunk!(chunkData);\n          }\n\n          // Check if stream is complete\n          if (chunk.choices?.[0]?.finish_reason) {\n            timeCost = Date.now() - startTime;\n\n            // If usage is not available from the stream, provide a basic usage info\n            if (!usage) {\n              // Estimate token counts based on content length (rough approximation)\n              const estimatedTokens = Math.max(\n                1,\n                Math.floor(accumulated.length / 4),\n              );\n              usage = {\n                prompt_tokens: estimatedTokens,\n                completion_tokens: estimatedTokens,\n                total_tokens: estimatedTokens * 2,\n              };\n            }\n\n            // Send final chunk\n            const finalChunk: CodeGenerationChunk = {\n              content: '',\n              accumulated,\n              reasoning_content: '',\n              isComplete: true,\n              usage: {\n                prompt_tokens: usage.prompt_tokens ?? 0,\n                completion_tokens: usage.completion_tokens ?? 0,\n                total_tokens: usage.total_tokens ?? 0,\n                time_cost: timeCost ?? 0,\n                model_name: modelName,\n              },\n            };\n            options.onChunk!(finalChunk);\n            break;\n          }\n        }\n        content = accumulated;\n        debugProfileStats(\n          `streaming model, ${modelName}, mode, ${vlLocateMode(modelPreferences) || 'default'}, cost-ms, ${timeCost}`,\n        );\n      } else {\n        const result = await completion.create({\n          model: modelName,\n          messages,\n          response_format: responseFormat,\n          ...commonConfig,\n        } as any);\n        timeCost = Date.now() - startTime;\n\n        debugProfileStats(\n          `model, ${modelName}, mode, ${vlLocateMode(modelPreferences) || 'default'}, ui-tars-version, ${uiTarsModelVersion(modelPreferences)}, prompt-tokens, ${result.usage?.prompt_tokens || ''}, completion-tokens, ${result.usage?.completion_tokens || ''}, total-tokens, ${result.usage?.total_tokens || ''}, cost-ms, ${timeCost}, requestId, ${result._request_id || ''}`,\n        );\n\n        debugProfileDetail(\n          `model usage detail: ${JSON.stringify(result.usage)}`,\n        );\n\n        assert(\n          result.choices,\n          `invalid response from LLM service: ${JSON.stringify(result)}`,\n        );\n        content = result.choices[0].message.content!;\n        usage = result.usage;\n      }\n\n      debugCall(`response: ${content}`);\n      assert(content, 'empty content');\n    } else if (style === 'anthropic') {\n      const convertImageContent = (content: any) => {\n        if (content.type === 'image_url') {\n          const imgBase64 = content.image_url.url;\n          assert(imgBase64, 'image_url is required');\n          const { mimeType, body } = parseBase64(content.image_url.url);\n          return {\n            source: {\n              type: 'base64',\n              media_type: mimeType,\n              data: body,\n            },\n            type: 'image',\n          };\n        }\n        return content;\n      };\n\n      if (isStreaming) {\n        const stream = (await completion.create({\n          model: modelName,\n          system: 'You are a versatile professional in software UI automation',\n          messages: messages.map((m) => ({\n            role: 'user',\n            content: Array.isArray(m.content)\n              ? (m.content as any).map(convertImageContent)\n              : m.content,\n          })),\n          response_format: responseFormat,\n          ...commonConfig,\n        } as any)) as any;\n\n        for await (const chunk of stream) {\n          const content = chunk.delta?.text || '';\n          if (content) {\n            accumulated += content;\n            const chunkData: CodeGenerationChunk = {\n              content,\n              accumulated,\n              reasoning_content: '',\n              isComplete: false,\n              usage: undefined,\n            };\n            options.onChunk!(chunkData);\n          }\n\n          // Check if stream is complete\n          if (chunk.type === 'message_stop') {\n            timeCost = Date.now() - startTime;\n            const anthropicUsage = chunk.usage;\n\n            // Send final chunk\n            const finalChunk: CodeGenerationChunk = {\n              content: '',\n              accumulated,\n              reasoning_content: '',\n              isComplete: true,\n              usage: anthropicUsage\n                ? {\n                    prompt_tokens: anthropicUsage.input_tokens ?? 0,\n                    completion_tokens: anthropicUsage.output_tokens ?? 0,\n                    total_tokens:\n                      (anthropicUsage.input_tokens ?? 0) +\n                      (anthropicUsage.output_tokens ?? 0),\n                    time_cost: timeCost ?? 0,\n                    model_name: modelName,\n                  }\n                : undefined,\n            };\n            options.onChunk!(finalChunk);\n            break;\n          }\n        }\n        content = accumulated;\n      } else {\n        const result = await completion.create({\n          model: modelName,\n          system: 'You are a versatile professional in software UI automation',\n          messages: messages.map((m) => ({\n            role: 'user',\n            content: Array.isArray(m.content)\n              ? (m.content as any).map(convertImageContent)\n              : m.content,\n          })),\n          response_format: responseFormat,\n          ...commonConfig,\n        } as any);\n        timeCost = Date.now() - startTime;\n        content = (result as any).content[0].text as string;\n        usage = result.usage;\n      }\n\n      assert(content, 'empty content');\n    }\n    // Ensure we always have usage info for streaming responses\n    if (isStreaming && !usage) {\n      // Estimate token counts based on content length (rough approximation)\n      const estimatedTokens = Math.max(\n        1,\n        Math.floor((content || '').length / 4),\n      );\n      usage = {\n        prompt_tokens: estimatedTokens,\n        completion_tokens: estimatedTokens,\n        total_tokens: estimatedTokens * 2,\n      };\n    }\n\n    return {\n      content: content || '',\n      usage: usage\n        ? {\n            prompt_tokens: usage.prompt_tokens ?? 0,\n            completion_tokens: usage.completion_tokens ?? 0,\n            total_tokens: usage.total_tokens ?? 0,\n            time_cost: timeCost ?? 0,\n            model_name: modelName,\n          }\n        : undefined,\n      isStreamed: !!isStreaming,\n    };\n  } catch (e: any) {\n    console.error(' call AI error', e);\n    const newError = new Error(\n      `failed to call ${isStreaming ? 'streaming ' : ''}AI model service: ${e.message}. Trouble shooting: https://midscenejs.com/model-provider.html`,\n      {\n        cause: e,\n      },\n    );\n    throw newError;\n  }\n}\n\nexport const getResponseFormat = (\n  modelName: string,\n  AIActionTypeValue: AIActionType,\n):\n  | OpenAI.ChatCompletionCreateParams['response_format']\n  | OpenAI.ResponseFormatJSONObject => {\n  let responseFormat:\n    | OpenAI.ChatCompletionCreateParams['response_format']\n    | OpenAI.ResponseFormatJSONObject\n    | undefined;\n\n  if (modelName.includes('gpt-4')) {\n    switch (AIActionTypeValue) {\n      case AIActionType.ASSERT:\n        responseFormat = assertSchema;\n        break;\n      case AIActionType.INSPECT_ELEMENT:\n        responseFormat = locatorSchema;\n        break;\n      case AIActionType.PLAN:\n        responseFormat = planSchema;\n        break;\n      case AIActionType.EXTRACT_DATA:\n      case AIActionType.DESCRIBE_ELEMENT:\n        responseFormat = { type: AIResponseFormat.JSON };\n        break;\n    }\n  }\n\n  // gpt-4o-2024-05-13 only supports json_object response format\n  if (modelName === 'gpt-4o-2024-05-13') {\n    responseFormat = { type: AIResponseFormat.JSON };\n  }\n\n  return responseFormat;\n};\n\nexport async function callToGetJSONObject<T>(\n  messages: ChatCompletionMessageParam[],\n  AIActionTypeValue: AIActionType,\n  modelPreferences: IModelPreferences,\n): Promise<{ content: T; usage?: AIUsageInfo }> {\n  const response = await call(messages, AIActionTypeValue, modelPreferences);\n  assert(response, 'empty response');\n  const jsonContent = safeParseJson(response.content, modelPreferences);\n  return { content: jsonContent, usage: response.usage };\n}\n\nexport async function callAiFnWithStringResponse<T>(\n  msgs: AIArgs,\n  AIActionTypeValue: AIActionType,\n  modelPreferences: IModelPreferences,\n): Promise<{ content: string; usage?: AIUsageInfo }> {\n  const { content, usage } = await call(\n    msgs,\n    AIActionTypeValue,\n    modelPreferences,\n  );\n  return { content, usage };\n}\n\nexport function extractJSONFromCodeBlock(response: string) {\n  try {\n    // First, try to match a JSON object directly in the response\n    const jsonMatch = response.match(/^\\s*(\\{[\\s\\S]*\\})\\s*$/);\n    if (jsonMatch) {\n      return jsonMatch[1];\n    }\n\n    // If no direct JSON object is found, try to extract JSON from a code block\n    const codeBlockMatch = response.match(\n      /```(?:json)?\\s*(\\{[\\s\\S]*?\\})\\s*```/,\n    );\n    if (codeBlockMatch) {\n      return codeBlockMatch[1];\n    }\n\n    // If no code block is found, try to find a JSON-like structure in the text\n    const jsonLikeMatch = response.match(/\\{[\\s\\S]*\\}/);\n    if (jsonLikeMatch) {\n      return jsonLikeMatch[0];\n    }\n  } catch {}\n  // If no JSON-like structure is found, return the original response\n  return response;\n}\n\nexport function preprocessDoubaoBboxJson(input: string) {\n  if (input.includes('bbox')) {\n    // when its values like 940 445 969 490, replace all /\\d+\\s+\\d+/g with /$1,$2/g\n    while (/\\d+\\s+\\d+/.test(input)) {\n      input = input.replace(/(\\d+)\\s+(\\d+)/g, '$1,$2');\n    }\n  }\n  return input;\n}\n\nexport function safeParseJson(\n  input: string,\n  modelPreferences: IModelPreferences,\n) {\n  const cleanJsonString = extractJSONFromCodeBlock(input);\n  // match the point\n  if (cleanJsonString?.match(/\\((\\d+),(\\d+)\\)/)) {\n    return cleanJsonString\n      .match(/\\((\\d+),(\\d+)\\)/)\n      ?.slice(1)\n      .map(Number);\n  }\n  try {\n    return JSON.parse(cleanJsonString);\n  } catch {}\n  try {\n    return JSON.parse(jsonrepair(cleanJsonString));\n  } catch (e) {}\n\n  if (\n    vlLocateMode(modelPreferences) === 'doubao-vision' ||\n    vlLocateMode(modelPreferences) === 'vlm-ui-tars'\n  ) {\n    const jsonString = preprocessDoubaoBboxJson(cleanJsonString);\n    return JSON.parse(jsonrepair(jsonString));\n  }\n  throw Error(`failed to parse json response: ${input}`);\n}\n"],"names":["__webpack_require__","module","getter","definition","key","Object","obj","prop","Symbol","createChatClient","AIActionTypeValue","modelPreferences","socksProxy","httpProxy","modelName","openaiBaseURL","openaiApiKey","openaiExtraConfig","openaiUseAzureDeprecated","useAzureOpenai","azureOpenaiScope","azureOpenaiKey","azureOpenaiEndpoint","azureOpenaiApiVersion","azureOpenaiDeployment","azureExtraConfig","useAnthropicSdk","anthropicApiKey","decideModelConfig","openai","proxyAgent","debugProxy","getDebug","HttpsProxyAgent","SocksProxyAgent","AzureOpenAI","tokenProvider","assert","ifInBrowser","credential","DefaultAzureCredential","getBearerTokenProvider","OpenAI","MIDSCENE_API_TYPE","getAIConfigInBoolean","MIDSCENE_LANGSMITH_DEBUG","Error","console","wrapOpenAI","Anthropic","call","messages","options","completion","style","responseFormat","getResponseFormat","maxTokens","getAIConfig","OPENAI_MAX_TOKENS","debugCall","debugProfileStats","debugProfileDetail","startTime","Date","isStreaming","content","accumulated","usage","timeCost","commonConfig","vlLocateMode","Number","stream","chunk","_chunk_choices__delta","_chunk_choices__delta1","_chunk_choices_2","reasoning_content","chunkData","undefined","estimatedTokens","Math","finalChunk","_result_usage","_result_usage1","_result_usage2","result","uiTarsModelVersion","JSON","convertImageContent","imgBase64","mimeType","body","parseBase64","m","Array","_chunk_delta","anthropicUsage","e","newError","AIActionType","assertSchema","locatorSchema","planSchema","AIResponseFormat","callToGetJSONObject","response","jsonContent","safeParseJson","callAiFnWithStringResponse","msgs","extractJSONFromCodeBlock","jsonMatch","codeBlockMatch","jsonLikeMatch","preprocessDoubaoBboxJson","input","cleanJsonString","_cleanJsonString_match","jsonrepair","jsonString"],"mappings":";;;;;;;;;;;;;;;;;;;IACAA,oBAAoB,CAAC,GAAG,CAACC;QACxB,IAAIC,SAASD,UAAUA,OAAO,UAAU,GACvC,IAAOA,MAAM,CAAC,UAAU,GACxB,IAAOA;QACRD,oBAAoB,CAAC,CAACE,QAAQ;YAAE,GAAGA;QAAO;QAC1C,OAAOA;IACR;;;ICPAF,oBAAoB,CAAC,GAAG,CAAC,UAASG;QACjC,IAAI,IAAIC,OAAOD,WACR,IAAGH,oBAAoB,CAAC,CAACG,YAAYC,QAAQ,CAACJ,oBAAoB,CAAC,CAAC,UAASI,MACzEC,OAAO,cAAc,CAAC,UAASD,KAAK;YAAE,YAAY;YAAM,KAAKD,UAAU,CAACC,IAAI;QAAC;IAGzF;;;ICNAJ,oBAAoB,CAAC,GAAG,CAACM,KAAKC,OAAUF,OAAO,SAAS,CAAC,cAAc,CAAC,IAAI,CAACC,KAAKC;;;ICClFP,oBAAoB,CAAC,GAAG,CAAC;QACxB,IAAG,AAAkB,eAAlB,OAAOQ,UAA0BA,OAAO,WAAW,EACrDH,OAAO,cAAc,CAAC,UAASG,OAAO,WAAW,EAAE;YAAE,OAAO;QAAS;QAEtEH,OAAO,cAAc,CAAC,UAAS,cAAc;YAAE,OAAO;QAAK;IAC5D;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;IC2BA,eAAeI,iBAAiB,EAC9BC,iBAAiB,EACjBC,gBAAgB,EAIjB;QAKC,MAAM,EACJC,UAAU,EACVC,SAAS,EACTC,SAAS,EACTC,aAAa,EACbC,YAAY,EACZC,iBAAiB,EACjBC,wBAAwB,EACxBC,cAAc,EACdC,gBAAgB,EAChBC,cAAc,EACdC,mBAAmB,EACnBC,qBAAqB,EACrBC,qBAAqB,EACrBC,gBAAgB,EAChBC,eAAe,EACfC,eAAe,EAChB,GAAGC,AAAAA,IAAAA,oBAAAA,iBAAAA,AAAAA,EAAkBjB,kBAAkB;QAExC,IAAIkB;QAEJ,IAAIC;QACJ,MAAMC,aAAaC,AAAAA,IAAAA,uBAAAA,QAAAA,AAAAA,EAAS;QAC5B,IAAInB,WAAW;YACbkB,WAAW,oBAAoBlB;YAC/BiB,aAAa,IAAIG,2CAAAA,eAAeA,CAACpB;QACnC,OAAO,IAAID,YAAY;YACrBmB,WAAW,qBAAqBnB;YAChCkB,aAAa,IAAII,2CAAAA,eAAeA,CAACtB;QACnC;QAEA,IAAIM,0BAEFW,SAAS,IAAIM,gCAAAA,WAAWA,CAAC;YACvB,SAASpB;YACT,QAAQC;YACR,WAAWc;YACX,GAAGb,iBAAiB;YACpB,yBAAyB;QAC3B;aACK,IAAIE,gBAAgB;YAGzB,IAAIiB;YACJ,IAAIhB,kBAAkB;gBACpBiB,IAAAA,sBAAAA,MAAAA,AAAAA,EACE,CAACC,sBAAAA,WAAWA,EACZ;gBAEF,MAAMC,aAAa,IAAIC,yBAAAA,sBAAsBA;gBAE7CJ,gBAAgBK,AAAAA,IAAAA,yBAAAA,sBAAAA,AAAAA,EAAuBF,YAAYnB;gBAEnDS,SAAS,IAAIM,gCAAAA,WAAWA,CAAC;oBACvB,sBAAsBC;oBACtB,UAAUd;oBACV,YAAYC;oBACZ,YAAYC;oBACZ,GAAGP,iBAAiB;oBACpB,GAAGQ,gBAAgB;gBACrB;YACF,OAEEI,SAAS,IAAIM,gCAAAA,WAAWA,CAAC;gBACvB,QAAQd;gBACR,UAAUC;gBACV,YAAYC;gBACZ,YAAYC;gBACZ,yBAAyB;gBACzB,GAAGP,iBAAiB;gBACpB,GAAGQ,gBAAgB;YACrB;QAEJ,OAAO,IAAI,CAACC,iBACVG,SAAS,IAAIa,CAAAA,yBAAAA,EAAO;YAClB,SAAS3B;YACT,QAAQC;YACR,WAAWc;YACX,GAAGb,iBAAiB;YACpB,gBAAgB;gBACd,GAAIA,AAAAA,CAAAA,QAAAA,oBAAAA,KAAAA,IAAAA,kBAAmB,cAAc,AAAD,KAAK,CAAC,CAAC;gBAC3C,CAAC0B,oBAAAA,iBAAiBA,CAAC,EAAEjC,kBAAkB,QAAQ;YACjD;YACA,yBAAyB;QAC3B;QAGF,IAAImB,UAAUe,AAAAA,IAAAA,oBAAAA,oBAAAA,AAAAA,EAAqBC,oBAAAA,wBAAwBA,GAAG;YAC5D,IAAIP,sBAAAA,WAAWA,EACb,MAAM,IAAIQ,MAAM;YAElBC,QAAQ,GAAG,CAAC;YACZ,MAAM,EAAEC,UAAU,EAAE,GAAG,MAAM;YAC7BnB,SAASmB,WAAWnB;QACtB;QAEA,IAAI,AAAkB,WAAXA,QACT,OAAO;YACL,YAAYA,OAAO,IAAI,CAAC,WAAW;YACnC,OAAO;YACPf;QACF;QAIF,IAAIY,iBACFG,SAAS,IAAIoB,oBAAAA,SAASA,CAAC;YACrB,QAAQtB;YACR,WAAWG;YACX,yBAAyB;QAC3B;QAGF,IAAI,AAAkB,WAAXD,UAA2BA,OAAe,QAAQ,EAC3D,OAAO;YACL,YAAaA,OAAe,QAAQ;YACpC,OAAO;YACPf;QACF;QAGF,MAAM,IAAIgC,MAAM;IAClB;IAEO,eAAeI,KACpBC,QAAsC,EACtCzC,iBAA+B,EAC/BC,gBAAmC,EACnCyC,OAGC;QAED,MAAM,EAAEC,UAAU,EAAEC,KAAK,EAAExC,SAAS,EAAE,GAAG,MAAML,iBAAiB;YAC9DC;YACAC;QACF;QAEA,MAAM4C,iBAAiBC,kBAAkB1C,WAAWJ;QAEpD,MAAM+C,YAAYC,AAAAA,IAAAA,oBAAAA,WAAAA,AAAAA,EAAYC,oBAAAA,iBAAiBA;QAC/C,MAAMC,YAAY5B,AAAAA,IAAAA,uBAAAA,QAAAA,AAAAA,EAAS;QAC3B,MAAM6B,oBAAoB7B,AAAAA,IAAAA,uBAAAA,QAAAA,AAAAA,EAAS;QACnC,MAAM8B,qBAAqB9B,AAAAA,IAAAA,uBAAAA,QAAAA,AAAAA,EAAS;QAEpC,MAAM+B,YAAYC,KAAK,GAAG;QAE1B,MAAMC,cAAcb,AAAAA,CAAAA,QAAAA,UAAAA,KAAAA,IAAAA,QAAS,MAAM,AAAD,KAAKA,CAAAA,QAAAA,UAAAA,KAAAA,IAAAA,QAAS,OAAO,AAAD;QACtD,IAAIc;QACJ,IAAIC,cAAc;QAClB,IAAIC;QACJ,IAAIC;QAEJ,MAAMC,eAAe;YACnB,aAAaC,AAAmC,kBAAnCA,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA,EAAa5D,oBAAsC,MAAM;YACtE,QAAQ,CAAC,CAACsD;YACV,YACE,AAAqB,YAArB,OAAOR,YACHA,YACAe,OAAO,QAAQ,CAACf,aAAa,QAAQ;YAC3C,GAAIc,AAAmC,cAAnCA,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA,EAAa5D,oBACb;gBACE,2BAA2B;YAC7B,IACA,CAAC,CAAC;QACR;QAEA,IAAI;YACF,IAAI2C,AAAU,aAAVA,OAAoB;gBACtBM,UACE,CAAC,QAAQ,EAAEK,cAAc,eAAe,GAAG,WAAW,EAAEnD,WAAW;gBAGrE,IAAImD,aAAa;oBACf,MAAMQ,SAAU,MAAMpB,WAAW,MAAM,CACrC;wBACE,OAAOvC;wBACPqC;wBACA,iBAAiBI;wBACjB,GAAGe,YAAY;oBACjB,GACA;wBACE,QAAQ;oBACV;oBAKF,WAAW,MAAMI,SAASD,OAAQ;4BAChBE,uBAAAA,iBAAAA,gBAEbC,wBAAAA,kBAAAA,iBAoBCC,kBAAAA;wBAtBJ,MAAMX,UAAUS,AAAAA,SAAAA,CAAAA,iBAAAA,MAAM,OAAO,AAAD,IAAZA,KAAAA,IAAAA,QAAAA,CAAAA,kBAAAA,cAAe,CAAC,EAAE,AAAD,IAAjBA,KAAAA,IAAAA,QAAAA,CAAAA,wBAAAA,gBAAoB,KAAK,AAAD,IAAxBA,KAAAA,IAAAA,sBAA2B,OAAO,AAAD,KAAK;wBACtD,MAAMG,oBACJ,AAAC,SAAAF,CAAAA,kBAAAA,MAAM,OAAO,AAAD,IAAZA,KAAAA,IAAAA,QAAAA,CAAAA,mBAAAA,eAAe,CAAC,EAAE,AAAD,IAAjBA,KAAAA,IAAAA,QAAAA,CAAAA,yBAAAA,iBAAoB,KAAK,AAAD,IAAxBA,KAAAA,IAAAA,uBAAmC,iBAAiB,AAAD,KAAK;wBAG3D,IAAIF,MAAM,KAAK,EACbN,QAAQM,MAAM,KAAK;wBAGrB,IAAIR,WAAWY,mBAAmB;4BAChCX,eAAeD;4BACf,MAAMa,YAAiC;gCACrCb;gCACAY;gCACAX;gCACA,YAAY;gCACZ,OAAOa;4BACT;4BACA5B,QAAQ,OAAO,CAAE2B;wBACnB;wBAGA,IAAI,QAAAF,CAAAA,kBAAAA,MAAM,OAAO,AAAD,IAAZA,KAAAA,IAAAA,QAAAA,CAAAA,mBAAAA,eAAe,CAAC,EAAE,AAAD,IAAjBA,KAAAA,IAAAA,iBAAoB,aAAa,EAAE;4BACrCR,WAAWL,KAAK,GAAG,KAAKD;4BAGxB,IAAI,CAACK,OAAO;gCAEV,MAAMa,kBAAkBC,KAAK,GAAG,CAC9B,GACAA,KAAK,KAAK,CAACf,YAAY,MAAM,GAAG;gCAElCC,QAAQ;oCACN,eAAea;oCACf,mBAAmBA;oCACnB,cAAcA,AAAkB,IAAlBA;gCAChB;4BACF;4BAGA,MAAME,aAAkC;gCACtC,SAAS;gCACThB;gCACA,mBAAmB;gCACnB,YAAY;gCACZ,OAAO;oCACL,eAAeC,MAAM,aAAa,IAAI;oCACtC,mBAAmBA,MAAM,iBAAiB,IAAI;oCAC9C,cAAcA,MAAM,YAAY,IAAI;oCACpC,WAAWC,YAAY;oCACvB,YAAYvD;gCACd;4BACF;4BACAsC,QAAQ,OAAO,CAAE+B;4BACjB;wBACF;oBACF;oBACAjB,UAAUC;oBACVN,kBACE,CAAC,iBAAiB,EAAE/C,UAAU,QAAQ,EAAEyD,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA,EAAa5D,qBAAqB,UAAU,WAAW,EAAE0D,UAAU;gBAE/G,OAAO;wBAUoJe,eAAyDC,gBAAwDC;oBAT1Q,MAAMC,SAAS,MAAMlC,WAAW,MAAM,CAAC;wBACrC,OAAOvC;wBACPqC;wBACA,iBAAiBI;wBACjB,GAAGe,YAAY;oBACjB;oBACAD,WAAWL,KAAK,GAAG,KAAKD;oBAExBF,kBACE,CAAC,OAAO,EAAE/C,UAAU,QAAQ,EAAEyD,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA,EAAa5D,qBAAqB,UAAU,mBAAmB,EAAE6E,AAAAA,IAAAA,oBAAAA,kBAAAA,AAAAA,EAAmB7E,kBAAkB,iBAAiB,EAAEyE,AAAAA,SAAAA,CAAAA,gBAAAA,OAAO,KAAK,AAAD,IAAXA,KAAAA,IAAAA,cAAc,aAAa,AAAD,KAAK,GAAG,qBAAqB,EAAEC,AAAAA,SAAAA,CAAAA,iBAAAA,OAAO,KAAK,AAAD,IAAXA,KAAAA,IAAAA,eAAc,iBAAiB,AAAD,KAAK,GAAG,gBAAgB,EAAEC,AAAAA,SAAAA,CAAAA,iBAAAA,OAAO,KAAK,AAAD,IAAXA,KAAAA,IAAAA,eAAc,YAAY,AAAD,KAAK,GAAG,WAAW,EAAEjB,SAAS,aAAa,EAAEkB,OAAO,WAAW,IAAI,IAAI;oBAG1WzB,mBACE,CAAC,oBAAoB,EAAE2B,KAAK,SAAS,CAACF,OAAO,KAAK,GAAG;oBAGvDlD,IAAAA,sBAAAA,MAAAA,AAAAA,EACEkD,OAAO,OAAO,EACd,CAAC,mCAAmC,EAAEE,KAAK,SAAS,CAACF,SAAS;oBAEhErB,UAAUqB,OAAO,OAAO,CAAC,EAAE,CAAC,OAAO,CAAC,OAAO;oBAC3CnB,QAAQmB,OAAO,KAAK;gBACtB;gBAEA3B,UAAU,CAAC,UAAU,EAAEM,SAAS;gBAChC7B,IAAAA,sBAAAA,MAAAA,AAAAA,EAAO6B,SAAS;YAClB,OAAO,IAAIZ,AAAU,gBAAVA,OAAuB;gBAChC,MAAMoC,sBAAsB,CAACxB;oBAC3B,IAAIA,AAAiB,gBAAjBA,QAAQ,IAAI,EAAkB;wBAChC,MAAMyB,YAAYzB,QAAQ,SAAS,CAAC,GAAG;wBACvC7B,IAAAA,sBAAAA,MAAAA,AAAAA,EAAOsD,WAAW;wBAClB,MAAM,EAAEC,QAAQ,EAAEC,IAAI,EAAE,GAAGC,AAAAA,IAAAA,oBAAAA,WAAAA,AAAAA,EAAY5B,QAAQ,SAAS,CAAC,GAAG;wBAC5D,OAAO;4BACL,QAAQ;gCACN,MAAM;gCACN,YAAY0B;gCACZ,MAAMC;4BACR;4BACA,MAAM;wBACR;oBACF;oBACA,OAAO3B;gBACT;gBAEA,IAAID,aAAa;oBACf,MAAMQ,SAAU,MAAMpB,WAAW,MAAM,CAAC;wBACtC,OAAOvC;wBACP,QAAQ;wBACR,UAAUqC,SAAS,GAAG,CAAC,CAAC4C,IAAO;gCAC7B,MAAM;gCACN,SAASC,MAAM,OAAO,CAACD,EAAE,OAAO,IAC3BA,EAAE,OAAO,CAAS,GAAG,CAACL,uBACvBK,EAAE,OAAO;4BACf;wBACA,iBAAiBxC;wBACjB,GAAGe,YAAY;oBACjB;oBAEA,WAAW,MAAMI,SAASD,OAAQ;4BAChBwB;wBAAhB,MAAM/B,UAAU+B,AAAAA,SAAAA,CAAAA,eAAAA,MAAM,KAAK,AAAD,IAAVA,KAAAA,IAAAA,aAAa,IAAI,AAAD,KAAK;wBACrC,IAAI/B,SAAS;4BACXC,eAAeD;4BACf,MAAMa,YAAiC;gCACrCb;gCACAC;gCACA,mBAAmB;gCACnB,YAAY;gCACZ,OAAOa;4BACT;4BACA5B,QAAQ,OAAO,CAAE2B;wBACnB;wBAGA,IAAIL,AAAe,mBAAfA,MAAM,IAAI,EAAqB;4BACjCL,WAAWL,KAAK,GAAG,KAAKD;4BACxB,MAAMmC,iBAAiBxB,MAAM,KAAK;4BAGlC,MAAMS,aAAkC;gCACtC,SAAS;gCACThB;gCACA,mBAAmB;gCACnB,YAAY;gCACZ,OAAO+B,iBACH;oCACE,eAAeA,eAAe,YAAY,IAAI;oCAC9C,mBAAmBA,eAAe,aAAa,IAAI;oCACnD,cACGA,AAAAA,CAAAA,eAAe,YAAY,IAAI,KAC/BA,CAAAA,eAAe,aAAa,IAAI;oCACnC,WAAW7B,YAAY;oCACvB,YAAYvD;gCACd,IACAkE;4BACN;4BACA5B,QAAQ,OAAO,CAAE+B;4BACjB;wBACF;oBACF;oBACAjB,UAAUC;gBACZ,OAAO;oBACL,MAAMoB,SAAS,MAAMlC,WAAW,MAAM,CAAC;wBACrC,OAAOvC;wBACP,QAAQ;wBACR,UAAUqC,SAAS,GAAG,CAAC,CAAC4C,IAAO;gCAC7B,MAAM;gCACN,SAASC,MAAM,OAAO,CAACD,EAAE,OAAO,IAC3BA,EAAE,OAAO,CAAS,GAAG,CAACL,uBACvBK,EAAE,OAAO;4BACf;wBACA,iBAAiBxC;wBACjB,GAAGe,YAAY;oBACjB;oBACAD,WAAWL,KAAK,GAAG,KAAKD;oBACxBG,UAAWqB,OAAe,OAAO,CAAC,EAAE,CAAC,IAAI;oBACzCnB,QAAQmB,OAAO,KAAK;gBACtB;gBAEAlD,IAAAA,sBAAAA,MAAAA,AAAAA,EAAO6B,SAAS;YAClB;YAEA,IAAID,eAAe,CAACG,OAAO;gBAEzB,MAAMa,kBAAkBC,KAAK,GAAG,CAC9B,GACAA,KAAK,KAAK,CAAEhB,AAAAA,CAAAA,WAAW,EAAC,EAAG,MAAM,GAAG;gBAEtCE,QAAQ;oBACN,eAAea;oBACf,mBAAmBA;oBACnB,cAAcA,AAAkB,IAAlBA;gBAChB;YACF;YAEA,OAAO;gBACL,SAASf,WAAW;gBACpB,OAAOE,QACH;oBACE,eAAeA,MAAM,aAAa,IAAI;oBACtC,mBAAmBA,MAAM,iBAAiB,IAAI;oBAC9C,cAAcA,MAAM,YAAY,IAAI;oBACpC,WAAWC,YAAY;oBACvB,YAAYvD;gBACd,IACAkE;gBACJ,YAAY,CAAC,CAACf;YAChB;QACF,EAAE,OAAOkC,GAAQ;YACfpD,QAAQ,KAAK,CAAC,kBAAkBoD;YAChC,MAAMC,WAAW,IAAItD,MACnB,CAAC,eAAe,EAAEmB,cAAc,eAAe,GAAG,kBAAkB,EAAEkC,EAAE,OAAO,CAAC,8DAA8D,CAAC,EAC/I;gBACE,OAAOA;YACT;YAEF,MAAMC;QACR;IACF;IAEO,MAAM5C,oBAAoB,CAC/B1C,WACAJ;QAIA,IAAI6C;QAKJ,IAAIzC,UAAU,QAAQ,CAAC,UACrB,OAAQJ;YACN,KAAK2F,mCAAAA,YAAAA,CAAAA,MAAmB;gBACtB9C,iBAAiB+C,6BAAAA,YAAYA;gBAC7B;YACF,KAAKD,mCAAAA,YAAAA,CAAAA,eAA4B;gBAC/B9C,iBAAiBgD,+BAAAA,aAAaA;gBAC9B;YACF,KAAKF,mCAAAA,YAAAA,CAAAA,IAAiB;gBACpB9C,iBAAiBiD,gCAAAA,UAAUA;gBAC3B;YACF,KAAKH,mCAAAA,YAAAA,CAAAA,YAAyB;YAC9B,KAAKA,mCAAAA,YAAAA,CAAAA,gBAA6B;gBAChC9C,iBAAiB;oBAAE,MAAMkD,kCAAAA,gBAAAA,CAAAA,IAAqB;gBAAC;gBAC/C;QACJ;QAIF,IAAI3F,AAAc,wBAAdA,WACFyC,iBAAiB;YAAE,MAAMkD,kCAAAA,gBAAAA,CAAAA,IAAqB;QAAC;QAGjD,OAAOlD;IACT;IAEO,eAAemD,oBACpBvD,QAAsC,EACtCzC,iBAA+B,EAC/BC,gBAAmC;QAEnC,MAAMgG,WAAW,MAAMzD,KAAKC,UAAUzC,mBAAmBC;QACzD0B,IAAAA,sBAAAA,MAAAA,AAAAA,EAAOsE,UAAU;QACjB,MAAMC,cAAcC,cAAcF,SAAS,OAAO,EAAEhG;QACpD,OAAO;YAAE,SAASiG;YAAa,OAAOD,SAAS,KAAK;QAAC;IACvD;IAEO,eAAeG,2BACpBC,IAAY,EACZrG,iBAA+B,EAC/BC,gBAAmC;QAEnC,MAAM,EAAEuD,OAAO,EAAEE,KAAK,EAAE,GAAG,MAAMlB,KAC/B6D,MACArG,mBACAC;QAEF,OAAO;YAAEuD;YAASE;QAAM;IAC1B;IAEO,SAAS4C,yBAAyBL,QAAgB;QACvD,IAAI;YAEF,MAAMM,YAAYN,SAAS,KAAK,CAAC;YACjC,IAAIM,WACF,OAAOA,SAAS,CAAC,EAAE;YAIrB,MAAMC,iBAAiBP,SAAS,KAAK,CACnC;YAEF,IAAIO,gBACF,OAAOA,cAAc,CAAC,EAAE;YAI1B,MAAMC,gBAAgBR,SAAS,KAAK,CAAC;YACrC,IAAIQ,eACF,OAAOA,aAAa,CAAC,EAAE;QAE3B,EAAE,OAAM,CAAC;QAET,OAAOR;IACT;IAEO,SAASS,yBAAyBC,KAAa;QACpD,IAAIA,MAAM,QAAQ,CAAC,SAEjB,MAAO,YAAY,IAAI,CAACA,OACtBA,QAAQA,MAAM,OAAO,CAAC,kBAAkB;QAG5C,OAAOA;IACT;IAEO,SAASR,cACdQ,KAAa,EACb1G,gBAAmC;QAEnC,MAAM2G,kBAAkBN,yBAAyBK;QAEjD,IAAIC,QAAAA,kBAAAA,KAAAA,IAAAA,gBAAiB,KAAK,CAAC,oBAAoB;gBACtCC;YAAP,OAAO,QAAAA,CAAAA,yBAAAA,gBACJ,KAAK,CAAC,kBAAiB,IADnBA,KAAAA,IAAAA,uBAEH,KAAK,CAAC,GACP,GAAG,CAAC/C;QACT;QACA,IAAI;YACF,OAAOiB,KAAK,KAAK,CAAC6B;QACpB,EAAE,OAAM,CAAC;QACT,IAAI;YACF,OAAO7B,KAAK,KAAK,CAAC+B,AAAAA,IAAAA,oCAAAA,UAAAA,AAAAA,EAAWF;QAC/B,EAAE,OAAOnB,GAAG,CAAC;QAEb,IACE5B,AAAmC,oBAAnCA,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA,EAAa5D,qBACb4D,AAAmC,kBAAnCA,AAAAA,IAAAA,oBAAAA,YAAAA,AAAAA,EAAa5D,mBACb;YACA,MAAM8G,aAAaL,yBAAyBE;YAC5C,OAAO7B,KAAK,KAAK,CAAC+B,AAAAA,IAAAA,oCAAAA,UAAAA,AAAAA,EAAWC;QAC/B;QACA,MAAM3E,MAAM,CAAC,+BAA+B,EAAEuE,OAAO;IACvD"}

package/dist/lib/ai-model/ui-tars-planning.js

@@ -45,7 +45,7 @@ const pointToBbox = (point, width, height)=>[
         Math.round(Math.min(point.y + bboxSize / 2, height))
     ];
 async function vlmPlanning(options) {
-    const { conversationHistory, userInstruction, size } = options;
+    const { conversationHistory, userInstruction, size, modelPreferences } = options;
     const systemPrompt = (0, ui_tars_planning_js_namespaceObject.getUiTarsPlanningPrompt)() + userInstruction;
     const res = await (0, index_js_namespaceObject.call)([
         {
@@ -53,9 +53,9 @@ async function vlmPlanning(options) {
             content: systemPrompt
         },
         ...conversationHistory
-    ], external_common_js_namespaceObject.AIActionType.INSPECT_ELEMENT);
+    ], external_common_js_namespaceObject.AIActionType.INSPECT_ELEMENT, modelPreferences);
     const convertedText = convertBboxToCoordinates(res.content);
-    const modelVer = (0, env_namespaceObject.uiTarsModelVersion)();
+    const modelVer = (0, env_namespaceObject.uiTarsModelVersion)(modelPreferences);
     const { parsed } = (0, action_parser_namespaceObject.actionParser)({
         prediction: convertedText,
         factor: [
@@ -240,8 +240,8 @@ function getPoint(startBox, size) {
         y * size.height
     ];
 }
-async function resizeImageForUiTars(imageBase64, size) {
-    if ('vlm-ui-tars' === (0, env_namespaceObject.vlLocateMode)() && (0, env_namespaceObject.uiTarsModelVersion)() === env_namespaceObject.UITarsModelVersion.V1_5) {
+async function resizeImageForUiTars(imageBase64, size, modelPreferences) {
+    if ('vlm-ui-tars' === (0, env_namespaceObject.vlLocateMode)(modelPreferences) && (0, env_namespaceObject.uiTarsModelVersion)(modelPreferences) === env_namespaceObject.UITarsModelVersion.V1_5) {
         debug('ui-tars-v1.5, will check image size', size);
         const currentPixels = size.width * size.height;
         const maxPixels = 12845056;