@livekit/agents 1.0.31 → 1.0.32

package/dist/voice/agent_activity.js

@@ -12,7 +12,7 @@ import { isSameToolChoice, isSameToolContext } from "../llm/tool_context.js";
 import { log } from "../log.js";
 import { DeferredReadableStream } from "../stream/deferred_stream.js";
 import { STT } from "../stt/stt.js";
-import { traceTypes, tracer } from "../telemetry/index.js";
+import { recordRealtimeMetrics, traceTypes, tracer } from "../telemetry/index.js";
 import { splitWords } from "../tokenize/basic/word.js";
 import { TTS } from "../tts/tts.js";
 import { Future, Task, cancelAndWait, waitFor } from "../utils.js";
@@ -46,6 +46,8 @@ class AgentActivity {
   started = false;
   audioRecognition;
   realtimeSession;
+  realtimeSpans;
+  // Maps response_id to OTEL span for metrics recording
   turnDetectionMode;
   logger = log();
   _draining = false;
@@ -134,6 +136,7 @@ class AgentActivity {
       this.agent._agentActivity = this;
       if (this.llm instanceof RealtimeModel) {
         this.realtimeSession = this.llm.session();
+        this.realtimeSpans = /* @__PURE__ */ new Map();
         this.realtimeSession.on("generation_created", (ev) => this.onGenerationCreated(ev));
         this.realtimeSession.on("input_speech_started", (ev) => this.onInputSpeechStarted(ev));
         this.realtimeSession.on("input_speech_stopped", (ev) => this.onInputSpeechStopped(ev));
@@ -346,6 +349,13 @@ class AgentActivity {
     if (speechHandle && (ev.type === "llm_metrics" || ev.type === "tts_metrics")) {
       ev.speechId = speechHandle.id;
     }
+    if (ev.type === "realtime_model_metrics" && this.realtimeSpans) {
+      const span = this.realtimeSpans.get(ev.requestId);
+      if (span) {
+        recordRealtimeMetrics(span, ev);
+        this.realtimeSpans.delete(ev.requestId);
+      }
+    }
     this.agentSession.emit(
       AgentSessionEventTypes.MetricsCollected,
       createMetricsCollectedEvent({ metrics: ev })
@@ -1259,6 +1269,10 @@ ${instructions}` : instructions,
     if (!(this.llm instanceof RealtimeModel)) {
       throw new Error("llm is not a realtime model");
     }
+    span.setAttribute(traceTypes.ATTR_GEN_AI_REQUEST_MODEL, this.llm.model);
+    if (this.realtimeSpans && ev.responseId) {
+      this.realtimeSpans.set(ev.responseId, span);
+    }
     this.logger.debug(
       { speech_id: speechHandle.id, stepIndex: speechHandle.numSteps },
       "realtime generation started"
@@ -1486,7 +1500,12 @@ ${instructions}` : instructions,
       this.agentSession._updateAgentState("thinking");
     });
     await executeToolsTask.result;
-    if (toolOutput.output.length === 0) return;
+    if (toolOutput.output.length === 0) {
+      if (!speechHandle.interrupted) {
+        this.agentSession._updateAgentState("listening");
+      }
+      return;
+    }
     const { maxToolSteps } = this.agentSession.options;
     if (speechHandle.numSteps >= maxToolSteps) {
       this.logger.warn(
@@ -1664,7 +1683,7 @@ ${instructions}` : instructions,
     }
   }
   async close() {
-    var _a, _b, _c;
+    var _a, _b, _c, _d;
     const unlock = await this.lock.lock();
     try {
       if (!this._draining) {
@@ -1694,9 +1713,10 @@ ${instructions}` : instructions,
         this.vad.off("metrics_collected", this.onMetricsCollected);
       }
       this.detachAudioInput();
-      await ((_a = this.realtimeSession) == null ? void 0 : _a.close());
-      await ((_b = this.audioRecognition) == null ? void 0 : _b.close());
-      await ((_c = this._mainTask) == null ? void 0 : _c.cancelAndWait());
+      (_a = this.realtimeSpans) == null ? void 0 : _a.clear();
+      await ((_b = this.realtimeSession) == null ? void 0 : _b.close());
+      await ((_c = this.audioRecognition) == null ? void 0 : _c.close());
+      await ((_d = this._mainTask) == null ? void 0 : _d.cancelAndWait());
     } finally {
       unlock();
     }