@ax-llm/ax 11.0.34 → 11.0.36

package/index.cjs

@@ -114,6 +114,7 @@ __export(index_exports, {
   AxStringUtil: () => AxStringUtil,
   AxTestPrompt: () => AxTestPrompt,
   axAIAnthropicDefaultConfig: () => axAIAnthropicDefaultConfig,
+  axAIAnthropicVertexDefaultConfig: () => axAIAnthropicVertexDefaultConfig,
   axAIAzureOpenAIBestConfig: () => axAIAzureOpenAIBestConfig,
   axAIAzureOpenAICreativeConfig: () => axAIAzureOpenAICreativeConfig,
   axAIAzureOpenAIDefaultConfig: () => axAIAzureOpenAIDefaultConfig,
@@ -884,8 +885,8 @@ var AxBaseAI = class {
     this.modelInfo = modelInfo;
     this.models = models;
     this.id = crypto.randomUUID();
-    const model = this.models?.find((v) => v.key === defaults.model)?.model ?? defaults.model;
-    const embedModel = defaults.embedModel;
+    const model = this.getModel(defaults.model) ?? defaults.model;
+    const embedModel = this.getEmbedModel(defaults.embedModel);
     this.defaults = { model, embedModel };
     if (!defaults.model || typeof defaults.model !== "string" || defaults.model === "") {
       throw new Error("No model defined");
@@ -951,18 +952,10 @@ var AxBaseAI = class {
     this.headers = headers;
   }
   setOptions(options) {
-    if (options.debug) {
-      this.debug = options.debug;
-    }
-    if (options.rateLimiter) {
-      this.rt = options.rateLimiter;
-    }
-    if (options.fetch) {
-      this.fetch = options.fetch;
-    }
-    if (options.tracer) {
-      this.tracer = options.tracer;
-    }
+    this.debug = options.debug ?? false;
+    this.rt = options.rateLimiter;
+    this.fetch = options.fetch;
+    this.tracer = options.tracer;
   }
   getOptions() {
     return {
@@ -973,17 +966,27 @@ var AxBaseAI = class {
     };
   }
   getModelList() {
-    return this.models?.filter((model) => !model.isInternal)?.map((model) => ({
-      key: model.key,
-      description: model.description,
-      model: model.model
-    }));
-  }
-  getDefaultModels() {
-    return {
-      model: this.defaults.model,
-      embedModel: this.defaults.embedModel
-    };
+    const models = [];
+    for (const model of this.models ?? []) {
+      if (model.isInternal) {
+        continue;
+      }
+      if ("model" in model && model.model) {
+        models.push({
+          key: model.key,
+          description: model.description,
+          model: model.model
+        });
+      }
+      if ("embedModel" in model && model.embedModel) {
+        models.push({
+          key: model.key,
+          description: model.description,
+          embedModel: model.embedModel
+        });
+      }
+    }
+    return models;
   }
   getName() {
     return this.name;
@@ -1038,7 +1041,7 @@ var AxBaseAI = class {
     }
   }
   async _chat1(req, options) {
-    const model = req.model ? this.models?.find((v) => v.key === req.model)?.model ?? req.model : this.defaults.model;
+    const model = this.getModel(req.model) ?? req.model ?? this.defaults.model;
     const modelConfig = {
       ...this.aiImpl.getModelConfig(),
       ...req.modelConfig
@@ -1102,6 +1105,7 @@ var AxBaseAI = class {
     if (!this.aiImpl.createChatReq) {
       throw new Error("generateChatReq not implemented");
     }
+    const debug = options?.debug ?? this.debug;
     let functions;
     if (chatReq.functions && chatReq.functions.length > 0) {
       functions = chatReq.functions.map((fn2) => this.cleanupFunctionSchema(fn2));
@@ -1123,7 +1127,7 @@ var AxBaseAI = class {
           url: this.apiURL,
           headers: await this.buildHeaders(apiConfig.headers),
           stream: modelConfig.stream,
-          debug: this.debug,
+          debug,
           fetch: this.fetch,
           span
         },
@@ -1131,7 +1135,7 @@ var AxBaseAI = class {
       );
       return res2;
     };
-    if (options?.debug ?? this.debug) {
+    if (debug) {
       logChatRequest(req.chatPrompt, options?.debugHideSystemPrompt);
     }
     const rt = options?.rateLimiter ?? this.rt;
@@ -1140,23 +1144,28 @@ var AxBaseAI = class {
       if (!this.aiImpl.createChatStreamResp) {
         throw new Error("generateChatResp not implemented");
       }
-      const respFn = this.aiImpl.createChatStreamResp;
+      const respFn = this.aiImpl.createChatStreamResp.bind(this);
       const wrappedRespFn = (state) => (resp) => {
         const res2 = respFn(resp, state);
         res2.sessionId = options?.sessionId;
-        if (res2.modelUsage) {
-          this.modelUsage = res2.modelUsage;
+        if (!res2.modelUsage) {
+          res2.modelUsage = {
+            ai: this.name,
+            model,
+            tokens: this.aiImpl.getTokenUsage()
+          };
         }
+        this.modelUsage = res2.modelUsage;
         if (span?.isRecording()) {
           setResponseAttr(res2, span);
         }
-        if (options?.debug ?? this.debug) {
+        if (debug) {
           logResponse(res2);
         }
         return res2;
       };
       const doneCb = async (_values) => {
-        if (options?.debug ?? this.debug) {
+        if (debug) {
           process.stdout.write("\n");
         }
       };
@@ -1179,7 +1188,7 @@ var AxBaseAI = class {
     if (span?.isRecording()) {
       setResponseAttr(res, span);
     }
-    if (options?.debug ?? this.debug) {
+    if (debug) {
       logResponse(res);
     }
     span?.end();
@@ -1200,7 +1209,7 @@ var AxBaseAI = class {
     }
   }
   async _embed1(req, options) {
-    const embedModel = req.embedModel ?? this.defaults.embedModel;
+    const embedModel = this.getEmbedModel(req.embedModel) ?? req.embedModel ?? this.defaults.embedModel;
     if (!embedModel) {
       throw new Error("No embed model defined");
     }
@@ -1232,6 +1241,7 @@ var AxBaseAI = class {
     if (!this.aiImpl.createEmbedResp) {
       throw new Error("generateEmbedResp not implemented");
     }
+    const debug = options?.debug ?? this.debug;
     const req = {
       ...embedReq,
       embedModel
@@ -1243,7 +1253,7 @@ var AxBaseAI = class {
           name: apiConfig.name,
           url: this.apiURL,
           headers: await this.buildHeaders(apiConfig.headers),
-          debug: this.debug,
+          debug,
           fetch: this.fetch,
           span
         },
@@ -1251,15 +1261,23 @@ var AxBaseAI = class {
       );
       return res2;
     };
-    const resValue = this.rt ? await this.rt(fn, { embedModelUsage: this.embedModelUsage }) : await fn();
+    const resValue = this.rt ? await this.rt(fn, { modelUsage: this.embedModelUsage }) : await fn();
     const res = this.aiImpl.createEmbedResp(resValue);
     res.sessionId = options?.sessionId;
+    if (!res.modelUsage) {
+      res.modelUsage = {
+        ai: this.name,
+        model: embedModel,
+        tokens: this.aiImpl.getTokenUsage()
+      };
+    }
+    this.embedModelUsage = res.modelUsage;
     if (span?.isRecording()) {
       if (res.modelUsage) {
         this.embedModelUsage = res.modelUsage;
         span.setAttributes({
-          [axSpanAttributes.LLM_USAGE_COMPLETION_TOKENS]: res.modelUsage.completionTokens ?? 0,
-          [axSpanAttributes.LLM_USAGE_PROMPT_TOKENS]: res.modelUsage.promptTokens
+          [axSpanAttributes.LLM_USAGE_COMPLETION_TOKENS]: res.modelUsage.tokens?.completionTokens ?? 0,
+          [axSpanAttributes.LLM_USAGE_PROMPT_TOKENS]: res.modelUsage.tokens?.promptTokens
         });
       }
     }
@@ -1269,12 +1287,27 @@ var AxBaseAI = class {
   async buildHeaders(headers = {}) {
     return { ...headers, ...await this.headers() };
   }
+  getModelByKey(modelName) {
+    if (!modelName) {
+      return void 0;
+    }
+    const item = this.models?.find((v) => v.key === modelName);
+    return item;
+  }
+  getModel(modelName) {
+    const item = this.getModelByKey(modelName);
+    return item && "model" in item ? item.model : void 0;
+  }
+  getEmbedModel(modelName) {
+    const item = this.getModelByKey(modelName);
+    return item && "embedModel" in item ? item.embedModel : void 0;
+  }
 };
 function setResponseAttr(res, span) {
   if (res.modelUsage) {
     span.setAttributes({
-      [axSpanAttributes.LLM_USAGE_COMPLETION_TOKENS]: res.modelUsage.completionTokens ?? 0,
-      [axSpanAttributes.LLM_USAGE_PROMPT_TOKENS]: res.modelUsage.promptTokens
+      [axSpanAttributes.LLM_USAGE_COMPLETION_TOKENS]: res.modelUsage.tokens?.completionTokens ?? 0,
+      [axSpanAttributes.LLM_USAGE_PROMPT_TOKENS]: res.modelUsage.tokens?.promptTokens
     });
   }
 }
@@ -1329,14 +1362,14 @@ var AxAIAnthropicModel = /* @__PURE__ */ ((AxAIAnthropicModel2) => {
   AxAIAnthropicModel2["ClaudeInstant12"] = "claude-instant-1.2";
   return AxAIAnthropicModel2;
 })(AxAIAnthropicModel || {});
-var AxAIAnthropicVertexModel = /* @__PURE__ */ ((AxAIAnthropicVertexModel3) => {
-  AxAIAnthropicVertexModel3["Claude37Sonnet"] = "claude-3-7-sonnet";
-  AxAIAnthropicVertexModel3["Claude35Haiku"] = "claude-3-5-haiku";
-  AxAIAnthropicVertexModel3["Claude35Sonnet"] = "claude-3-5-sonnet";
-  AxAIAnthropicVertexModel3["Claude35SonnetV2"] = "claude-3-5-sonnet-v2";
-  AxAIAnthropicVertexModel3["Claude3Haiku"] = "claude-3-haiku";
-  AxAIAnthropicVertexModel3["Claude3Opus"] = "claude-3-opus";
-  return AxAIAnthropicVertexModel3;
+var AxAIAnthropicVertexModel = /* @__PURE__ */ ((AxAIAnthropicVertexModel2) => {
+  AxAIAnthropicVertexModel2["Claude37Sonnet"] = "claude-3-7-sonnet";
+  AxAIAnthropicVertexModel2["Claude35Haiku"] = "claude-3-5-haiku";
+  AxAIAnthropicVertexModel2["Claude35Sonnet"] = "claude-3-5-sonnet";
+  AxAIAnthropicVertexModel2["Claude35SonnetV2"] = "claude-3-5-sonnet-v2";
+  AxAIAnthropicVertexModel2["Claude3Haiku"] = "claude-3-haiku";
+  AxAIAnthropicVertexModel2["Claude3Opus"] = "claude-3-opus";
+  return AxAIAnthropicVertexModel2;
 })(AxAIAnthropicVertexModel || {});
 
 // ai/anthropic/info.ts
@@ -1390,7 +1423,11 @@ var axModelInfoAnthropic = [
 
 // ai/anthropic/api.ts
 var axAIAnthropicDefaultConfig = () => structuredClone({
-  model: "claude-3-5-sonnet-latest" /* Claude35Sonnet */,
+  model: "claude-3-7-sonnet-latest" /* Claude37Sonnet */,
+  ...axBaseAIDefaultConfig()
+});
+var axAIAnthropicVertexDefaultConfig = () => structuredClone({
+  model: "claude-3-7-sonnet" /* Claude37Sonnet */,
   ...axBaseAIDefaultConfig()
 });
 var AxAIAnthropicImpl = class {
@@ -1398,6 +1435,10 @@ var AxAIAnthropicImpl = class {
     this.config = config;
     this.isVertex = isVertex;
   }
+  tokensUsed;
+  getTokenUsage() {
+    return this.tokensUsed;
+  }
   getModelConfig() {
     const { config } = this;
     return {
@@ -1507,16 +1548,12 @@ var AxAIAnthropicImpl = class {
         finishReason
       };
     });
-    const modelUsage = {
+    this.tokensUsed = {
       promptTokens: resp.usage.input_tokens,
       completionTokens: resp.usage.output_tokens,
       totalTokens: resp.usage.input_tokens + resp.usage.output_tokens
     };
-    return {
-      results,
-      modelUsage,
-      remoteId: resp.id
-    };
+    return { results, remoteId: resp.id };
   };
   createChatStreamResp = (resp, state) => {
     if (!("type" in resp)) {
@@ -1533,15 +1570,12 @@ var AxAIAnthropicImpl = class {
     if (resp.type === "message_start") {
       const { message } = resp;
       const results = [{ content: "", id: message.id }];
-      const modelUsage = {
+      this.tokensUsed = {
         promptTokens: message.usage?.input_tokens ?? 0,
         completionTokens: message.usage?.output_tokens ?? 0,
         totalTokens: (message.usage?.input_tokens ?? 0) + (message.usage?.output_tokens ?? 0)
       };
-      return {
-        results,
-        modelUsage
-      };
+      return { results };
     }
     if (resp.type === "content_block_start") {
       const { content_block: contentBlock } = resp;
@@ -1598,19 +1632,15 @@ var AxAIAnthropicImpl = class {
     }
     if (resp.type === "message_delta") {
       const { delta, usage } = resp;
-      return {
-        results: [
-          {
-            content: "",
-            finishReason: mapFinishReason(delta.stop_reason)
-          }
-        ],
-        modelUsage: {
-          promptTokens: 0,
-          completionTokens: usage.output_tokens,
-          totalTokens: usage.output_tokens
-        }
+      this.tokensUsed = {
+        promptTokens: 0,
+        completionTokens: usage.output_tokens,
+        totalTokens: usage.output_tokens
       };
+      const results = [
+        { content: "", finishReason: mapFinishReason(delta.stop_reason) }
+      ];
+      return { results };
     }
     return {
       results: [{ content: "" }]
@@ -1795,10 +1825,13 @@ function mapFinishReason(stopReason) {
 // ai/openai/types.ts
 var AxAIOpenAIModel = /* @__PURE__ */ ((AxAIOpenAIModel2) => {
   AxAIOpenAIModel2["O1"] = "o1";
+  AxAIOpenAIModel2["O3"] = "o3";
   AxAIOpenAIModel2["O1Mini"] = "o1-mini";
   AxAIOpenAIModel2["O3Mini"] = "o3-mini";
+  AxAIOpenAIModel2["O4Mini"] = "o4-mini";
   AxAIOpenAIModel2["GPT4"] = "gpt-4";
-  AxAIOpenAIModel2["GPT45"] = "gpt-4.5-preview";
+  AxAIOpenAIModel2["GPT41"] = "gpt-4.1";
+  AxAIOpenAIModel2["GPT41Mini"] = "gpt-4.1-mini";
   AxAIOpenAIModel2["GPT4O"] = "gpt-4o";
   AxAIOpenAIModel2["GPT4OMini"] = "gpt-4o-mini";
   AxAIOpenAIModel2["GPT4ChatGPT4O"] = "chatgpt-4o-latest";
@@ -1819,12 +1852,6 @@ var AxAIOpenAIEmbedModel = /* @__PURE__ */ ((AxAIOpenAIEmbedModel2) => {
 
 // ai/openai/info.ts
 var axModelInfoOpenAI = [
-  {
-    name: "gpt-4.5-preview" /* GPT45 */,
-    currency: "usd",
-    promptTokenCostPer1M: 75,
-    completionTokenCostPer1M: 150
-  },
   {
     name: "o1" /* O1 */,
     currency: "usd",
@@ -1843,12 +1870,30 @@ var axModelInfoOpenAI = [
     promptTokenCostPer1M: 1.1,
     completionTokenCostPer1M: 4.4
   },
+  {
+    name: "o4-mini" /* O4Mini */,
+    currency: "usd",
+    promptTokenCostPer1M: 1.1,
+    completionTokenCostPer1M: 4.4
+  },
   {
     name: "gpt-4" /* GPT4 */,
     currency: "usd",
     promptTokenCostPer1M: 30,
     completionTokenCostPer1M: 60
   },
+  {
+    name: "gpt-4.1" /* GPT41 */,
+    currency: "usd",
+    promptTokenCostPer1M: 2,
+    completionTokenCostPer1M: 8
+  },
+  {
+    name: "gpt-4.1-mini" /* GPT41Mini */,
+    currency: "usd",
+    promptTokenCostPer1M: 0.4,
+    completionTokenCostPer1M: 1.6
+  },
   {
     name: "gpt-4o" /* GPT4O */,
     currency: "usd",
@@ -1901,28 +1946,32 @@ var axModelInfoOpenAI = [
 
 // ai/openai/api.ts
 var axAIOpenAIDefaultConfig = () => structuredClone({
-  model: "gpt-4o" /* GPT4O */,
+  model: "gpt-4.1" /* GPT41 */,
   embedModel: "text-embedding-3-small" /* TextEmbedding3Small */,
   ...axBaseAIDefaultConfig()
 });
 var axAIOpenAIBestConfig = () => structuredClone({
   ...axAIOpenAIDefaultConfig(),
-  model: "gpt-4o" /* GPT4O */
+  model: "gpt-4.1" /* GPT41 */
 });
 var axAIOpenAICreativeConfig = () => structuredClone({
-  model: "gpt-4o" /* GPT4O */,
+  model: "gpt-4.1" /* GPT41 */,
   embedModel: "text-embedding-3-small" /* TextEmbedding3Small */,
   ...axBaseAIDefaultCreativeConfig()
 });
 var axAIOpenAIFastConfig = () => ({
   ...axAIOpenAIDefaultConfig(),
-  model: "gpt-4o-mini" /* GPT4OMini */
+  model: "gpt-4.1-mini" /* GPT41Mini */
 });
 var AxAIOpenAIImpl = class {
   constructor(config, streamingUsage) {
     this.config = config;
     this.streamingUsage = streamingUsage;
   }
+  tokensUsed;
+  getTokenUsage() {
+    return this.tokensUsed;
+  }
   getModelConfig() {
     const { config } = this;
     return {
@@ -2002,7 +2051,7 @@ var AxAIOpenAIImpl = class {
     if (error) {
       throw error;
     }
-    const modelUsage = usage ? {
+    this.tokensUsed = usage ? {
       promptTokens: usage.prompt_tokens,
       completionTokens: usage.completion_tokens,
       totalTokens: usage.total_tokens
@@ -2024,14 +2073,13 @@ var AxAIOpenAIImpl = class {
       };
     });
     return {
-      modelUsage,
       results,
       remoteId: id
     };
   }
   createChatStreamResp(resp, state) {
     const { id, usage, choices } = resp;
-    const modelUsage = usage ? {
+    this.tokensUsed = usage ? {
       promptTokens: usage.prompt_tokens,
       completionTokens: usage.completion_tokens,
       totalTokens: usage.total_tokens
@@ -2069,22 +2117,16 @@ var AxAIOpenAIImpl = class {
         };
       }
     );
-    return {
-      results,
-      modelUsage
-    };
+    return { results };
   }
   createEmbedResp(resp) {
     const { data, usage } = resp;
-    const modelUsage = usage ? {
+    this.tokensUsed = usage ? {
       promptTokens: usage.prompt_tokens,
       completionTokens: usage.completion_tokens,
       totalTokens: usage.total_tokens
     } : void 0;
-    return {
-      embeddings: data.map((v) => v.embedding),
-      modelUsage
-    };
+    return { embeddings: data.map((v) => v.embedding) };
   }
 };
 var mapFinishReason2 = (finishReason) => {
@@ -2349,6 +2391,10 @@ var AxAICohereImpl = class {
   constructor(config) {
     this.config = config;
   }
+  tokensUsed;
+  getTokenUsage() {
+    return this.tokensUsed;
+  }
   getModelConfig() {
     const { config } = this;
     return {
@@ -2440,7 +2486,7 @@ var AxAICohereImpl = class {
     return [apiConfig, reqValue];
   };
   createChatResp = (resp) => {
-    const modelUsage = resp.meta.billed_units ? {
+    this.tokensUsed = resp.meta.billed_units ? {
       promptTokens: resp.meta.billed_units.input_tokens,
       completionTokens: resp.meta.billed_units.output_tokens,
       totalTokens: resp.meta.billed_units.input_tokens + resp.meta.billed_units.output_tokens
@@ -2483,17 +2529,18 @@ var AxAICohereImpl = class {
         finishReason
       }
     ];
-    return {
-      results,
-      modelUsage,
-      remoteId: resp.response_id
-    };
+    return { results, remoteId: resp.response_id };
   };
   createChatStreamResp = (resp, state) => {
     const ss = state;
     if (resp.event_type === "stream-start") {
       ss.generation_id = resp.generation_id;
     }
+    this.tokensUsed = {
+      promptTokens: 0,
+      completionTokens: resp.meta.billed_units?.output_tokens ?? 0,
+      totalTokens: resp.meta.billed_units?.output_tokens ?? 0
+    };
     const { results } = this.createChatResp(resp);
     const result = results[0];
     if (!result) {
@@ -2653,8 +2700,8 @@ var AxAIDeepSeek = class extends AxAIOpenAIBase {
 
 // ai/google-gemini/types.ts
 var AxAIGoogleGeminiModel = /* @__PURE__ */ ((AxAIGoogleGeminiModel2) => {
-  AxAIGoogleGeminiModel2["Gemini25Pro"] = "gemini-2.5-pro-exp-03-25";
-  AxAIGoogleGeminiModel2["Gemini20Pro"] = "gemini-2.0-pro-exp-02-05";
+  AxAIGoogleGeminiModel2["Gemini25Pro"] = "gemini-2.5-pro-preview-03-25";
+  AxAIGoogleGeminiModel2["Gemini25Flash"] = "gemini-2.5-flash-preview-04-17";
   AxAIGoogleGeminiModel2["Gemini20Flash"] = "gemini-2.0-flash";
   AxAIGoogleGeminiModel2["Gemini20FlashLite"] = "gemini-2.0-flash-lite-preview-02-05";
   AxAIGoogleGeminiModel2["Gemini20FlashThinking"] = "gemini-2.0-flash-thinking-exp-01-21";
@@ -2702,11 +2749,18 @@ var AxAIGoogleGeminiEmbedTypes = /* @__PURE__ */ ((AxAIGoogleGeminiEmbedTypes2)
 // ai/google-gemini/info.ts
 var axModelInfoGoogleGemini = [
   {
-    name: "gemini-2.0-pro-exp-02-05" /* Gemini20Pro */,
+    name: "gemini-2.5-pro-preview-03-25" /* Gemini25Pro */,
     currency: "usd",
     characterIsToken: false,
-    promptTokenCostPer1M: 0,
-    completionTokenCostPer1M: 0
+    promptTokenCostPer1M: 2.5,
+    completionTokenCostPer1M: 15
+  },
+  {
+    name: "gemini-2.5-flash-preview-04-17" /* Gemini25Flash */,
+    currency: "usd",
+    characterIsToken: false,
+    promptTokenCostPer1M: 15,
+    completionTokenCostPer1M: 3.5
   },
   {
     name: "gemini-2.0-flash" /* Gemini20Flash */,
@@ -2780,13 +2834,13 @@ var safetySettings = [
 ];
 var axAIGoogleGeminiDefaultConfig = () => structuredClone({
   model: "gemini-2.0-flash" /* Gemini20Flash */,
-  embedModel: "text-embedding-004" /* TextEmbedding004 */,
+  embedModel: "text-embedding-005" /* TextEmbedding005 */,
   safetySettings,
   ...axBaseAIDefaultConfig()
 });
 var axAIGoogleGeminiDefaultCreativeConfig = () => structuredClone({
   model: "gemini-2.0-flash" /* Gemini20Flash */,
-  embedModel: "text-embedding-004" /* TextEmbedding004 */,
+  embedModel: "text-embedding-005" /* TextEmbedding005 */,
   safetySettings,
   ...axBaseAIDefaultCreativeConfig()
 });
@@ -2801,6 +2855,10 @@ var AxAIGoogleGeminiImpl = class {
       throw new Error("Auto truncate is not supported for GoogleGemini");
     }
   }
+  tokensUsed;
+  getTokenUsage() {
+    return this.tokensUsed;
+  }
   getModelConfig() {
     const { config } = this;
     return {
@@ -2960,7 +3018,10 @@ var AxAIGoogleGeminiImpl = class {
       frequencyPenalty: req.modelConfig?.frequencyPenalty ?? this.config.frequencyPenalty,
       candidateCount: 1,
       stopSequences: req.modelConfig?.stopSequences ?? this.config.stopSequences,
-      responseMimeType: "text/plain"
+      responseMimeType: "text/plain",
+      ...this.config.thinkingConfig && {
+        thinkingConfig: this.config.thinkingConfig
+      }
     };
     const safetySettings2 = this.config.safetySettings;
     const reqValue = {
@@ -3060,18 +3121,14 @@ var AxAIGoogleGeminiImpl = class {
         return result;
       }
     );
-    let modelUsage;
     if (resp.usageMetadata) {
-      modelUsage = {
+      this.tokensUsed = {
         totalTokens: resp.usageMetadata.totalTokenCount,
         promptTokens: resp.usageMetadata.promptTokenCount,
         completionTokens: resp.usageMetadata.candidatesTokenCount
       };
     }
-    return {
-      results,
-      modelUsage
-    };
+    return { results };
   };
   createChatStreamResp = (resp) => {
     return this.createChatResp(resp);
@@ -3289,7 +3346,7 @@ var AxAIGroq = class extends AxAIOpenAIBase {
       debug: options?.debug
     });
     const rtFunc = async (func, info) => {
-      const totalTokens = info.modelUsage?.totalTokens || 0;
+      const totalTokens = info.modelUsage?.tokens?.totalTokens || 0;
       await rt.acquire(totalTokens);
       return await func();
     };
@@ -3319,6 +3376,10 @@ var AxAIHuggingFaceImpl = class {
   constructor(config) {
     this.config = config;
   }
+  tokensUsed;
+  getTokenUsage() {
+    return this.tokensUsed;
+  }
   getModelConfig() {
     const { config } = this;
     return {
@@ -3612,6 +3673,10 @@ var AxAIRekaImpl = class {
   constructor(config) {
     this.config = config;
   }
+  tokensUsed;
+  getTokenUsage() {
+    return this.tokensUsed;
+  }
   getModelConfig() {
     const { config } = this;
     return {
@@ -3652,7 +3717,7 @@ var AxAIRekaImpl = class {
   };
   createChatResp = (resp) => {
     const { id, usage, responses } = resp;
-    const modelUsage = usage ? {
+    this.tokensUsed = usage ? {
       promptTokens: usage.input_tokens,
       completionTokens: usage.output_tokens,
       totalTokens: usage.input_tokens + usage.output_tokens
@@ -3671,15 +3736,11 @@ var AxAIRekaImpl = class {
         finishReason
       };
     });
-    return {
-      modelUsage,
-      results,
-      remoteId: id
-    };
+    return { results, remoteId: id };
   };
   createChatStreamResp = (resp) => {
     const { id, usage, responses } = resp;
-    const modelUsage = usage ? {
+    this.tokensUsed = usage ? {
       promptTokens: usage.input_tokens,
       completionTokens: usage.output_tokens,
       totalTokens: usage.input_tokens + usage.output_tokens
@@ -3698,10 +3759,7 @@ var AxAIRekaImpl = class {
         finishReason
       };
     });
-    return {
-      results,
-      modelUsage
-    };
+    return { results };
   };
 };
 var mapFinishReason3 = (finishReason) => {
@@ -3887,9 +3945,6 @@ var AxAI = class {
   getModelList() {
     return this.ai.getModelList();
   }
-  getDefaultModels() {
-    return this.ai.getDefaultModels();
-  }
   getMetrics() {
     return this.ai.getMetrics();
   }
@@ -4826,9 +4881,15 @@ function mergeProgramUsage(usages) {
     }
     const currentUsage = usageMap[key];
     if (currentUsage) {
-      currentUsage.promptTokens += usage.promptTokens;
-      currentUsage.completionTokens += usage.completionTokens;
-      currentUsage.totalTokens += usage.totalTokens;
+      const tokens = currentUsage.tokens ?? {
+        promptTokens: 0,
+        completionTokens: 0,
+        totalTokens: 0
+      };
+      tokens.promptTokens += usage?.tokens?.promptTokens ?? 0;
+      tokens.completionTokens += usage?.tokens?.completionTokens ?? 0;
+      tokens.totalTokens += usage?.tokens?.totalTokens ?? 0;
+      currentUsage.tokens = tokens;
     }
   }
   return Object.values(usageMap);
@@ -6348,13 +6409,9 @@ var AxGen = class extends AxProgramWithSignature {
     mem,
     options
   }) {
-    const { sessionId, traceId, model, functions: _functions } = options ?? {};
+    const { sessionId, traceId, functions: _functions } = options ?? {};
     const fastFail = options?.fastFail ?? this.options?.fastFail;
-    const modelName = model ?? ai.getDefaultModels().model;
-    const usageInfo = {
-      ai: ai.getName(),
-      model: modelName
-    };
+    const model = options.model;
     const functions = _functions?.map((f) => "toFunction" in f ? f.toFunction() : f)?.flat();
     const res = await this.forwardSendRequest({
       ai,
@@ -6366,7 +6423,6 @@ var AxGen = class extends AxProgramWithSignature {
         ai,
         model,
         res,
-        usageInfo,
         mem,
         traceId,
         sessionId,
@@ -6378,7 +6434,6 @@ var AxGen = class extends AxProgramWithSignature {
         ai,
         model,
         res,
-        usageInfo,
         mem,
         traceId,
         sessionId,
@@ -6390,14 +6445,13 @@ var AxGen = class extends AxProgramWithSignature {
     ai,
     model,
     res,
-    usageInfo,
     mem,
     sessionId,
     traceId,
     functions,
     fastFail
   }) {
-    const streamingValidation = fastFail ?? ai.getFeatures().functionCot !== true;
+    const streamingValidation = fastFail ?? ai.getFeatures(model).functionCot !== true;
     const functionCalls = [];
     const values = {};
     const xstate = {
@@ -6412,7 +6466,7 @@ var AxGen = class extends AxProgramWithSignature {
         continue;
       }
       if (v.modelUsage) {
-        this.usage.push({ ...usageInfo, ...v.modelUsage });
+        this.usage.push(v.modelUsage);
       }
       if (result.functionCalls) {
         mergeFunctionCalls(functionCalls, result.functionCalls);
@@ -6513,7 +6567,6 @@ var AxGen = class extends AxProgramWithSignature {
   async processResponse({
     ai,
     res,
-    usageInfo,
     mem,
     sessionId,
     traceId,
@@ -6526,7 +6579,7 @@ var AxGen = class extends AxProgramWithSignature {
     }
     for (const result of results) {
       if (res.modelUsage) {
-        this.usage.push({ ...usageInfo, ...res.modelUsage });
+        this.usage.push(res.modelUsage);
       }
       mem.addResult(result, sessionId);
       if (result.functionCalls?.length) {
@@ -6574,10 +6627,8 @@ var AxGen = class extends AxProgramWithSignature {
     const maxRetries = options.maxRetries ?? this.options?.maxRetries ?? 10;
     const maxSteps = options.maxSteps ?? this.options?.maxSteps ?? 10;
     const debug = options.debug ?? ai.getOptions().debug;
-    const memOptions = {
-      debug: options.debug,
-      debugHideSystemPrompt: options.debugHideSystemPrompt
-    };
+    const debugHideSystemPrompt = options.debugHideSystemPrompt;
+    const memOptions = { debug, debugHideSystemPrompt };
     const mem = options.mem ?? this.options?.mem ?? new AxMemory(1e4, memOptions);
     let err;
     if (options?.functions && options.functions.length > 0) {
@@ -6655,7 +6706,7 @@ var AxGen = class extends AxProgramWithSignature {
     return false;
   }
   async *_forward1(ai, values, options) {
-    const tracer = this.options?.tracer ?? options?.tracer;
+    const tracer = options?.tracer ?? this.options?.tracer;
     let functions = this.functions;
     if (options?.functions) {
       functions = parseFunctions(options.functions, this.functions);
@@ -6691,9 +6742,7 @@ var AxGen = class extends AxProgramWithSignature {
     }
   }
   async forward(ai, values, options) {
-    const generator = this._forward1(ai, values, {
-      ...options
-    });
+    const generator = this._forward1(ai, values, options ?? {});
     let buffer = {};
     let currentVersion = 0;
     for await (const item of generator) {
@@ -7122,9 +7171,6 @@ var AxBalancer = class _AxBalancer {
   getModelList() {
     return this.currentService.getModelList();
   }
-  getDefaultModels() {
-    return this.currentService.getDefaultModels();
-  }
   getNextService() {
     const cs = this.services[++this.currentServiceIndex];
     if (cs === void 0) {
@@ -9276,12 +9322,6 @@ var AxMockAIService = class {
   getModelList() {
     return this.config.models;
   }
-  getDefaultModels() {
-    return {
-      model: this.config.modelInfo?.name ?? "mock-model",
-      embedModel: this.config.embedModelInfo?.name
-    };
-  }
   getMetrics() {
     return this.metrics;
   }
@@ -9304,9 +9344,13 @@ var AxMockAIService = class {
         }
       ],
       modelUsage: {
-        promptTokens: 10,
-        completionTokens: 5,
-        totalTokens: 15
+        ai: this.getName(),
+        model: "mock-model",
+        tokens: {
+          promptTokens: 10,
+          completionTokens: 5,
+          totalTokens: 15
+        }
       }
     };
   }
@@ -9324,9 +9368,13 @@ var AxMockAIService = class {
     return this.config.embedResponse ?? {
       embeddings: [[0.1, 0.2, 0.3]],
       modelUsage: {
-        promptTokens: 5,
-        completionTokens: 0,
-        totalTokens: 5
+        ai: this.getName(),
+        model: "mock-model",
+        tokens: {
+          promptTokens: 5,
+          completionTokens: 0,
+          totalTokens: 5
+        }
       }
     };
   }
@@ -9357,7 +9405,7 @@ var AxMockAIService = class {
   }
 };
 
-// dsp/router.ts
+// dsp/classifier.ts
 var colorLog6 = new ColorLog();
 var AxSimpleClassifierClass = class {
   name;
@@ -11166,6 +11214,7 @@ var AxMCPStdioTransport = class {
 
 // ai/multiservice.ts
 var AxMultiServiceRouter = class {
+  options;
   services = /* @__PURE__ */ new Map();
   /**
    * Constructs a new multi-service router.
@@ -11186,9 +11235,7 @@ var AxMultiServiceRouter = class {
         this.services.set(item.key, {
           service,
           description,
-          isInternal,
-          model: item.service.getDefaultModels().model,
-          useDefaultModel: true
+          isInternal
         });
       } else {
         const modelList = item.getModelList();
@@ -11197,18 +11244,31 @@ var AxMultiServiceRouter = class {
             `Service ${index} \`${item.getName()}\` has no model list.`
           );
         }
-        for (const { key, description, model } of modelList ?? []) {
-          if (this.services.has(key)) {
-            const otherService = this.services.get(key)?.service;
+        for (const v of modelList) {
+          if (this.services.has(v.key)) {
+            const otherService = this.services.get(v.key)?.service;
             throw new Error(
-              `Service ${index} \`${item.getName()}\` has duplicate model key: ${key} as service ${otherService?.getName()}`
+              `Service ${index} \`${item.getName()}\` has duplicate model key: ${v.key} as service ${otherService?.getName()}`
             );
+          } else {
+            if ("model" in v && typeof v.model) {
+              this.services.set(v.key, {
+                description: v.description,
+                service: item,
+                model: v.model
+              });
+            } else if ("embedModel" in v && v.embedModel) {
+              this.services.set(v.key, {
+                description: v.description,
+                service: item,
+                embedModel: v.embedModel
+              });
+            } else {
+              throw new Error(
+                `Key ${v.key} in model list for service ${index} \`${item.getName()}\` is missing a model or embedModel property.`
+              );
+            }
           }
-          this.services.set(key, {
-            description,
-            service: item,
-            model
-          });
         }
       }
     }
@@ -11225,25 +11285,32 @@ var AxMultiServiceRouter = class {
     if (!item) {
       throw new Error(`No service found for model key: ${modelKey}`);
     }
-    const service = item.service;
-    const model = item.useDefaultModel ? req.model : modelKey;
-    return await service.chat({ model, ...req }, options);
+    if (!item.model) {
+      const { model, ...reqWithoutModel } = req;
+      return await item.service.chat(reqWithoutModel, options);
+    }
+    return await item.service.chat({ model: modelKey, ...req }, options);
   }
   /**
    * Delegates the embed call to the service matching the provided embed model key.
    */
   async embed(req, options) {
-    const modelKey = req.embedModel;
-    if (!modelKey) {
+    const embedModelKey = req.embedModel;
+    if (!embedModelKey) {
       throw new Error("Embed model key must be specified for multi-service");
     }
-    const item = this.services.get(modelKey);
+    const item = this.services.get(embedModelKey);
     if (!item) {
-      throw new Error(`No service found for embed model key: ${modelKey}`);
+      throw new Error(`No service found for embed model key: ${embedModelKey}`);
     }
-    const service = item.service;
-    const embedModel = item.useDefaultModel ? req.embedModel : modelKey;
-    return await service.embed({ embedModel, ...req }, options);
+    if (!item.model) {
+      const { embedModel, ...reqWithoutEmbedModel } = req;
+      return await item.service.embed(reqWithoutEmbedModel, options);
+    }
+    return await item.service.embed(
+      { embedModel: embedModelKey, ...req },
+      options
+    );
   }
   /**
    * Returns a composite ID built from the IDs of the underlying services.
@@ -11261,16 +11328,15 @@ var AxMultiServiceRouter = class {
    * Aggregates all available models across the underlying services.
    */
   getModelList() {
-    return Array.from(this.services).filter(([, value]) => !value.isInternal).map(([key, { description, model }]) => ({
-      key,
-      description,
-      model
-    }));
-  }
-  getDefaultModels() {
-    throw new Error(
-      "getDefaultModels is not supported for multi-service router."
-    );
+    return Array.from(this.services).filter(([, value]) => !value.isInternal).map(([key, v]) => {
+      if (v.model) {
+        return { key, description: v.description, model: v.model };
+      } else if (v.embedModel) {
+        return { key, description: v.description, embedModel: v.embedModel };
+      } else {
+        throw new Error(`Service ${key} has no model or embedModel`);
+      }
+    });
   }
   /**
    * If a model key is provided, delegate to the corresponding service's features.
@@ -11304,17 +11370,14 @@ var AxMultiServiceRouter = class {
     for (const service of this.services.values()) {
       service.service.setOptions(options);
     }
+    this.options = options;
   }
   /**
    * Returns the options from the last used service,
    * or falls back to the first service if none has been used.
    */
   getOptions() {
-    const service = this.services.values().next().value;
-    if (!service) {
-      throw new Error("No service available to get options.");
-    }
-    return service.service.getOptions();
+    return this.options ?? {};
   }
 };
 
@@ -11437,6 +11500,7 @@ var AxRAG = class extends AxChainOfThought {
   AxStringUtil,
   AxTestPrompt,
   axAIAnthropicDefaultConfig,
+  axAIAnthropicVertexDefaultConfig,
   axAIAzureOpenAIBestConfig,
   axAIAzureOpenAICreativeConfig,
   axAIAzureOpenAIDefaultConfig,