@ax-llm/ax 12.0.19 → 12.0.20

package/index.js

@@ -586,6 +586,16 @@ var apiCall = async (api, json) => {
 import crypto2 from "crypto";
 import { context, SpanKind } from "@opentelemetry/api";
 
+// dsp/globals.ts
+var axGlobals = {
+  signatureStrict: true,
+  // Controls reservedNames enforcement in signature parsing/validation
+  tracer: void 0,
+  // Global OpenTelemetry tracer for all AI operations
+  meter: void 0
+  // Global OpenTelemetry meter for metrics collection
+};
+
 // trace/trace.ts
 var axSpanAttributes = {
   // LLM
@@ -1021,6 +1031,291 @@ var logResponseDelta = (delta, logger = defaultLogger) => {
   logger(delta, { tags: ["responseContent"] });
 };
 
+// ai/metrics.ts
+var createMetricsInstruments = (meter) => {
+  return {
+    latencyHistogram: meter.createHistogram("ax_llm_request_duration_ms", {
+      description: "Duration of LLM requests in milliseconds",
+      unit: "ms"
+    }),
+    errorCounter: meter.createCounter("ax_llm_errors_total", {
+      description: "Total number of LLM request errors"
+    }),
+    requestCounter: meter.createCounter("ax_llm_requests_total", {
+      description: "Total number of LLM requests"
+    }),
+    tokenCounter: meter.createCounter("ax_llm_tokens_total", {
+      description: "Total number of LLM tokens consumed"
+    }),
+    inputTokenCounter: meter.createCounter("ax_llm_input_tokens_total", {
+      description: "Total number of input/prompt tokens consumed"
+    }),
+    outputTokenCounter: meter.createCounter("ax_llm_output_tokens_total", {
+      description: "Total number of output/completion tokens generated"
+    }),
+    errorRateGauge: meter.createGauge("ax_llm_error_rate", {
+      description: "Current error rate as a percentage (0-100)"
+    }),
+    meanLatencyGauge: meter.createGauge("ax_llm_mean_latency_ms", {
+      description: "Mean latency of LLM requests in milliseconds",
+      unit: "ms"
+    }),
+    p95LatencyGauge: meter.createGauge("ax_llm_p95_latency_ms", {
+      description: "95th percentile latency of LLM requests in milliseconds",
+      unit: "ms"
+    }),
+    p99LatencyGauge: meter.createGauge("ax_llm_p99_latency_ms", {
+      description: "99th percentile latency of LLM requests in milliseconds",
+      unit: "ms"
+    }),
+    streamingRequestsCounter: meter.createCounter(
+      "ax_llm_streaming_requests_total",
+      {
+        description: "Total number of streaming LLM requests"
+      }
+    ),
+    functionCallsCounter: meter.createCounter("ax_llm_function_calls_total", {
+      description: "Total number of function/tool calls made"
+    }),
+    functionCallLatencyHistogram: meter.createHistogram(
+      "ax_llm_function_call_latency_ms",
+      {
+        description: "Latency of function calls in milliseconds",
+        unit: "ms"
+      }
+    ),
+    requestSizeHistogram: meter.createHistogram("ax_llm_request_size_bytes", {
+      description: "Size of LLM request payloads in bytes",
+      unit: "By"
+    }),
+    responseSizeHistogram: meter.createHistogram("ax_llm_response_size_bytes", {
+      description: "Size of LLM response payloads in bytes",
+      unit: "By"
+    }),
+    temperatureGauge: meter.createGauge("ax_llm_temperature_gauge", {
+      description: "Temperature setting used for LLM requests"
+    }),
+    maxTokensGauge: meter.createGauge("ax_llm_max_tokens_gauge", {
+      description: "Maximum tokens setting used for LLM requests"
+    }),
+    estimatedCostCounter: meter.createCounter("ax_llm_estimated_cost_total", {
+      description: "Estimated cost of LLM requests in USD",
+      unit: "$"
+    }),
+    promptLengthHistogram: meter.createHistogram("ax_llm_prompt_length_chars", {
+      description: "Length of prompts in characters"
+    }),
+    contextWindowUsageGauge: meter.createGauge(
+      "ax_llm_context_window_usage_ratio",
+      {
+        description: "Context window utilization ratio (0-1)"
+      }
+    ),
+    timeoutsCounter: meter.createCounter("ax_llm_timeouts_total", {
+      description: "Total number of timed out LLM requests"
+    }),
+    abortsCounter: meter.createCounter("ax_llm_aborts_total", {
+      description: "Total number of aborted LLM requests"
+    }),
+    thinkingBudgetUsageCounter: meter.createCounter(
+      "ax_llm_thinking_budget_usage_total",
+      {
+        description: "Total thinking budget tokens used"
+      }
+    ),
+    multimodalRequestsCounter: meter.createCounter(
+      "ax_llm_multimodal_requests_total",
+      {
+        description: "Total number of multimodal requests (with images/audio)"
+      }
+    )
+  };
+};
+var recordLatencyMetric = (instruments, type, duration, aiService, model) => {
+  if (instruments.latencyHistogram) {
+    instruments.latencyHistogram.record(duration, {
+      operation: type,
+      ai_service: aiService,
+      ...model ? { model } : {}
+    });
+  }
+};
+var recordLatencyStatsMetrics = (instruments, type, meanLatency, p95Latency, p99Latency, aiService, model) => {
+  const labels = {
+    operation: type,
+    ai_service: aiService,
+    ...model ? { model } : {}
+  };
+  if (instruments.meanLatencyGauge) {
+    instruments.meanLatencyGauge.record(meanLatency, labels);
+  }
+  if (instruments.p95LatencyGauge) {
+    instruments.p95LatencyGauge.record(p95Latency, labels);
+  }
+  if (instruments.p99LatencyGauge) {
+    instruments.p99LatencyGauge.record(p99Latency, labels);
+  }
+};
+var recordErrorMetric = (instruments, type, aiService, model) => {
+  if (instruments.errorCounter) {
+    instruments.errorCounter.add(1, {
+      operation: type,
+      ai_service: aiService,
+      ...model ? { model } : {}
+    });
+  }
+};
+var recordErrorRateMetric = (instruments, type, errorRate, aiService, model) => {
+  if (instruments.errorRateGauge) {
+    instruments.errorRateGauge.record(errorRate * 100, {
+      // Convert to percentage
+      operation: type,
+      ai_service: aiService,
+      ...model ? { model } : {}
+    });
+  }
+};
+var recordRequestMetric = (instruments, type, aiService, model) => {
+  if (instruments.requestCounter) {
+    instruments.requestCounter.add(1, {
+      operation: type,
+      ai_service: aiService,
+      ...model ? { model } : {}
+    });
+  }
+};
+var recordTokenMetric = (instruments, type, tokens, aiService, model) => {
+  const labels = {
+    ai_service: aiService,
+    ...model ? { model } : {}
+  };
+  if (instruments.tokenCounter) {
+    instruments.tokenCounter.add(tokens, {
+      token_type: type,
+      ...labels
+    });
+  }
+  if (type === "input" && instruments.inputTokenCounter) {
+    instruments.inputTokenCounter.add(tokens, labels);
+  }
+  if (type === "output" && instruments.outputTokenCounter) {
+    instruments.outputTokenCounter.add(tokens, labels);
+  }
+};
+var recordStreamingRequestMetric = (instruments, type, isStreaming, aiService, model) => {
+  if (isStreaming && instruments.streamingRequestsCounter) {
+    instruments.streamingRequestsCounter.add(1, {
+      operation: type,
+      ai_service: aiService,
+      ...model ? { model } : {}
+    });
+  }
+};
+var recordFunctionCallMetric = (instruments, functionName, latency, aiService, model) => {
+  const labels = {
+    function_name: functionName,
+    ...aiService ? { ai_service: aiService } : {},
+    ...model ? { model } : {}
+  };
+  if (instruments.functionCallsCounter) {
+    instruments.functionCallsCounter.add(1, labels);
+  }
+  if (latency && instruments.functionCallLatencyHistogram) {
+    instruments.functionCallLatencyHistogram.record(latency, labels);
+  }
+};
+var recordRequestSizeMetric = (instruments, type, sizeBytes, aiService, model) => {
+  if (instruments.requestSizeHistogram) {
+    instruments.requestSizeHistogram.record(sizeBytes, {
+      operation: type,
+      ai_service: aiService,
+      ...model ? { model } : {}
+    });
+  }
+};
+var recordResponseSizeMetric = (instruments, type, sizeBytes, aiService, model) => {
+  if (instruments.responseSizeHistogram) {
+    instruments.responseSizeHistogram.record(sizeBytes, {
+      operation: type,
+      ai_service: aiService,
+      ...model ? { model } : {}
+    });
+  }
+};
+var recordModelConfigMetrics = (instruments, temperature, maxTokens, aiService, model) => {
+  const labels = {
+    ...aiService ? { ai_service: aiService } : {},
+    ...model ? { model } : {}
+  };
+  if (temperature !== void 0 && instruments.temperatureGauge) {
+    instruments.temperatureGauge.record(temperature, labels);
+  }
+  if (maxTokens !== void 0 && instruments.maxTokensGauge) {
+    instruments.maxTokensGauge.record(maxTokens, labels);
+  }
+};
+var recordEstimatedCostMetric = (instruments, type, costUSD, aiService, model) => {
+  if (instruments.estimatedCostCounter) {
+    instruments.estimatedCostCounter.add(costUSD, {
+      operation: type,
+      ai_service: aiService,
+      ...model ? { model } : {}
+    });
+  }
+};
+var recordPromptLengthMetric = (instruments, lengthChars, aiService, model) => {
+  if (instruments.promptLengthHistogram) {
+    instruments.promptLengthHistogram.record(lengthChars, {
+      ai_service: aiService,
+      ...model ? { model } : {}
+    });
+  }
+};
+var recordContextWindowUsageMetric = (instruments, usageRatio, aiService, model) => {
+  if (instruments.contextWindowUsageGauge) {
+    instruments.contextWindowUsageGauge.record(usageRatio, {
+      ai_service: aiService,
+      ...model ? { model } : {}
+    });
+  }
+};
+var recordTimeoutMetric = (instruments, type, aiService, model) => {
+  if (instruments.timeoutsCounter) {
+    instruments.timeoutsCounter.add(1, {
+      operation: type,
+      ai_service: aiService,
+      ...model ? { model } : {}
+    });
+  }
+};
+var recordAbortMetric = (instruments, type, aiService, model) => {
+  if (instruments.abortsCounter) {
+    instruments.abortsCounter.add(1, {
+      operation: type,
+      ai_service: aiService,
+      ...model ? { model } : {}
+    });
+  }
+};
+var recordThinkingBudgetUsageMetric = (instruments, tokensUsed, aiService, model) => {
+  if (instruments.thinkingBudgetUsageCounter) {
+    instruments.thinkingBudgetUsageCounter.add(tokensUsed, {
+      ai_service: aiService,
+      ...model ? { model } : {}
+    });
+  }
+};
+var recordMultimodalRequestMetric = (instruments, hasImages, hasAudio, aiService, model) => {
+  if ((hasImages || hasAudio) && instruments.multimodalRequestsCounter) {
+    instruments.multimodalRequestsCounter.add(1, {
+      ai_service: aiService,
+      has_images: hasImages.toString(),
+      has_audio: hasAudio.toString(),
+      ...model ? { model } : {}
+    });
+  }
+};
+
 // ai/base.ts
 var axBaseAIDefaultConfig = () => structuredClone({
   temperature: 0,
@@ -1051,7 +1346,8 @@ var AxBaseAI = class {
     this.apiURL = apiURL;
     this.headers = headers;
     this.supportFor = supportFor;
-    this.tracer = options.tracer;
+    this.tracer = options.tracer ?? axGlobals.tracer;
+    this.meter = options.meter ?? axGlobals.meter;
     this.modelInfo = modelInfo;
     this.models = models;
     this.id = crypto2.randomUUID();
@@ -1062,6 +1358,7 @@ var AxBaseAI = class {
       throw new Error("No model defined");
     }
     this.setOptions(options);
+    this.initializeMetricsInstruments();
     if (models) {
       validateModels(models);
     }
@@ -1070,11 +1367,14 @@ var AxBaseAI = class {
   rt;
   fetch;
   tracer;
+  meter;
   timeout;
   excludeContentFromTrace;
   models;
   abortSignal;
   logger = defaultLogger2;
+  // OpenTelemetry metrics instruments
+  metricsInstruments;
   modelInfo;
   modelUsage;
   embedModelUsage;
@@ -1116,6 +1416,11 @@ var AxBaseAI = class {
       }
     }
   };
+  initializeMetricsInstruments() {
+    if (this.meter) {
+      this.metricsInstruments = createMetricsInstruments(this.meter);
+    }
+  }
   setName(name) {
     this.name = name;
   }
@@ -1133,10 +1438,12 @@ var AxBaseAI = class {
     this.rt = options.rateLimiter;
     this.fetch = options.fetch;
     this.timeout = options.timeout;
-    this.tracer = options.tracer;
+    this.tracer = options.tracer ?? axGlobals.tracer;
+    this.meter = options.meter ?? axGlobals.meter;
     this.excludeContentFromTrace = options.excludeContentFromTrace;
     this.abortSignal = options.abortSignal;
     this.logger = options.logger ?? defaultLogger2;
+    this.initializeMetricsInstruments();
   }
   getOptions() {
     return {
@@ -1144,6 +1451,7 @@ var AxBaseAI = class {
       rateLimiter: this.rt,
       fetch: this.fetch,
       tracer: this.tracer,
+      meter: this.meter,
       timeout: this.timeout,
       excludeContentFromTrace: this.excludeContentFromTrace,
       abortSignal: this.abortSignal,
@@ -1208,6 +1516,25 @@ var AxBaseAI = class {
     metrics.mean = metrics.samples.reduce((a, b) => a + b, 0) / metrics.samples.length;
     metrics.p95 = this.calculatePercentile(metrics.samples, 95);
     metrics.p99 = this.calculatePercentile(metrics.samples, 99);
+    if (this.metricsInstruments) {
+      const model = type === "chat" ? this.lastUsedChatModel : this.lastUsedEmbedModel;
+      recordLatencyMetric(
+        this.metricsInstruments,
+        type,
+        duration,
+        this.name,
+        model
+      );
+      recordLatencyStatsMetrics(
+        this.metricsInstruments,
+        type,
+        metrics.mean,
+        metrics.p95,
+        metrics.p99,
+        this.name,
+        model
+      );
+    }
   }
   // Method to update error metrics
   updateErrorMetrics(type, isError) {
@@ -1217,6 +1544,317 @@ var AxBaseAI = class {
       metrics.count++;
     }
     metrics.rate = metrics.count / metrics.total;
+    if (this.metricsInstruments) {
+      const model = type === "chat" ? this.lastUsedChatModel : this.lastUsedEmbedModel;
+      recordRequestMetric(this.metricsInstruments, type, this.name, model);
+      if (isError) {
+        recordErrorMetric(this.metricsInstruments, type, this.name, model);
+      }
+      recordErrorRateMetric(
+        this.metricsInstruments,
+        type,
+        metrics.rate,
+        this.name,
+        model
+      );
+    }
+  }
+  // Method to record token usage metrics
+  recordTokenUsage(modelUsage) {
+    if (this.metricsInstruments && modelUsage?.tokens) {
+      const { promptTokens, completionTokens, totalTokens, thoughtsTokens } = modelUsage.tokens;
+      if (promptTokens) {
+        recordTokenMetric(
+          this.metricsInstruments,
+          "input",
+          promptTokens,
+          this.name,
+          modelUsage.model
+        );
+      }
+      if (completionTokens) {
+        recordTokenMetric(
+          this.metricsInstruments,
+          "output",
+          completionTokens,
+          this.name,
+          modelUsage.model
+        );
+      }
+      if (totalTokens) {
+        recordTokenMetric(
+          this.metricsInstruments,
+          "total",
+          totalTokens,
+          this.name,
+          modelUsage.model
+        );
+      }
+      if (thoughtsTokens) {
+        recordTokenMetric(
+          this.metricsInstruments,
+          "thoughts",
+          thoughtsTokens,
+          this.name,
+          modelUsage.model
+        );
+      }
+    }
+  }
+  // Helper method to calculate request size in bytes
+  calculateRequestSize(req) {
+    try {
+      return new TextEncoder().encode(JSON.stringify(req)).length;
+    } catch {
+      return 0;
+    }
+  }
+  // Helper method to calculate response size in bytes
+  calculateResponseSize(response) {
+    try {
+      return new TextEncoder().encode(JSON.stringify(response)).length;
+    } catch {
+      return 0;
+    }
+  }
+  // Helper method to detect multimodal content
+  detectMultimodalContent(req) {
+    let hasImages = false;
+    let hasAudio = false;
+    if (req.chatPrompt && Array.isArray(req.chatPrompt)) {
+      for (const message of req.chatPrompt) {
+        if (message.role === "user" && Array.isArray(message.content)) {
+          for (const part of message.content) {
+            if (part.type === "image") {
+              hasImages = true;
+            } else if (part.type === "audio") {
+              hasAudio = true;
+            }
+          }
+        }
+      }
+    }
+    return { hasImages, hasAudio };
+  }
+  // Helper method to calculate prompt length
+  calculatePromptLength(req) {
+    let totalLength = 0;
+    if (req.chatPrompt && Array.isArray(req.chatPrompt)) {
+      for (const message of req.chatPrompt) {
+        if (message.role === "system" || message.role === "assistant") {
+          if (message.content) {
+            totalLength += message.content.length;
+          }
+        } else if (message.role === "user") {
+          if (typeof message.content === "string") {
+            totalLength += message.content.length;
+          } else if (Array.isArray(message.content)) {
+            for (const part of message.content) {
+              if (part.type === "text") {
+                totalLength += part.text.length;
+              }
+            }
+          }
+        } else if (message.role === "function") {
+          if (message.result) {
+            totalLength += message.result.length;
+          }
+        }
+      }
+    }
+    return totalLength;
+  }
+  // Helper method to calculate context window usage
+  calculateContextWindowUsage(model, modelUsage) {
+    if (!modelUsage?.tokens?.promptTokens) return 0;
+    const modelInfo = this.modelInfo.find(
+      (info) => info.name === model
+    );
+    if (!modelInfo?.contextWindow) return 0;
+    return modelUsage.tokens.promptTokens / modelInfo.contextWindow;
+  }
+  // Helper method to estimate cost
+  estimateCost(model, modelUsage) {
+    if (!modelUsage?.tokens) return 0;
+    const modelInfo = this.modelInfo.find(
+      (info) => info.name === model
+    );
+    if (!modelInfo || !modelInfo.promptTokenCostPer1M && !modelInfo.completionTokenCostPer1M)
+      return 0;
+    const { promptTokens = 0, completionTokens = 0 } = modelUsage.tokens;
+    const promptCostPer1M = modelInfo.promptTokenCostPer1M || 0;
+    const completionCostPer1M = modelInfo.completionTokenCostPer1M || 0;
+    return promptTokens * promptCostPer1M / 1e6 + completionTokens * completionCostPer1M / 1e6;
+  }
+  // Helper method to estimate cost by model name
+  estimateCostByName(modelName, modelUsage) {
+    if (!modelUsage?.tokens) return 0;
+    const modelInfo = this.modelInfo.find((info) => info.name === modelName);
+    if (!modelInfo || !modelInfo.promptTokenCostPer1M && !modelInfo.completionTokenCostPer1M)
+      return 0;
+    const { promptTokens = 0, completionTokens = 0 } = modelUsage.tokens;
+    const promptCostPer1M = modelInfo.promptTokenCostPer1M || 0;
+    const completionCostPer1M = modelInfo.completionTokenCostPer1M || 0;
+    return promptTokens * promptCostPer1M / 1e6 + completionTokens * completionCostPer1M / 1e6;
+  }
+  // Helper method to record function call metrics
+  recordFunctionCallMetrics(functionCalls, model) {
+    if (!this.metricsInstruments || !functionCalls) return;
+    for (const call of functionCalls) {
+      if (call && typeof call === "object" && "function" in call && call.function && typeof call.function === "object" && "name" in call.function) {
+        recordFunctionCallMetric(
+          this.metricsInstruments,
+          call.function.name,
+          void 0,
+          // latency would need to be tracked separately
+          this.name,
+          model
+        );
+      }
+    }
+  }
+  // Helper method to record timeout metrics
+  recordTimeoutMetric(type) {
+    if (this.metricsInstruments) {
+      const model = type === "chat" ? this.lastUsedChatModel : this.lastUsedEmbedModel;
+      recordTimeoutMetric(this.metricsInstruments, type, this.name, model);
+    }
+  }
+  // Helper method to record abort metrics
+  recordAbortMetric(type) {
+    if (this.metricsInstruments) {
+      const model = type === "chat" ? this.lastUsedChatModel : this.lastUsedEmbedModel;
+      recordAbortMetric(this.metricsInstruments, type, this.name, model);
+    }
+  }
+  // Comprehensive method to record all chat-related metrics
+  recordChatMetrics(req, options, result) {
+    if (!this.metricsInstruments) return;
+    const model = this.lastUsedChatModel;
+    const modelConfig = this.lastUsedModelConfig;
+    const isStreaming = modelConfig?.stream ?? false;
+    recordStreamingRequestMetric(
+      this.metricsInstruments,
+      "chat",
+      isStreaming,
+      this.name,
+      model
+    );
+    const { hasImages, hasAudio } = this.detectMultimodalContent(req);
+    recordMultimodalRequestMetric(
+      this.metricsInstruments,
+      hasImages,
+      hasAudio,
+      this.name,
+      model
+    );
+    const promptLength = this.calculatePromptLength(req);
+    recordPromptLengthMetric(
+      this.metricsInstruments,
+      promptLength,
+      this.name,
+      model
+    );
+    recordModelConfigMetrics(
+      this.metricsInstruments,
+      modelConfig?.temperature,
+      modelConfig?.maxTokens,
+      this.name,
+      model
+    );
+    if (options?.thinkingTokenBudget && this.modelUsage?.tokens?.thoughtsTokens) {
+      recordThinkingBudgetUsageMetric(
+        this.metricsInstruments,
+        this.modelUsage.tokens.thoughtsTokens,
+        this.name,
+        model
+      );
+    }
+    const requestSize = this.calculateRequestSize(req);
+    recordRequestSizeMetric(
+      this.metricsInstruments,
+      "chat",
+      requestSize,
+      this.name,
+      model
+    );
+    if (result && !isStreaming) {
+      const chatResponse = result;
+      const responseSize = this.calculateResponseSize(chatResponse);
+      recordResponseSizeMetric(
+        this.metricsInstruments,
+        "chat",
+        responseSize,
+        this.name,
+        model
+      );
+      if (chatResponse.results) {
+        for (const chatResult of chatResponse.results) {
+          if (chatResult.functionCalls) {
+            this.recordFunctionCallMetrics(
+              chatResult.functionCalls,
+              this.lastUsedChatModel
+            );
+          }
+        }
+      }
+      const contextUsage = this.calculateContextWindowUsage(
+        this.lastUsedChatModel,
+        chatResponse.modelUsage
+      );
+      if (contextUsage > 0) {
+        recordContextWindowUsageMetric(
+          this.metricsInstruments,
+          contextUsage,
+          this.name,
+          model
+        );
+      }
+      const estimatedCost = this.estimateCost(
+        this.lastUsedChatModel,
+        chatResponse.modelUsage
+      );
+      if (estimatedCost > 0) {
+        recordEstimatedCostMetric(
+          this.metricsInstruments,
+          "chat",
+          estimatedCost,
+          this.name,
+          model
+        );
+      }
+    }
+  }
+  // Comprehensive method to record all embed-related metrics
+  recordEmbedMetrics(req, result) {
+    if (!this.metricsInstruments) return;
+    const model = this.lastUsedEmbedModel;
+    const requestSize = this.calculateRequestSize(req);
+    recordRequestSizeMetric(
+      this.metricsInstruments,
+      "embed",
+      requestSize,
+      this.name,
+      model
+    );
+    const responseSize = this.calculateResponseSize(result);
+    recordResponseSizeMetric(
+      this.metricsInstruments,
+      "embed",
+      responseSize,
+      this.name,
+      model
+    );
+    const estimatedCost = this.estimateCostByName(model, result.modelUsage);
+    if (estimatedCost > 0) {
+      recordEstimatedCostMetric(
+        this.metricsInstruments,
+        "embed",
+        estimatedCost,
+        this.name,
+        model
+      );
+    }
   }
   // Public method to get metrics
   getMetrics() {
@@ -1225,16 +1863,27 @@ var AxBaseAI = class {
   async chat(req, options) {
     const startTime = performance.now();
     let isError = false;
+    let result;
     try {
-      const result = await this._chat1(req, options);
+      result = await this._chat1(req, options);
       return result;
     } catch (error) {
       isError = true;
+      if (error instanceof Error) {
+        if (error.message.includes("timeout") || error.name === "TimeoutError") {
+          this.recordTimeoutMetric("chat");
+        } else if (error.message.includes("abort") || error.name === "AbortError") {
+          this.recordAbortMetric("chat");
+        }
+      }
       throw error;
     } finally {
       const duration = performance.now() - startTime;
       this.updateLatencyMetrics("chat", duration);
       this.updateErrorMetrics("chat", isError);
+      if (!isError) {
+        this.recordChatMetrics(req, options, result);
+      }
     }
   }
   async _chat1(req, options) {
@@ -1381,6 +2030,7 @@ var AxBaseAI = class {
           }
         }
         this.modelUsage = res2.modelUsage;
+        this.recordTokenUsage(res2.modelUsage);
         if (span?.isRecording()) {
           setChatResponseEvents(res2, span, this.excludeContentFromTrace);
         }
@@ -1423,6 +2073,7 @@ var AxBaseAI = class {
     }
     if (res.modelUsage) {
       this.modelUsage = res.modelUsage;
+      this.recordTokenUsage(res.modelUsage);
     }
     if (span?.isRecording()) {
       setChatResponseEvents(res, span, this.excludeContentFromTrace);
@@ -1439,15 +2090,27 @@ var AxBaseAI = class {
   async embed(req, options) {
     const startTime = performance.now();
     let isError = false;
+    let result;
     try {
-      return this._embed1(req, options);
+      result = await this._embed1(req, options);
+      return result;
     } catch (error) {
       isError = true;
+      if (error instanceof Error) {
+        if (error.message.includes("timeout") || error.name === "TimeoutError") {
+          this.recordTimeoutMetric("embed");
+        } else if (error.message.includes("abort") || error.name === "AbortError") {
+          this.recordAbortMetric("embed");
+        }
+      }
       throw error;
     } finally {
       const duration = performance.now() - startTime;
       this.updateLatencyMetrics("embed", duration);
       this.updateErrorMetrics("embed", isError);
+      if (!isError) {
+        this.recordEmbedMetrics(req, result);
+      }
     }
   }
   async _embed1(req, options) {
@@ -1522,6 +2185,7 @@ var AxBaseAI = class {
       }
     }
     this.embedModelUsage = res.modelUsage;
+    this.recordTokenUsage(res.modelUsage);
     if (span?.isRecording() && res.modelUsage?.tokens) {
       span.addEvent(axSpanEvents.GEN_AI_USAGE, {
         [axSpanAttributes.LLM_USAGE_INPUT_TOKENS]: res.modelUsage.tokens.promptTokens,
@@ -7884,12 +8548,6 @@ function mergeFunctionCalls(functionCalls, functionCallDeltas) {
 // dsp/sig.ts
 import { createHash } from "crypto";
 
-// dsp/globals.ts
-var axGlobals = {
-  signatureStrict: true
-  // Controls reservedNames enforcement in signature parsing/validation
-};
-
 // dsp/parser.ts
 var SignatureValidationError = class extends Error {
   constructor(message, position, context3, suggestion) {