@ax-llm/ax 11.0.44 → 11.0.46

package/index.cjs

@@ -46,6 +46,9 @@ __export(index_exports, {
   AxAIGoogleGeminiModel: () => AxAIGoogleGeminiModel,
   AxAIGoogleGeminiSafetyCategory: () => AxAIGoogleGeminiSafetyCategory,
   AxAIGoogleGeminiSafetyThreshold: () => AxAIGoogleGeminiSafetyThreshold,
+  AxAIGrok: () => AxAIGrok,
+  AxAIGrokEmbedModels: () => AxAIGrokEmbedModels,
+  AxAIGrokModel: () => AxAIGrokModel,
   AxAIGroq: () => AxAIGroq,
   AxAIGroqModel: () => AxAIGroqModel,
   AxAIHuggingFace: () => AxAIHuggingFace,
@@ -126,6 +129,8 @@ __export(index_exports, {
   axAIDeepSeekDefaultConfig: () => axAIDeepSeekDefaultConfig,
   axAIGoogleGeminiDefaultConfig: () => axAIGoogleGeminiDefaultConfig,
   axAIGoogleGeminiDefaultCreativeConfig: () => axAIGoogleGeminiDefaultCreativeConfig,
+  axAIGrokBestConfig: () => axAIGrokBestConfig,
+  axAIGrokDefaultConfig: () => axAIGrokDefaultConfig,
   axAIHuggingFaceCreativeConfig: () => axAIHuggingFaceCreativeConfig,
   axAIHuggingFaceDefaultConfig: () => axAIHuggingFaceDefaultConfig,
   axAIMistralBestConfig: () => axAIMistralBestConfig,
@@ -147,6 +152,7 @@ __export(index_exports, {
   axModelInfoCohere: () => axModelInfoCohere,
   axModelInfoDeepSeek: () => axModelInfoDeepSeek,
   axModelInfoGoogleGemini: () => axModelInfoGoogleGemini,
+  axModelInfoGrok: () => axModelInfoGrok,
   axModelInfoGroq: () => axModelInfoGroq,
   axModelInfoHuggingFace: () => axModelInfoHuggingFace,
   axModelInfoMistral: () => axModelInfoMistral,
@@ -1068,15 +1074,16 @@ var AxBaseAI = class {
       ...this.aiImpl.getModelConfig(),
       ...req.modelConfig
     };
+    if (options?.thinkingTokenBudget && !this.getFeatures(model).hasThinkingBudget) {
+      throw new Error(
+        `Model ${model} does not support thinkingTokenBudget.`
+      );
+    }
     modelConfig.stream = (options?.stream !== void 0 ? options.stream : modelConfig.stream) ?? true;
     const canStream = this.getFeatures(model).streaming;
     if (!canStream) {
       modelConfig.stream = false;
     }
-    const canSetThinkingTokenBudget = this.getFeatures(model).thinkingTokenBudget;
-    if (!canSetThinkingTokenBudget && options?.thinkingTokenBudget) {
-      throw new Error("Thinking token budget is not supported for this model");
-    }
     if (this.tracer) {
       return await this.tracer?.startActiveSpan(
         "AI Chat Request",
@@ -1174,6 +1181,13 @@ var AxBaseAI = class {
       const wrappedRespFn = (state) => (resp) => {
         const res2 = respFn(resp, state);
         res2.sessionId = options?.sessionId;
+        if (options?.hideThought) {
+          res2.results.forEach((result) => {
+            if (result.thought) {
+              result.thought = void 0;
+            }
+          });
+        }
         if (!res2.modelUsage) {
           res2.modelUsage = {
             ai: this.name,
@@ -1211,6 +1225,13 @@ var AxBaseAI = class {
     }
     const res = this.aiImpl.createChatResp(rv);
     res.sessionId = options?.sessionId;
+    if (options?.hideThought) {
+      res.results.forEach((result) => {
+        if (result.thought) {
+          result.thought = void 0;
+        }
+      });
+    }
     if (!res.modelUsage) {
       const tokenUsage = this.aiImpl.getTokenUsage();
       if (tokenUsage) {
@@ -1998,25 +2019,29 @@ var axModelInfoOpenAI = [
     name: "o1" /* O1 */,
     currency: "usd",
     promptTokenCostPer1M: 15,
-    completionTokenCostPer1M: 60
+    completionTokenCostPer1M: 60,
+    hasThinkingBudget: true
   },
   {
     name: "o1-mini" /* O1Mini */,
     currency: "usd",
     promptTokenCostPer1M: 1.1,
-    completionTokenCostPer1M: 14.4
+    completionTokenCostPer1M: 14.4,
+    hasThinkingBudget: true
   },
   {
     name: "o3-mini" /* O3Mini */,
     currency: "usd",
     promptTokenCostPer1M: 1.1,
-    completionTokenCostPer1M: 4.4
+    completionTokenCostPer1M: 4.4,
+    hasThinkingBudget: true
   },
   {
     name: "o4-mini" /* O4Mini */,
     currency: "usd",
     promptTokenCostPer1M: 1.1,
-    completionTokenCostPer1M: 4.4
+    completionTokenCostPer1M: 4.4,
+    hasThinkingBudget: true
   },
   {
     name: "gpt-4" /* GPT4 */,
@@ -2106,9 +2131,10 @@ var axAIOpenAIFastConfig = () => ({
   model: "gpt-4.1-mini" /* GPT41Mini */
 });
 var AxAIOpenAIImpl = class {
-  constructor(config, streamingUsage) {
+  constructor(config, streamingUsage, chatReqUpdater) {
     this.config = config;
     this.streamingUsage = streamingUsage;
+    this.chatReqUpdater = chatReqUpdater;
   }
   tokensUsed;
   getTokenUsage() {
@@ -2128,7 +2154,7 @@ var AxAIOpenAIImpl = class {
       stream: config.stream
     };
   }
-  createChatReq(req, _config) {
+  createChatReq(req, config) {
     const model = req.model;
     if (!req.chatPrompt || req.chatPrompt.length === 0) {
       throw new Error("Chat prompt is empty");
@@ -2148,12 +2174,11 @@ var AxAIOpenAIImpl = class {
     const messages = createMessages2(req);
     const frequencyPenalty = req.modelConfig?.frequencyPenalty ?? this.config.frequencyPenalty;
     const stream = req.modelConfig?.stream ?? this.config.stream;
-    const reasoningEffort = isReasoningModel(model) ? this.config.reasoningEffort : void 0;
     const store = this.config.store;
-    const reqValue = {
+    let reqValue = {
       model,
       messages,
-      response_format: this.config?.responseFormat ? { type: this.config?.responseFormat } : void 0,
+      response_format: this.config?.responseFormat ? { type: this.config.responseFormat } : void 0,
       tools,
       tool_choice: toolsChoice,
       max_completion_tokens: req.modelConfig?.maxTokens ?? this.config.maxTokens ?? 500,
@@ -2165,9 +2190,55 @@ var AxAIOpenAIImpl = class {
       logit_bias: this.config.logitBias,
       ...frequencyPenalty ? { frequency_penalty: frequencyPenalty } : {},
       ...stream && this.streamingUsage ? { stream: true, stream_options: { include_usage: true } } : {},
-      ...reasoningEffort ? { reasoning_effort: reasoningEffort } : {},
       ...store ? { store } : {}
     };
+    if (this.config.reasoningEffort) {
+      reqValue.reasoning_effort = this.config.reasoningEffort;
+    }
+    if (this.config.webSearchOptions) {
+      reqValue.web_search_options = {
+        ...this.config.webSearchOptions.searchContextSize && {
+          search_context_size: this.config.webSearchOptions.searchContextSize
+        },
+        ...this.config.webSearchOptions.userLocation && {
+          user_location: {
+            approximate: {
+              type: "approximate",
+              ...this.config.webSearchOptions.userLocation.approximate.city && {
+                city: this.config.webSearchOptions.userLocation.approximate.city
+              },
+              ...this.config.webSearchOptions.userLocation.approximate.country && {
+                country: this.config.webSearchOptions.userLocation.approximate.country
+              },
+              ...this.config.webSearchOptions.userLocation.approximate.region && {
+                region: this.config.webSearchOptions.userLocation.approximate.region
+              },
+              ...this.config.webSearchOptions.userLocation.approximate.timezone && {
+                timezone: this.config.webSearchOptions.userLocation.approximate.timezone
+              }
+            }
+          }
+        }
+      };
+    }
+    if (config.thinkingTokenBudget) {
+      switch (config.thinkingTokenBudget) {
+        case "minimal":
+          reqValue.reasoning_effort = "low";
+          break;
+        case "low":
+          reqValue.reasoning_effort = "medium";
+          break;
+        case "medium":
+          reqValue.reasoning_effort = "high";
+          break;
+        case "high":
+          reqValue.reasoning_effort = "high";
+      }
+    }
+    if (this.chatReqUpdater) {
+      reqValue = this.chatReqUpdater(reqValue);
+    }
     return [apiConfig, reqValue];
   }
   createEmbedReq(req) {
@@ -2210,6 +2281,7 @@ var AxAIOpenAIImpl = class {
       return {
         id: `${choice.index}`,
         content: choice.message.content,
+        thought: choice.message.reasoning_content,
         functionCalls,
         finishReason
       };
@@ -2232,7 +2304,12 @@ var AxAIOpenAIImpl = class {
     }
     const results = choices.map(
       ({
-        delta: { content, role, tool_calls: toolCalls },
+        delta: {
+          content,
+          role,
+          tool_calls: toolCalls,
+          reasoning_content: thought
+        },
         finish_reason: oaiFinishReason
       }) => {
         const finishReason = mapFinishReason2(oaiFinishReason);
@@ -2253,6 +2330,7 @@ var AxAIOpenAIImpl = class {
         return {
           content,
           role,
+          thought,
           functionCalls,
           finishReason,
           id
@@ -2350,14 +2428,17 @@ var AxAIOpenAIBase = class extends AxBaseAI {
     options,
     apiURL,
     modelInfo,
-    models
+    models,
+    chatReqUpdater,
+    supportFor
   }) {
     if (!apiKey || apiKey === "") {
       throw new Error("OpenAI API key not set");
     }
     const aiImpl = new AxAIOpenAIImpl(
       config,
-      options?.streamingUsage ?? true
+      options?.streamingUsage ?? true,
+      chatReqUpdater
     );
     super(aiImpl, {
       name: "OpenAI",
@@ -2369,16 +2450,19 @@ var AxAIOpenAIBase = class extends AxBaseAI {
         embedModel: config.embedModel
       },
       options,
-      supportFor: () => {
-        return { functions: true, streaming: true };
-      },
+      supportFor: supportFor ?? ((model) => {
+        const modelInf = modelInfo.find((m) => m.name === model);
+        return {
+          functions: true,
+          streaming: true,
+          hasThinkingBudget: modelInf?.hasThinkingBudget ?? false,
+          hasShowThoughts: modelInf?.hasShowThoughts ?? false
+        };
+      }),
       models
     });
   }
 };
-var isReasoningModel = (model) => ["o1-mini" /* O1Mini */, "o1" /* O1 */, "o3-mini" /* O3Mini */].includes(
-  model
-);
 var AxAIOpenAI = class extends AxAIOpenAIBase {
   constructor({
     apiKey,
@@ -2389,6 +2473,15 @@ var AxAIOpenAI = class extends AxAIOpenAIBase {
     if (!apiKey || apiKey === "") {
       throw new Error("OpenAI API key not set");
     }
+    const supportForFn = (model) => {
+      const modelInf = axModelInfoOpenAI.find((m) => m.name === model);
+      return {
+        functions: true,
+        streaming: true,
+        hasThinkingBudget: modelInf?.hasThinkingBudget ?? false,
+        hasShowThoughts: modelInf?.hasShowThoughts ?? false
+      };
+    };
     super({
       apiKey,
       config: {
@@ -2397,7 +2490,8 @@ var AxAIOpenAI = class extends AxAIOpenAIBase {
       },
       options,
       modelInfo: axModelInfoOpenAI,
-      models
+      models,
+      supportFor: supportForFn
     });
     super.setName("OpenAI");
   }
@@ -2436,7 +2530,16 @@ var AxAIAzureOpenAI = class extends AxAIOpenAIBase {
       config: _config,
       options,
       models,
-      modelInfo: axModelInfoOpenAI
+      modelInfo: axModelInfoOpenAI,
+      supportFor: (model) => {
+        const modelInf = axModelInfoOpenAI.find((m) => m.name === model);
+        return {
+          functions: true,
+          streaming: true,
+          hasThinkingBudget: modelInf?.hasThinkingBudget ?? false,
+          hasShowThoughts: modelInf?.hasShowThoughts ?? false
+        };
+      }
     });
     const host = resourceName.includes("://") ? resourceName : `https://${resourceName}.openai.azure.com/`;
     super.setName("Azure OpenAI");
@@ -2895,14 +2998,18 @@ var axModelInfoGoogleGemini = [
     currency: "usd",
     characterIsToken: false,
     promptTokenCostPer1M: 2.5,
-    completionTokenCostPer1M: 15
+    completionTokenCostPer1M: 15,
+    hasThinkingBudget: true,
+    hasShowThoughts: true
   },
   {
     name: "gemini-2.5-flash-preview-04-17" /* Gemini25Flash */,
     currency: "usd",
     characterIsToken: false,
     promptTokenCostPer1M: 15,
-    completionTokenCostPer1M: 3.5
+    completionTokenCostPer1M: 3.5,
+    hasThinkingBudget: true,
+    hasShowThoughts: true
   },
   {
     name: "gemini-2.0-flash" /* Gemini20Flash */,
@@ -3163,7 +3270,20 @@ var AxAIGoogleGeminiImpl = class {
       thinkingConfig.thinkingBudget = this.config.thinking.thinkingTokenBudget;
     }
     if (config.thinkingTokenBudget) {
-      thinkingConfig.thinkingBudget = config.thinkingTokenBudget;
+      switch (config.thinkingTokenBudget) {
+        case "minimal":
+          thinkingConfig.thinkingBudget = 0;
+          break;
+        case "low":
+          thinkingConfig.thinkingBudget = 1024;
+          break;
+        case "medium":
+          thinkingConfig.thinkingBudget = 4096;
+          break;
+        case "high":
+          thinkingConfig.thinkingBudget = 8192;
+          break;
+      }
     }
     const generationConfig = {
       maxOutputTokens: req.modelConfig?.maxTokens ?? this.config.maxTokens,
@@ -3364,7 +3484,16 @@ var AxAIGoogleGemini = class extends AxBaseAI {
         embedModel: _config.embedModel
       },
       options,
-      supportFor: { functions: true, streaming: true },
+      supportFor: (model) => {
+        const modelInf = axModelInfoGoogleGemini.find((m) => m.name === model);
+        return {
+          functions: true,
+          streaming: true,
+          hasThinkingBudget: modelInf?.hasThinkingBudget ?? false,
+          hasShowThoughts: modelInf?.hasShowThoughts ?? false,
+          functionCot: false
+        };
+      },
       models
     });
   }
@@ -4129,9 +4258,96 @@ var AxAI = class {
   }
 };
 
+// ai/x-grok/types.ts
+var AxAIGrokModel = /* @__PURE__ */ ((AxAIGrokModel2) => {
+  AxAIGrokModel2["Grok3"] = "grok-3";
+  AxAIGrokModel2["Grok3Mini"] = "grok-3-mini";
+  AxAIGrokModel2["Grok3Fast"] = "grok-3-fast";
+  AxAIGrokModel2["Grok3MiniFast"] = "grok-3-mini-fast";
+  return AxAIGrokModel2;
+})(AxAIGrokModel || {});
+var AxAIGrokEmbedModels = /* @__PURE__ */ ((AxAIGrokEmbedModels3) => {
+  AxAIGrokEmbedModels3["GrokEmbedSmall"] = "grok-embed-small";
+  return AxAIGrokEmbedModels3;
+})(AxAIGrokEmbedModels || {});
+
+// ai/x-grok/info.ts
+var axModelInfoGrok = [
+  {
+    name: "grok-3" /* Grok3 */,
+    currency: "USD",
+    promptTokenCostPer1M: 3,
+    completionTokenCostPer1M: 15
+  },
+  {
+    name: "grok-3-mini" /* Grok3Mini */,
+    currency: "USD",
+    promptTokenCostPer1M: 0.3,
+    completionTokenCostPer1M: 0.5,
+    hasThinkingBudget: true
+  },
+  {
+    name: "grok-3-fast" /* Grok3Fast */,
+    currency: "USD",
+    promptTokenCostPer1M: 5,
+    completionTokenCostPer1M: 25
+  },
+  {
+    name: "grok-3-mini-fast" /* Grok3MiniFast */,
+    currency: "USD",
+    promptTokenCostPer1M: 0.6,
+    completionTokenCostPer1M: 4,
+    hasThinkingBudget: true
+  }
+];
+
+// ai/x-grok/api.ts
+var axAIGrokDefaultConfig = () => structuredClone({
+  model: "grok-3-mini" /* Grok3Mini */,
+  ...axBaseAIDefaultConfig()
+});
+var axAIGrokBestConfig = () => structuredClone({
+  ...axAIGrokDefaultConfig(),
+  model: "grok-3" /* Grok3 */
+});
+var AxAIGrok = class extends AxAIOpenAIBase {
+  constructor({
+    apiKey,
+    config,
+    options,
+    models
+  }) {
+    if (!apiKey || apiKey === "") {
+      throw new Error("Grok API key not set");
+    }
+    const _config = {
+      ...axAIGrokDefaultConfig(),
+      ...config
+    };
+    super({
+      apiKey,
+      config: _config,
+      options,
+      apiURL: "https://api.x.ai/v1",
+      modelInfo: axModelInfoGrok,
+      models,
+      supportFor: (model) => {
+        const modelInf = axModelInfoGrok.find((m) => m.name === model);
+        return {
+          functions: true,
+          streaming: true,
+          hasThinkingBudget: modelInf?.hasThinkingBudget ?? false,
+          hasShowThoughts: modelInf?.hasShowThoughts ?? false
+        };
+      }
+    });
+    super.setName("Grok");
+  }
+};
+
 // dsp/generate.ts
 var import_web5 = require("stream/web");
-var import_api21 = require("@opentelemetry/api");
+var import_api22 = require("@opentelemetry/api");
 
 // ai/util.ts
 function mergeFunctionCalls(functionCalls, functionCallDeltas) {
@@ -6986,7 +7202,7 @@ Content: ${result.content}`
     const traceLabel = options.traceLabel ?? this.options?.traceLabel;
     const spanName = traceLabel ? `${traceLabel} (AxGen)` : "AxGen";
     const span = tracer.startSpan(spanName, {
-      kind: import_api21.SpanKind.SERVER,
+      kind: import_api22.SpanKind.SERVER,
       attributes
     });
     try {
@@ -7861,7 +8077,7 @@ var randomSample = (array, n) => {
 };
 
 // db/base.ts
-var import_api22 = require("@opentelemetry/api");
+var import_api23 = require("@opentelemetry/api");
 var AxDBBase = class {
   name;
   fetch;
@@ -7888,7 +8104,7 @@ var AxDBBase = class {
     return await this.tracer?.startActiveSpan(
       "DB Upsert Request",
       {
-        kind: import_api22.SpanKind.SERVER,
+        kind: import_api23.SpanKind.SERVER,
         attributes: {
           [axSpanAttributes.DB_SYSTEM]: this.name,
           [axSpanAttributes.DB_OPERATION_NAME]: "upsert",
@@ -7922,7 +8138,7 @@ var AxDBBase = class {
     return await this.tracer?.startActiveSpan(
       "DB Batch Upsert Request",
       {
-        kind: import_api22.SpanKind.SERVER,
+        kind: import_api23.SpanKind.SERVER,
         attributes: {
           [axSpanAttributes.DB_SYSTEM]: this.name,
           [axSpanAttributes.DB_OPERATION_NAME]: "upsert",
@@ -7950,7 +8166,7 @@ var AxDBBase = class {
     return await this.tracer?.startActiveSpan(
       "DB Query Request",
       {
-        kind: import_api22.SpanKind.SERVER,
+        kind: import_api23.SpanKind.SERVER,
         attributes: {
           [axSpanAttributes.DB_SYSTEM]: this.name,
           [axSpanAttributes.DB_OPERATION_NAME]: "upsert",
@@ -11764,6 +11980,9 @@ var AxRAG = class extends AxChainOfThought {
   AxAIGoogleGeminiModel,
   AxAIGoogleGeminiSafetyCategory,
   AxAIGoogleGeminiSafetyThreshold,
+  AxAIGrok,
+  AxAIGrokEmbedModels,
+  AxAIGrokModel,
   AxAIGroq,
   AxAIGroqModel,
   AxAIHuggingFace,
@@ -11844,6 +12063,8 @@ var AxRAG = class extends AxChainOfThought {
   axAIDeepSeekDefaultConfig,
   axAIGoogleGeminiDefaultConfig,
   axAIGoogleGeminiDefaultCreativeConfig,
+  axAIGrokBestConfig,
+  axAIGrokDefaultConfig,
   axAIHuggingFaceCreativeConfig,
   axAIHuggingFaceDefaultConfig,
   axAIMistralBestConfig,
@@ -11865,6 +12086,7 @@ var AxRAG = class extends AxChainOfThought {
   axModelInfoCohere,
   axModelInfoDeepSeek,
   axModelInfoGoogleGemini,
+  axModelInfoGrok,
   axModelInfoGroq,
   axModelInfoHuggingFace,
   axModelInfoMistral,