feat(langfuse.py): Allow for individual call message/response redaction

docs/my-website/docs/observability/langfuse_integration.md

@@ -213,8 +213,20 @@ chat(messages)
 
 ## Redacting Messages, Response Content from Langfuse Logging 
 
+### Redact Messages and Responses from all Langfuse Logging
+
 Set `litellm.turn_off_message_logging=True` This will prevent the messages and responses from being logged to langfuse, but request metadata will still be logged.
 
+### Redact Messages and Responses from specific Langfuse Logging
+
+In the metadata typically passed for text completion or embedding calls you can set specific keys to mask the messages and responses for this call.
+
+Setting `mask_input` to `True` will mask the input from being logged for this call 
+
+Setting `mask_output` to `True` will make the output from being logged for this call.
+
+Be aware that if you are continuing an existing trace, and you set `update_trace_keys` to include either `input` or `output` and you set the corresponding `mask_input` or `mask_output`, then that trace will have its existing input and/or output replaced with a redacted message.
+
 ## Troubleshooting & Errors
 ### Data not getting logged to Langfuse ? 
 - Ensure you're on the latest version of langfuse `pip install langfuse -U`. The latest version allows litellm to log JSON input/outputs to langfuse

litellm/integrations/langfuse.py

@@ -322,6 +322,8 @@ class LangFuseLogger:
             existing_trace_id = clean_metadata.pop("existing_trace_id", None)
             update_trace_keys = clean_metadata.pop("update_trace_keys", [])
             debug = clean_metadata.pop("debug_langfuse", None)
+            mask_input = clean_metadata.pop("mask_input", False)
+            mask_output = clean_metadata.pop("mask_output", False)
 
             if trace_name is None and existing_trace_id is None:
                 # just log `litellm-{call_type}` as the trace name
@@ -349,15 +351,15 @@ class LangFuseLogger:
 
                 # Special keys that are found in the function arguments and not the metadata
                 if "input" in update_trace_keys:
-                    trace_params["input"] = input
+                    trace_params["input"] = input if not mask_input else "redacted-by-litellm"
                 if "output" in update_trace_keys:
-                    trace_params["output"] = output
+                    trace_params["output"] = output if not mask_output else "redacted-by-litellm"
             else:  # don't overwrite an existing trace
                 trace_params = {
                     "id": trace_id,
                     "name": trace_name,
                     "session_id": session_id,
-                    "input": input,
+                    "input": input if not mask_input else "redacted-by-litellm",
                     "version": clean_metadata.pop(
                         "trace_version", clean_metadata.get("version", None)
                     ),  # If provided just version, it will applied to the trace as well, if applied a trace version it will take precedence
@@ -373,7 +375,7 @@ class LangFuseLogger:
                 if level == "ERROR":
                     trace_params["status_message"] = output
                 else:
-                    trace_params["output"] = output
+                    trace_params["output"] = output if not mask_output else "redacted-by-litellm"
 
             if debug == True or (isinstance(debug, str) and debug.lower() == "true"):
                 if "metadata" in trace_params:
@@ -463,8 +465,8 @@ class LangFuseLogger:
                 "end_time": end_time,
                 "model": kwargs["model"],
                 "model_parameters": optional_params,
-                "input": input,
-                "output": output,
+                "input": input if not mask_input else "redacted-by-litellm",
+                "output": output if not mask_output else "redacted-by-litellm",
                 "usage": usage,
                 "metadata": clean_metadata,
                 "level": level,

litellm/tests/test_alangfuse.py

@@ -228,6 +228,40 @@ async def test_langfuse_logging_without_request_response(stream, langfuse_client
         pytest.fail(f"An exception occurred - {e}")
 
 
+@pytest.mark.asyncio
+async def test_langfuse_masked_input_output(langfuse_client):
+    """
+    Test that creates a trace with masked input and output
+    """
+    import uuid
+
+    for mask_value in [True, False]:
+        _unique_trace_name = f"litellm-test-{str(uuid.uuid4())}"
+        litellm.set_verbose = True
+        litellm.success_callback = ["langfuse"]
+        response = await create_async_task(
+            model="gpt-3.5-turbo",
+            messages=[{"role": "user", "content": "This is a test"}],
+            metadata={"trace_id": _unique_trace_name, "mask_input": mask_value, "mask_output": mask_value},
+            mock_response="This is a test response"
+        )
+        print(response)
+        expected_input = "redacted-by-litellm" if mask_value else {'messages': [{'content': 'This is a test', 'role': 'user'}]}
+        expected_output = "redacted-by-litellm" if mask_value else {'content': 'This is a test response', 'role': 'assistant'}
+        langfuse_client.flush()
+        await asyncio.sleep(2)
+
+        # get trace with _unique_trace_name
+        trace = langfuse_client.get_trace(id=_unique_trace_name)
+        generations = list(
+            reversed(langfuse_client.get_generations(trace_id=_unique_trace_name).data)
+        )
+
+        assert trace.input == expected_input
+        assert trace.output == expected_output
+        assert generations[0].input == expected_input
+        assert generations[0].output == expected_output
+
 @pytest.mark.asyncio
 async def test_langfuse_logging_metadata(langfuse_client):
     """