diff --git a/py/src/braintrust/integrations/pipecat/test_pipecat.py b/py/src/braintrust/integrations/pipecat/test_pipecat.py
index 8c9064e1f..cc25a21e9 100644
--- a/py/src/braintrust/integrations/pipecat/test_pipecat.py
+++ b/py/src/braintrust/integrations/pipecat/test_pipecat.py
@@ -6,6 +6,7 @@
 import os
 import tempfile
 from pathlib import Path
+from types import SimpleNamespace
 
 import pytest
 from braintrust import SpanCustomizer, logger, set_span_customizers
@@ -58,6 +59,33 @@ def _single_span(logs, name):
     return matches[0]
 
 
+def test_pipecat_observer_filters_metrics_from_other_processors():
+    LLMTokenUsage = _import("pipecat.metrics.metrics.LLMTokenUsage")
+    LLMUsageMetricsData = _import("pipecat.metrics.metrics.LLMUsageMetricsData")
+    MetricsFrame = _import("pipecat.frames.frames.MetricsFrame")
+
+    observer = BraintrustPipecatObserver()
+    processor = SimpleNamespace(name="OpenAILLMService#1")
+    frame = MetricsFrame(
+        data=[
+            LLMUsageMetricsData(
+                processor=processor.name,
+                model="gpt-4o-mini",
+                value=LLMTokenUsage(prompt_tokens=10, completion_tokens=5, total_tokens=15),
+            ),
+            LLMUsageMetricsData(
+                processor="JevClassifier#1",
+                model="gpt-4o-mini",
+                value=LLMTokenUsage(prompt_tokens=100, completion_tokens=20, total_tokens=120),
+            ),
+        ]
+    )
+
+    observer._capture_metrics(frame, processor)
+
+    assert observer._llm_metrics == {"prompt_tokens": 10, "completion_tokens": 5, "tokens": 15}
+
+
 @pytest.mark.asyncio
 async def test_span_customizer_redacts_incremental_tts_input(memory_logger):
     TTSStartedFrame = _import("pipecat.frames.frames.TTSStartedFrame")
diff --git a/py/src/braintrust/integrations/pipecat/tracing.py b/py/src/braintrust/integrations/pipecat/tracing.py
index a60c04d2f..d84d94466 100644
--- a/py/src/braintrust/integrations/pipecat/tracing.py
+++ b/py/src/braintrust/integrations/pipecat/tracing.py
@@ -456,9 +456,13 @@ def _discard_tts_audio(self, context_id: str) -> None:
         self._tts_audio_metadata.pop(context_id, None)
 
     def _capture_metrics(self, frame: Any, processor: Any) -> None:
+        processor_name = _processor_name(processor)
         for metric in getattr(frame, "data", []) or []:
             metric_type = type(metric).__name__
             if metric_type == "LLMUsageMetricsData":
+                metric_processor = getattr(metric, "processor", None)
+                if processor_name is not None and metric_processor is not None and metric_processor != processor_name:
+                    continue
                 self._llm_metrics.update(_llm_usage_metrics(getattr(metric, "value", None)))
                 self._llm_metadata.update(_metadata_from_metric(metric))
             elif metric_type == "TTFBMetricsData" and self._llm_span is not None: