Fix Galileo logging to match Langfuse across all endpoint types.

Stop skipping ingest when output is empty and log embeddings with a placeholder so embedding, speech, and other non-text responses are recorded like Langfuse.

Co-authored-by: Cursor <cursoragent@cursor.com>

litellm/integrations/galileo.py

@@ -517,7 +517,7 @@ class GalileoObserve(CustomLogger):
         """
         Mirror Langfuse _get_langfuse_input_output_content for Galileo ingest.
 
-        Returns (input_text, output_text, messages_for_span). output_text None skips ingest.
+        Returns (input_text, output_text, messages_for_span).
         """
         call_type = kwargs.get("call_type")
         prompt = self._build_prompt(kwargs)
@@ -530,10 +530,11 @@ class GalileoObserve(CustomLogger):
             return self._prompt_to_input_text(prompt), status_message, prompt
 
         if response_obj is not None and (
-            call_type == "embedding"
+            call_type in ("embedding", "aembedding")
             or isinstance(response_obj, litellm.EmbeddingResponse)
         ):
-            return self._prompt_to_input_text(prompt), None, prompt
+            # Match Langfuse OTEL: log embeddings without serializing vectors.
+            return self._prompt_to_input_text(prompt), "embedding-output", prompt
 
         if response_obj is not None and isinstance(response_obj, litellm.ModelResponse):
             output = self._get_chat_content_for_galileo(response_obj)
@@ -627,7 +628,7 @@ class GalileoObserve(CustomLogger):
                 kwargs.get("messages") or [],
             )
 
-        return self._prompt_to_input_text(prompt), None, kwargs.get("messages") or []
+        return self._prompt_to_input_text(prompt), "", kwargs.get("messages") or []
 
     def get_output_str_from_response(
         self, response_obj: Any, kwargs: Dict[str, Any]
@@ -713,10 +714,7 @@ class GalileoObserve(CustomLogger):
             kwargs=kwargs, response_obj=response_obj
         )
         if output_text is None:
-            verbose_logger.debug(
-                "Galileo Logger: skipping %s — no text output to log", _call_type
-            )
-            return
+            output_text = ""
 
         raw_start = slo.get("startTime")
         raw_end = slo.get("endTime")

tests/test_litellm/integrations/test_galileo.py

@@ -357,12 +357,18 @@ def test_galileo_record_to_v2_span_with_tags_and_offset():
 
 def test_galileo_get_output_str_variants(galileo_v2_env):
     logger = GalileoObserve()
-    assert logger.get_output_str_from_response(None, {}) is None
+    assert logger.get_output_str_from_response(None, {}) == ""
     assert (
         logger.get_output_str_from_response(
             EmbeddingResponse(), {"call_type": "embedding"}
         )
-        is None
+        == "embedding-output"
+    )
+    assert (
+        logger.get_output_str_from_response(
+            EmbeddingResponse(), {"call_type": "aembedding"}
+        )
+        == "embedding-output"
     )
 
     text_resp = TextCompletionResponse()
@@ -414,7 +420,7 @@ def test_galileo_get_output_str_variants(galileo_v2_env):
         {"call_type": "acompletion", "messages": [{"role": "user", "content": "hi"}]},
     )
 
-    assert logger.get_output_str_from_response("not-a-supported-type", {}) is None
+    assert logger.get_output_str_from_response("not-a-supported-type", {}) == ""
 
 
 def test_galileo_get_input_output_error_status_message(galileo_v2_env):
@@ -445,6 +451,48 @@ def test_galileo_get_output_str_rerank_response(galileo_v2_env):
     assert '"relevance_score": 0.98' in output
 
 
+@pytest.mark.asyncio
+async def test_galileo_async_log_success_embedding(galileo_v2_env):
+    import datetime
+
+    logger = GalileoObserve()
+    embedding_response = EmbeddingResponse(
+        data=[{"object": "embedding", "embedding": [0.1, 0.2, 0.3], "index": 0}]
+    )
+
+    mock_response = MagicMock()
+    mock_response.is_success = True
+    mock_response.status_code = 201
+
+    with patch.object(logger.async_httpx_handler, "post", return_value=mock_response):
+        await logger.async_log_success_event(
+            kwargs={
+                "call_type": "aembedding",
+                "model": "text-embedding-3-small",
+                "input": "hello world",
+                "standard_logging_object": {
+                    "call_type": "aembedding",
+                    "model": "text-embedding-3-small",
+                    "prompt_tokens": 2,
+                    "completion_tokens": 0,
+                    "total_tokens": 2,
+                    "response_cost": 0.0,
+                    "startTime": datetime.datetime(
+                        2026, 5, 25, 12, 0, 0, tzinfo=datetime.timezone.utc
+                    ).timestamp(),
+                    "endTime": datetime.datetime(
+                        2026, 5, 25, 12, 0, 1, tzinfo=datetime.timezone.utc
+                    ).timestamp(),
+                },
+            },
+            response_obj=embedding_response,
+            start_time=datetime.datetime(2026, 5, 25, 12, 0, 0),
+            end_time=datetime.datetime(2026, 5, 25, 12, 0, 1),
+        )
+
+    assert logger.in_memory_records == []
+
+
 @pytest.mark.asyncio
 async def test_galileo_async_log_success_rerank(galileo_v2_env):
     import datetime