@midscene/core 0.25.4-beta-20250807040242.0 → 0.25.4-beta-20250807062119.0

package/dist/lib/{chunk-I5LBWOQA.js → chunk-G2JTYWI6.js}

@@ -10,25 +10,6 @@ var _identity = require('@azure/identity');
 
 
 
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
 
 
 var _env = require('@midscene/shared/env');
@@ -55,10 +36,11 @@ var AIActionType = /* @__PURE__ */ ((AIActionType2) => {
   AIActionType2[AIActionType2["DESCRIBE_ELEMENT"] = 4] = "DESCRIBE_ELEMENT";
   return AIActionType2;
 })(AIActionType || {});
-async function callAiFn(msgs, AIActionTypeValue) {
+async function callAiFn(msgs, AIActionTypeValue, modelPreferences) {
   const { content, usage } = await callToGetJSONObject(
     msgs,
-    AIActionTypeValue
+    AIActionTypeValue,
+    modelPreferences
   );
   return { content, usage };
 }
@@ -1163,24 +1145,57 @@ pageDescription:
   });
 };
 
-// src/ai-model/service-caller/index.ts
-function checkAIConfig() {
-  const openaiKey = _env.getAIConfig.call(void 0, _env.OPENAI_API_KEY);
-  const azureConfig = _env.getAIConfig.call(void 0, _env.MIDSCENE_USE_AZURE_OPENAI);
-  const anthropicKey = _env.getAIConfig.call(void 0, _env.ANTHROPIC_API_KEY);
-  const initConfigJson = _env.getAIConfig.call(void 0, _env.MIDSCENE_OPENAI_INIT_CONFIG_JSON);
-  if (openaiKey)
-    return true;
-  if (azureConfig)
-    return true;
-  if (anthropicKey)
-    return true;
-  return Boolean(initConfigJson);
+// src/ai-model/service-caller/utils.ts
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+function getModelName() {
+  let modelName = "gpt-4o";
+  const nameInConfig = _env.getAIConfig.call(void 0, _env.MIDSCENE_MODEL_NAME);
+  if (nameInConfig) {
+    modelName = nameInConfig;
+  }
+  return modelName;
 }
-var debugConfigInitialized = false;
 function initDebugConfig() {
-  if (debugConfigInitialized)
-    return;
   const shouldPrintTiming = _env.getAIConfigInBoolean.call(void 0, _env.MIDSCENE_DEBUG_AI_PROFILE);
   let debugConfig = "";
   if (shouldPrintTiming) {
@@ -1205,25 +1220,230 @@ function initDebugConfig() {
   if (debugConfig) {
     _logger.enableDebug.call(void 0, debugConfig);
   }
-  debugConfigInitialized = true;
 }
-var defaultModel = "gpt-4o";
-function getModelName() {
-  let modelName = defaultModel;
-  const nameInConfig = _env.getAIConfig.call(void 0, _env.MIDSCENE_MODEL_NAME);
-  if (nameInConfig) {
-    modelName = nameInConfig;
+var createAssert = (modelNameKey, modelName) => (value, key, modelVendorFlag) => {
+  if (modelVendorFlag) {
+    _utils.assert.call(void 0, 
+      value,
+      `The ${key} must be a non-empty string because of the ${modelNameKey} is declared as ${modelName} and ${modelVendorFlag} has also been specified, but got: ${value}
+Please check your config.`
+    );
+  } else {
+    _utils.assert.call(void 0, 
+      value,
+      `The ${key} must be a non-empty string because of the ${modelNameKey} is declared as ${modelName}, but got: ${value}
+Please check your config.`
+    );
   }
-  return modelName;
-}
+};
+var getModelConfigFromEnv = (modelName, keys, valueAssert) => {
+  const socksProxy = _env.getAIConfig.call(void 0, keys.socksProxy);
+  const httpProxy = _env.getAIConfig.call(void 0, keys.httpProxy);
+  if (_env.getAIConfig.call(void 0, keys.openaiUseAzureDeprecated)) {
+    const openaiBaseURL = _env.getAIConfig.call(void 0, keys.openaiBaseURL);
+    const openaiApiKey = _env.getAIConfig.call(void 0, keys.openaiApiKey);
+    const openaiExtraConfig = _env.getAIConfigInJson.call(void 0, keys.openaiExtraConfig);
+    valueAssert(
+      openaiBaseURL,
+      keys.openaiBaseURL,
+      keys.openaiUseAzureDeprecated
+    );
+    valueAssert(openaiApiKey, keys.openaiApiKey, keys.openaiUseAzureDeprecated);
+    return {
+      socksProxy,
+      httpProxy,
+      modelName,
+      openaiUseAzureDeprecated: true,
+      openaiApiKey,
+      openaiBaseURL,
+      openaiExtraConfig
+    };
+  } else if (_env.getAIConfig.call(void 0, keys.useAzureOpenai)) {
+    const azureOpenaiScope = _env.getAIConfig.call(void 0, keys.azureOpenaiScope);
+    const azureOpenaiApiKey = _env.getAIConfig.call(void 0, keys.azureOpenaiApiKey);
+    const azureOpenaiEndpoint = _env.getAIConfig.call(void 0, keys.azureOpenaiEndpoint);
+    const azureOpenaiDeployment = _env.getAIConfig.call(void 0, keys.azureOpenaiDeployment);
+    const azureOpenaiApiVersion = _env.getAIConfig.call(void 0, keys.azureOpenaiApiVersion);
+    const azureExtraConfig = _env.getAIConfigInJson.call(void 0, keys.azureExtraConfig);
+    const openaiExtraConfig = _env.getAIConfigInJson.call(void 0, keys.openaiExtraConfig);
+    valueAssert(azureOpenaiApiKey, keys.azureOpenaiApiKey, keys.useAzureOpenai);
+    return {
+      socksProxy,
+      httpProxy,
+      modelName,
+      useAzureOpenai: true,
+      azureOpenaiScope,
+      azureOpenaiApiKey,
+      azureOpenaiEndpoint,
+      azureOpenaiDeployment,
+      azureOpenaiApiVersion,
+      azureExtraConfig,
+      openaiExtraConfig
+    };
+  } else if (_env.getAIConfig.call(void 0, keys.useAnthropicSdk)) {
+    const anthropicApiKey = _env.getAIConfig.call(void 0, keys.anthropicApiKey);
+    valueAssert(anthropicApiKey, keys.anthropicApiKey, keys.useAnthropicSdk);
+    return {
+      socksProxy,
+      httpProxy,
+      modelName,
+      useAnthropicSdk: true,
+      anthropicApiKey
+    };
+  } else {
+    const openaiBaseURL = _env.getAIConfig.call(void 0, keys.openaiBaseURL);
+    const openaiApiKey = _env.getAIConfig.call(void 0, keys.openaiApiKey);
+    const openaiExtraConfig = _env.getAIConfigInJson.call(void 0, keys.openaiExtraConfig);
+    valueAssert(openaiBaseURL, keys.openaiBaseURL);
+    valueAssert(openaiApiKey, keys.openaiApiKey);
+    return {
+      socksProxy,
+      httpProxy,
+      modelName,
+      openaiBaseURL,
+      openaiApiKey,
+      openaiExtraConfig
+    };
+  }
+};
+var maskKey = (key, maskChar = "*") => {
+  if (typeof key !== "string" || key.length === 0) {
+    return key;
+  }
+  const prefixLen = 3;
+  const suffixLen = 3;
+  const keepLength = prefixLen + suffixLen;
+  if (key.length <= keepLength) {
+    return key;
+  }
+  const prefix = key.substring(0, prefixLen);
+  const suffix = key.substring(key.length - suffixLen);
+  const maskLength = key.length - keepLength;
+  const mask = maskChar.repeat(maskLength);
+  return `${prefix}${mask}${suffix}`;
+};
+var maskConfig = (config) => {
+  return Object.fromEntries(
+    Object.entries(config).map(([key, value]) => [
+      key,
+      ["openaiApiKey", "azureOpenaiApiKey", "anthropicApiKey"].includes(key) ? maskKey(value) : value
+    ])
+  );
+};
+var decideModelConfig = (modelPreferences) => {
+  initDebugConfig();
+  const debugLog = _logger.getDebug.call(void 0, "ai:decideModelConfig");
+  debugLog("modelPreferences", modelPreferences);
+  const isVQAIntent = _optionalChain([modelPreferences, 'optionalAccess', _21 => _21.intent]) === "VQA";
+  const vqaModelName = _env.getAIConfig.call(void 0, _env.MIDSCENE_VQA_MODEL_NAME);
+  if (isVQAIntent && vqaModelName) {
+    debugLog(
+      `current action is a VQA action and detected ${_env.MIDSCENE_VQA_MODEL_NAME} ${vqaModelName}, will only read VQA related model config from process.env`
+    );
+    const config = getModelConfigFromEnv(
+      vqaModelName,
+      {
+        /**
+         * proxy
+         */
+        socksProxy: _env.MIDSCENE_VQA_OPENAI_SOCKS_PROXY,
+        httpProxy: _env.MIDSCENE_VQA_OPENAI_HTTP_PROXY,
+        /**
+         * OpenAI
+         */
+        openaiBaseURL: _env.MIDSCENE_VQA_OPENAI_BASE_URL,
+        openaiApiKey: _env.MIDSCENE_VQA_OPENAI_API_KEY,
+        openaiExtraConfig: _env.MIDSCENE_VQA_OPENAI_INIT_CONFIG_JSON,
+        /**
+         * Azure
+         */
+        openaiUseAzureDeprecated: _env.MIDSCENE_VQA_OPENAI_USE_AZURE,
+        useAzureOpenai: _env.MIDSCENE_VQA_USE_AZURE_OPENAI,
+        azureOpenaiScope: _env.MIDSCENE_VQA_AZURE_OPENAI_SCOPE,
+        azureOpenaiApiKey: _env.MIDSCENE_VQA_AZURE_OPENAI_KEY,
+        azureOpenaiEndpoint: _env.MIDSCENE_VQA_AZURE_OPENAI_ENDPOINT,
+        azureOpenaiApiVersion: _env.MIDSCENE_VQA_AZURE_OPENAI_API_VERSION,
+        azureOpenaiDeployment: _env.MIDSCENE_VQA_AZURE_OPENAI_DEPLOYMENT,
+        azureExtraConfig: _env.MIDSCENE_VQA_AZURE_OPENAI_INIT_CONFIG_JSON,
+        /**
+         * Anthropic
+         */
+        useAnthropicSdk: _env.MIDSCENE_VQA_USE_ANTHROPIC_SDK,
+        anthropicApiKey: _env.MIDSCENE_VQA_ANTHROPIC_API_KEY
+      },
+      createAssert(_env.MIDSCENE_VQA_MODEL_NAME, vqaModelName)
+    );
+    debugLog("got model config for VQA usage:", maskConfig(config));
+    return config;
+  } else {
+    debugLog("read model config from process.env as normal.");
+    const commonModelName = _env.getAIConfig.call(void 0, _env.MIDSCENE_MODEL_NAME);
+    _utils.assert.call(void 0, 
+      commonModelName,
+      `${_env.MIDSCENE_MODEL_NAME} is empty, please check your config.`
+    );
+    const config = getModelConfigFromEnv(
+      commonModelName,
+      {
+        /**
+         * proxy
+         */
+        socksProxy: _env.MIDSCENE_OPENAI_SOCKS_PROXY,
+        httpProxy: _env.MIDSCENE_OPENAI_HTTP_PROXY,
+        /**
+         * OpenAI
+         */
+        openaiBaseURL: _env.OPENAI_BASE_URL,
+        openaiApiKey: _env.OPENAI_API_KEY,
+        openaiExtraConfig: _env.MIDSCENE_OPENAI_INIT_CONFIG_JSON,
+        /**
+         * Azure
+         */
+        openaiUseAzureDeprecated: _env.OPENAI_USE_AZURE,
+        useAzureOpenai: _env.MIDSCENE_USE_AZURE_OPENAI,
+        azureOpenaiScope: _env.MIDSCENE_AZURE_OPENAI_SCOPE,
+        azureOpenaiApiKey: _env.AZURE_OPENAI_KEY,
+        azureOpenaiEndpoint: _env.AZURE_OPENAI_ENDPOINT,
+        azureOpenaiApiVersion: _env.AZURE_OPENAI_API_VERSION,
+        azureOpenaiDeployment: _env.AZURE_OPENAI_DEPLOYMENT,
+        azureExtraConfig: _env.MIDSCENE_AZURE_OPENAI_INIT_CONFIG_JSON,
+        /**
+         * Anthropic
+         */
+        useAnthropicSdk: _env.MIDSCENE_USE_ANTHROPIC_SDK,
+        anthropicApiKey: _env.ANTHROPIC_API_KEY
+      },
+      createAssert(_env.MIDSCENE_MODEL_NAME, commonModelName)
+    );
+    debugLog("got model config for common usage:", maskConfig(config));
+    return config;
+  }
+};
+
+// src/ai-model/service-caller/index.ts
 async function createChatClient({
-  AIActionTypeValue
+  AIActionTypeValue,
+  modelPreferences
 }) {
-  initDebugConfig();
+  const {
+    socksProxy,
+    httpProxy,
+    modelName,
+    openaiBaseURL,
+    openaiApiKey,
+    openaiExtraConfig,
+    openaiUseAzureDeprecated,
+    useAzureOpenai,
+    azureOpenaiScope,
+    azureOpenaiApiKey,
+    azureOpenaiEndpoint,
+    azureOpenaiApiVersion,
+    azureOpenaiDeployment,
+    azureExtraConfig,
+    useAnthropicSdk,
+    anthropicApiKey
+  } = decideModelConfig(modelPreferences);
   let openai;
-  const extraConfig = _env.getAIConfigInJson.call(void 0, _env.MIDSCENE_OPENAI_INIT_CONFIG_JSON);
-  const socksProxy = _env.getAIConfig.call(void 0, _env.MIDSCENE_OPENAI_SOCKS_PROXY);
-  const httpProxy = _env.getAIConfig.call(void 0, _env.MIDSCENE_OPENAI_HTTP_PROXY);
   let proxyAgent = void 0;
   const debugProxy = _logger.getDebug.call(void 0, "ai:call:proxy");
   if (httpProxy) {
@@ -1233,64 +1453,50 @@ async function createChatClient({
     debugProxy("using socks proxy", socksProxy);
     proxyAgent = new (0, _socksproxyagent.SocksProxyAgent)(socksProxy);
   }
-  if (_env.getAIConfig.call(void 0, _env.OPENAI_USE_AZURE)) {
+  if (openaiUseAzureDeprecated) {
     openai = new (0, _openai.AzureOpenAI)({
-      baseURL: _env.getAIConfig.call(void 0, _env.OPENAI_BASE_URL),
-      apiKey: _env.getAIConfig.call(void 0, _env.OPENAI_API_KEY),
+      baseURL: openaiBaseURL,
+      apiKey: openaiApiKey,
       httpAgent: proxyAgent,
-      ...extraConfig,
+      ...openaiExtraConfig,
       dangerouslyAllowBrowser: true
     });
-  } else if (_env.getAIConfig.call(void 0, _env.MIDSCENE_USE_AZURE_OPENAI)) {
-    const extraAzureConfig = _env.getAIConfigInJson.call(void 0, 
-      _env.MIDSCENE_AZURE_OPENAI_INIT_CONFIG_JSON
-    );
-    const scope = _env.getAIConfig.call(void 0, _env.MIDSCENE_AZURE_OPENAI_SCOPE);
+  } else if (useAzureOpenai) {
     let tokenProvider = void 0;
-    if (scope) {
+    if (azureOpenaiScope) {
       _utils.assert.call(void 0, 
         !_utils.ifInBrowser,
         "Azure OpenAI is not supported in browser with Midscene."
       );
       const credential = new (0, _identity.DefaultAzureCredential)();
-      _utils.assert.call(void 0, scope, "MIDSCENE_AZURE_OPENAI_SCOPE is required");
-      tokenProvider = _identity.getBearerTokenProvider.call(void 0, credential, scope);
+      tokenProvider = _identity.getBearerTokenProvider.call(void 0, credential, azureOpenaiScope);
       openai = new (0, _openai.AzureOpenAI)({
         azureADTokenProvider: tokenProvider,
-        endpoint: _env.getAIConfig.call(void 0, _env.AZURE_OPENAI_ENDPOINT),
-        apiVersion: _env.getAIConfig.call(void 0, _env.AZURE_OPENAI_API_VERSION),
-        deployment: _env.getAIConfig.call(void 0, _env.AZURE_OPENAI_DEPLOYMENT),
-        ...extraConfig,
-        ...extraAzureConfig
+        endpoint: azureOpenaiEndpoint,
+        apiVersion: azureOpenaiApiVersion,
+        deployment: azureOpenaiDeployment,
+        ...openaiExtraConfig,
+        ...azureExtraConfig
       });
     } else {
       openai = new (0, _openai.AzureOpenAI)({
-        apiKey: _env.getAIConfig.call(void 0, _env.AZURE_OPENAI_KEY),
-        endpoint: _env.getAIConfig.call(void 0, _env.AZURE_OPENAI_ENDPOINT),
-        apiVersion: _env.getAIConfig.call(void 0, _env.AZURE_OPENAI_API_VERSION),
-        deployment: _env.getAIConfig.call(void 0, _env.AZURE_OPENAI_DEPLOYMENT),
+        apiKey: azureOpenaiApiKey,
+        endpoint: azureOpenaiEndpoint,
+        apiVersion: azureOpenaiApiVersion,
+        deployment: azureOpenaiDeployment,
         dangerouslyAllowBrowser: true,
-        ...extraConfig,
-        ...extraAzureConfig
+        ...openaiExtraConfig,
+        ...azureExtraConfig
       });
     }
-  } else if (!_env.getAIConfig.call(void 0, _env.MIDSCENE_USE_ANTHROPIC_SDK)) {
-    const baseURL = _env.getAIConfig.call(void 0, _env.OPENAI_BASE_URL);
-    if (typeof baseURL === "string") {
-      if (!/^https?:\/\//.test(baseURL)) {
-        throw new Error(
-          `OPENAI_BASE_URL must be a valid URL starting with http:// or https://, but got: ${baseURL}
-Please check your config.`
-        );
-      }
-    }
+  } else if (!useAnthropicSdk) {
     openai = new (0, _openai2.default)({
-      baseURL: _env.getAIConfig.call(void 0, _env.OPENAI_BASE_URL),
-      apiKey: _env.getAIConfig.call(void 0, _env.OPENAI_API_KEY),
+      baseURL: openaiBaseURL,
+      apiKey: openaiApiKey,
       httpAgent: proxyAgent,
-      ...extraConfig,
+      ...openaiExtraConfig,
       defaultHeaders: {
-        ..._optionalChain([extraConfig, 'optionalAccess', _21 => _21.defaultHeaders]) || {},
+        ..._optionalChain([openaiExtraConfig, 'optionalAccess', _22 => _22.defaultHeaders]) || {},
         [_env.MIDSCENE_API_TYPE]: AIActionTypeValue.toString()
       },
       dangerouslyAllowBrowser: true
@@ -1307,14 +1513,13 @@ Please check your config.`
   if (typeof openai !== "undefined") {
     return {
       completion: openai.chat.completions,
-      style: "openai"
+      style: "openai",
+      modelName
     };
   }
-  if (_env.getAIConfig.call(void 0, _env.MIDSCENE_USE_ANTHROPIC_SDK)) {
-    const apiKey = _env.getAIConfig.call(void 0, _env.ANTHROPIC_API_KEY);
-    _utils.assert.call(void 0, apiKey, "ANTHROPIC_API_KEY is required");
+  if (useAnthropicSdk) {
     openai = new (0, _sdk.Anthropic)({
-      apiKey,
+      apiKey: anthropicApiKey,
       httpAgent: proxyAgent,
       dangerouslyAllowBrowser: true
     });
@@ -1322,26 +1527,24 @@ Please check your config.`
   if (typeof openai !== "undefined" && openai.messages) {
     return {
       completion: openai.messages,
-      style: "anthropic"
+      style: "anthropic",
+      modelName
     };
   }
   throw new Error("Openai SDK or Anthropic SDK is not initialized");
 }
-async function call2(messages, AIActionTypeValue, responseFormat, options) {
-  _utils.assert.call(void 0, 
-    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 call2(messages, AIActionTypeValue, options, modelPreferences) {
+  const { completion, style, modelName } = await createChatClient({
+    AIActionTypeValue,
+    modelPreferences
   });
+  const responseFormat = getResponseFormat(modelName, AIActionTypeValue);
   const maxTokens = _env.getAIConfig.call(void 0, _env.OPENAI_MAX_TOKENS);
   const debugCall = _logger.getDebug.call(void 0, "ai:call");
   const debugProfileStats = _logger.getDebug.call(void 0, "ai:profile:stats");
   const debugProfileDetail = _logger.getDebug.call(void 0, "ai:profile:detail");
   const startTime = Date.now();
-  const model = getModelName();
-  const isStreaming = _optionalChain([options, 'optionalAccess', _22 => _22.stream]) && _optionalChain([options, 'optionalAccess', _23 => _23.onChunk]);
+  const isStreaming = _optionalChain([options, 'optionalAccess', _23 => _23.stream]) && _optionalChain([options, 'optionalAccess', _24 => _24.onChunk]);
   let content;
   let accumulated = "";
   let usage;
@@ -1357,12 +1560,12 @@ async function call2(messages, AIActionTypeValue, responseFormat, options) {
   try {
     if (style === "openai") {
       debugCall(
-        `sending ${isStreaming ? "streaming " : ""}request to ${model}`
+        `sending ${isStreaming ? "streaming " : ""}request to ${modelName}`
       );
       if (isStreaming) {
         const stream = await completion.create(
           {
-            model,
+            model: modelName,
             messages,
             response_format: responseFormat,
             ...commonConfig
@@ -1372,8 +1575,8 @@ async function call2(messages, AIActionTypeValue, responseFormat, options) {
           }
         );
         for await (const chunk of stream) {
-          const content2 = _optionalChain([chunk, 'access', _24 => _24.choices, 'optionalAccess', _25 => _25[0], 'optionalAccess', _26 => _26.delta, 'optionalAccess', _27 => _27.content]) || "";
-          const reasoning_content = _optionalChain([chunk, 'access', _28 => _28.choices, 'optionalAccess', _29 => _29[0], 'optionalAccess', _30 => _30.delta, 'optionalAccess', _31 => _31.reasoning_content]) || "";
+          const content2 = _optionalChain([chunk, 'access', _25 => _25.choices, 'optionalAccess', _26 => _26[0], 'optionalAccess', _27 => _27.delta, 'optionalAccess', _28 => _28.content]) || "";
+          const reasoning_content = _optionalChain([chunk, 'access', _29 => _29.choices, 'optionalAccess', _30 => _30[0], 'optionalAccess', _31 => _31.delta, 'optionalAccess', _32 => _32.reasoning_content]) || "";
           if (chunk.usage) {
             usage = chunk.usage;
           }
@@ -1388,7 +1591,7 @@ async function call2(messages, AIActionTypeValue, responseFormat, options) {
             };
             options.onChunk(chunkData);
           }
-          if (_optionalChain([chunk, 'access', _32 => _32.choices, 'optionalAccess', _33 => _33[0], 'optionalAccess', _34 => _34.finish_reason])) {
+          if (_optionalChain([chunk, 'access', _33 => _33.choices, 'optionalAccess', _34 => _34[0], 'optionalAccess', _35 => _35.finish_reason])) {
             timeCost = Date.now() - startTime;
             if (!usage) {
               const estimatedTokens = Math.max(
@@ -1419,18 +1622,18 @@ async function call2(messages, AIActionTypeValue, responseFormat, options) {
         }
         content = accumulated;
         debugProfileStats(
-          `streaming model, ${model}, mode, ${_env.vlLocateMode.call(void 0, ) || "default"}, cost-ms, ${timeCost}`
+          `streaming model, ${modelName}, mode, ${_env.vlLocateMode.call(void 0, ) || "default"}, cost-ms, ${timeCost}`
         );
       } else {
         const result = await completion.create({
-          model,
+          model: modelName,
           messages,
           response_format: responseFormat,
           ...commonConfig
         });
         timeCost = Date.now() - startTime;
         debugProfileStats(
-          `model, ${model}, mode, ${_env.vlLocateMode.call(void 0, ) || "default"}, ui-tars-version, ${_env.uiTarsModelVersion.call(void 0, )}, prompt-tokens, ${_optionalChain([result, 'access', _35 => _35.usage, 'optionalAccess', _36 => _36.prompt_tokens]) || ""}, completion-tokens, ${_optionalChain([result, 'access', _37 => _37.usage, 'optionalAccess', _38 => _38.completion_tokens]) || ""}, total-tokens, ${_optionalChain([result, 'access', _39 => _39.usage, 'optionalAccess', _40 => _40.total_tokens]) || ""}, cost-ms, ${timeCost}, requestId, ${result._request_id || ""}`
+          `model, ${modelName}, mode, ${_env.vlLocateMode.call(void 0, ) || "default"}, ui-tars-version, ${_env.uiTarsModelVersion.call(void 0, )}, prompt-tokens, ${_optionalChain([result, 'access', _36 => _36.usage, 'optionalAccess', _37 => _37.prompt_tokens]) || ""}, completion-tokens, ${_optionalChain([result, 'access', _38 => _38.usage, 'optionalAccess', _39 => _39.completion_tokens]) || ""}, total-tokens, ${_optionalChain([result, 'access', _40 => _40.usage, 'optionalAccess', _41 => _41.total_tokens]) || ""}, cost-ms, ${timeCost}, requestId, ${result._request_id || ""}`
         );
         debugProfileDetail(
           `model usage detail: ${JSON.stringify(result.usage)}`
@@ -1462,7 +1665,7 @@ async function call2(messages, AIActionTypeValue, responseFormat, options) {
       };
       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",
@@ -1472,7 +1675,7 @@ async function call2(messages, AIActionTypeValue, responseFormat, options) {
           ...commonConfig
         });
         for await (const chunk of stream) {
-          const content2 = _optionalChain([chunk, 'access', _41 => _41.delta, 'optionalAccess', _42 => _42.text]) || "";
+          const content2 = _optionalChain([chunk, 'access', _42 => _42.delta, 'optionalAccess', _43 => _43.text]) || "";
           if (content2) {
             accumulated += content2;
             const chunkData = {
@@ -1506,7 +1709,7 @@ async function call2(messages, AIActionTypeValue, responseFormat, options) {
         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",
@@ -1553,10 +1756,9 @@ async function call2(messages, AIActionTypeValue, responseFormat, options) {
     throw newError;
   }
 }
-async function callToGetJSONObject(messages, AIActionTypeValue) {
+var getResponseFormat = (modelName, AIActionTypeValue) => {
   let responseFormat;
-  const model = getModelName();
-  if (model.includes("gpt-4")) {
+  if (modelName.includes("gpt-4")) {
     switch (AIActionTypeValue) {
       case 0 /* ASSERT */:
         responseFormat = assertSchema;
@@ -1573,10 +1775,18 @@ async function callToGetJSONObject(messages, AIActionTypeValue) {
         break;
     }
   }
-  if (model === "gpt-4o-2024-05-13") {
+  if (modelName === "gpt-4o-2024-05-13") {
     responseFormat = { type: "json_object" /* JSON */ };
   }
-  const response = await call2(messages, AIActionTypeValue, responseFormat);
+  return responseFormat;
+};
+async function callToGetJSONObject(messages, AIActionTypeValue, modelPreferences) {
+  const response = await call2(
+    messages,
+    AIActionTypeValue,
+    void 0,
+    modelPreferences
+  );
   _utils.assert.call(void 0, response, "empty response");
   const jsonContent = safeParseJson(response.content);
   return { content: jsonContent, usage: response.usage };
@@ -1615,8 +1825,8 @@ function preprocessDoubaoBboxJson(input) {
 }
 function safeParseJson(input) {
   const cleanJsonString = extractJSONFromCodeBlock(input);
-  if (_optionalChain([cleanJsonString, 'optionalAccess', _43 => _43.match, 'call', _44 => _44(/\((\d+),(\d+)\)/)])) {
-    return _optionalChain([cleanJsonString, 'access', _45 => _45.match, 'call', _46 => _46(/\((\d+),(\d+)\)/), 'optionalAccess', _47 => _47.slice, 'call', _48 => _48(1), 'access', _49 => _49.map, 'call', _50 => _50(Number)]);
+  if (_optionalChain([cleanJsonString, 'optionalAccess', _44 => _44.match, 'call', _45 => _45(/\((\d+),(\d+)\)/)])) {
+    return _optionalChain([cleanJsonString, 'access', _46 => _46.match, 'call', _47 => _47(/\((\d+),(\d+)\)/), 'optionalAccess', _48 => _48.slice, 'call', _49 => _49(1), 'access', _50 => _50.map, 'call', _51 => _51(Number)]);
   }
   try {
     return JSON.parse(cleanJsonString);
@@ -1800,7 +2010,7 @@ Respond with YAML only, no explanations.`
       });
     }
     const response = await call2(prompt, 2 /* EXTRACT_DATA */);
-    if (_optionalChain([response, 'optionalAccess', _51 => _51.content]) && typeof response.content === "string") {
+    if (_optionalChain([response, 'optionalAccess', _52 => _52.content]) && typeof response.content === "string") {
       return response.content;
     }
     throw new Error("Failed to generate YAML test configuration");
@@ -1861,13 +2071,13 @@ Respond with YAML only, no explanations.`
       });
     }
     if (options.stream && options.onChunk) {
-      return await call2(prompt, 2 /* EXTRACT_DATA */, void 0, {
+      return await call2(prompt, 2 /* EXTRACT_DATA */, {
         stream: true,
         onChunk: options.onChunk
       });
     } else {
       const response = await call2(prompt, 2 /* EXTRACT_DATA */);
-      if (_optionalChain([response, 'optionalAccess', _52 => _52.content]) && typeof response.content === "string") {
+      if (_optionalChain([response, 'optionalAccess', _53 => _53.content]) && typeof response.content === "string") {
         return {
           content: response.content,
           usage: response.usage,
@@ -1930,7 +2140,7 @@ ${_constants.PLAYWRIGHT_EXAMPLE_CODE}`;
     }
   ];
   const response = await call2(prompt, 2 /* EXTRACT_DATA */);
-  if (_optionalChain([response, 'optionalAccess', _53 => _53.content]) && typeof response.content === "string") {
+  if (_optionalChain([response, 'optionalAccess', _54 => _54.content]) && typeof response.content === "string") {
     return response.content;
   }
   throw new Error("Failed to generate Playwright test code");
@@ -1984,13 +2194,13 @@ ${_constants.PLAYWRIGHT_EXAMPLE_CODE}`;
     }
   ];
   if (options.stream && options.onChunk) {
-    return await call2(prompt, 2 /* EXTRACT_DATA */, void 0, {
+    return await call2(prompt, 2 /* EXTRACT_DATA */, {
       stream: true,
       onChunk: options.onChunk
     });
   } else {
     const response = await call2(prompt, 2 /* EXTRACT_DATA */);
-    if (_optionalChain([response, 'optionalAccess', _54 => _54.content]) && typeof response.content === "string") {
+    if (_optionalChain([response, 'optionalAccess', _55 => _55.content]) && typeof response.content === "string") {
       return {
         content: response.content,
         usage: response.usage,
@@ -2180,7 +2390,7 @@ var extraTextFromUserPrompt = (prompt) => {
 };
 var promptsToChatParam = async (multimodalPrompt) => {
   const msgs = [];
-  if (_optionalChain([multimodalPrompt, 'optionalAccess', _55 => _55.images, 'optionalAccess', _56 => _56.length])) {
+  if (_optionalChain([multimodalPrompt, 'optionalAccess', _56 => _56.images, 'optionalAccess', _57 => _57.length])) {
     msgs.push({
       role: "user",
       content: [
@@ -2289,10 +2499,10 @@ async function AiLocateElement(options) {
     if ("bbox" in res.content && Array.isArray(res.content.bbox)) {
       resRect = adaptBboxToRect(
         res.content.bbox,
-        _optionalChain([options, 'access', _57 => _57.searchConfig, 'optionalAccess', _58 => _58.rect, 'optionalAccess', _59 => _59.width]) || context.size.width,
-        _optionalChain([options, 'access', _60 => _60.searchConfig, 'optionalAccess', _61 => _61.rect, 'optionalAccess', _62 => _62.height]) || context.size.height,
-        _optionalChain([options, 'access', _63 => _63.searchConfig, 'optionalAccess', _64 => _64.rect, 'optionalAccess', _65 => _65.left]),
-        _optionalChain([options, 'access', _66 => _66.searchConfig, 'optionalAccess', _67 => _67.rect, 'optionalAccess', _68 => _68.top])
+        _optionalChain([options, 'access', _58 => _58.searchConfig, 'optionalAccess', _59 => _59.rect, 'optionalAccess', _60 => _60.width]) || context.size.width,
+        _optionalChain([options, 'access', _61 => _61.searchConfig, 'optionalAccess', _62 => _62.rect, 'optionalAccess', _63 => _63.height]) || context.size.height,
+        _optionalChain([options, 'access', _64 => _64.searchConfig, 'optionalAccess', _65 => _65.rect, 'optionalAccess', _66 => _66.left]),
+        _optionalChain([options, 'access', _67 => _67.searchConfig, 'optionalAccess', _68 => _68.rect, 'optionalAccess', _69 => _69.top])
       );
       debugInspect("resRect", resRect);
       const rectCenter = {
@@ -2311,7 +2521,7 @@ async function AiLocateElement(options) {
     }
   } catch (e) {
     const msg = e instanceof Error ? `Failed to parse bbox: ${e.message}` : "unknown error in locate";
-    if (!errors || _optionalChain([errors, 'optionalAccess', _69 => _69.length]) === 0) {
+    if (!errors || _optionalChain([errors, 'optionalAccess', _70 => _70.length]) === 0) {
       errors = [msg];
     } else {
       errors.push(`(${msg})`);
@@ -2403,21 +2613,27 @@ async function AiLocateSection(options) {
   };
 }
 async function AiExtractElementInfo(options) {
-  const { dataQuery, context, extractOption, multimodalPrompt } = options;
+  const {
+    dataQuery,
+    context,
+    extractOption,
+    multimodalPrompt,
+    modelPreferences
+  } = options;
   const systemPrompt = systemPromptToExtract();
   const { screenshotBase64 } = context;
   const { description, elementById } = await describeUserPage(context, {
     truncateTextLength: 200,
     filterNonTextContent: false,
     visibleOnly: false,
-    domIncluded: _optionalChain([extractOption, 'optionalAccess', _70 => _70.domIncluded])
+    domIncluded: _optionalChain([extractOption, 'optionalAccess', _71 => _71.domIncluded])
   });
   const extractDataPromptText = await extractDataQueryPrompt(
     description,
     dataQuery
   );
   const userContent = [];
-  if (_optionalChain([extractOption, 'optionalAccess', _71 => _71.screenshotIncluded]) !== false) {
+  if (_optionalChain([extractOption, 'optionalAccess', _72 => _72.screenshotIncluded]) !== false) {
     userContent.push({
       type: "image_url",
       image_url: {
@@ -2437,7 +2653,7 @@ async function AiExtractElementInfo(options) {
       content: userContent
     }
   ];
-  if (_optionalChain([options, 'access', _72 => _72.extractOption, 'optionalAccess', _73 => _73.returnThought])) {
+  if (_optionalChain([options, 'access', _73 => _73.extractOption, 'optionalAccess', _74 => _74.returnThought])) {
     msgs.push({
       role: "user",
       content: "Please provide reasons."
@@ -2452,7 +2668,8 @@ async function AiExtractElementInfo(options) {
   }
   const result = await callAiFn(
     msgs,
-    2 /* EXTRACT_DATA */
+    2 /* EXTRACT_DATA */,
+    modelPreferences
   );
   return {
     parseResult: result.content,
@@ -2566,7 +2783,7 @@ async function plan(userInstruction, opts) {
   const { content, usage } = await call3(msgs, 3 /* PLAN */);
   const rawResponse = JSON.stringify(content, void 0, 2);
   const planFromAI = content;
-  const actions = (_optionalChain([planFromAI, 'access', _74 => _74.action, 'optionalAccess', _75 => _75.type]) ? [planFromAI.action] : planFromAI.actions) || [];
+  const actions = (_optionalChain([planFromAI, 'access', _75 => _75.action, 'optionalAccess', _76 => _76.type]) ? [planFromAI.action] : planFromAI.actions) || [];
   const returnValue = {
     ...planFromAI,
     actions,
@@ -2593,7 +2810,7 @@ async function plan(userInstruction, opts) {
     _utils.assert.call(void 0, !planFromAI.error, `Failed to plan actions: ${planFromAI.error}`);
   } else {
     actions.forEach((action) => {
-      if (_optionalChain([action, 'access', _76 => _76.locate, 'optionalAccess', _77 => _77.id])) {
+      if (_optionalChain([action, 'access', _77 => _77.locate, 'optionalAccess', _78 => _78.id])) {
         const element = elementById(action.locate.id);
         if (element) {
           action.locate.id = element.id;
@@ -2922,4 +3139,4 @@ async function resizeImageForUiTars(imageBase64, size) {
 
 exports.systemPromptToLocateElement = systemPromptToLocateElement; exports.elementByPositionWithElementInfo = elementByPositionWithElementInfo; exports.describeUserPage = describeUserPage; exports.call = call2; exports.callToGetJSONObject = callToGetJSONObject; exports.callAiFnWithStringResponse = callAiFnWithStringResponse; exports.AIActionType = AIActionType; exports.callAiFn = callAiFn; exports.adaptBboxToRect = adaptBboxToRect; exports.expandSearchArea = expandSearchArea; exports.generateYamlTest = generateYamlTest; exports.generateYamlTestStream = generateYamlTestStream; exports.generatePlaywrightTest = generatePlaywrightTest; exports.generatePlaywrightTestStream = generatePlaywrightTestStream; exports.AiLocateElement = AiLocateElement; exports.AiLocateSection = AiLocateSection; exports.AiExtractElementInfo = AiExtractElementInfo; exports.AiAssert = AiAssert; exports.plan = plan; exports.vlmPlanning = vlmPlanning; exports.resizeImageForUiTars = resizeImageForUiTars;
 
-//# sourceMappingURL=chunk-I5LBWOQA.js.map
+//# sourceMappingURL=chunk-G2JTYWI6.js.map