Add namespace metadata to deriver LLM calls

src/deriver/deriver.py

@@ -154,6 +154,7 @@ async def process_representation_tasks_batch(
         enable_retry=True,
         retry_attempts=3,
         trace_name="minimal_deriver",
+        request_metadata={"namespace": settings.NAMESPACE},
         telemetry=LLMTelemetryContext(
             workspace_name=latest_message.workspace_name,
             call_purpose=CallPurpose.DERIVER_REPRESENTATION.value,

src/llm/api.py

@@ -75,6 +75,7 @@ async def honcho_llm_call(
     trace_name: str | None = None,
     iteration_callback: IterationCallback | None = None,
     telemetry: LLMTelemetryContext | None = None,
+    request_metadata: dict[str, str] | None = None,
 ) -> HonchoLLMCallResponse[M]: ...
 
 
@@ -105,6 +106,7 @@ async def honcho_llm_call(
     trace_name: str | None = None,
     iteration_callback: IterationCallback | None = None,
     telemetry: LLMTelemetryContext | None = None,
+    request_metadata: dict[str, str] | None = None,
 ) -> HonchoLLMCallResponse[str]: ...
 
 
@@ -135,6 +137,7 @@ async def honcho_llm_call(
     trace_name: str | None = None,
     iteration_callback: IterationCallback | None = None,
     telemetry: LLMTelemetryContext | None = None,
+    request_metadata: dict[str, str] | None = None,
 ) -> AsyncIterator[HonchoLLMCallStreamChunk] | StreamingResponseWithMetadata: ...
 
 
@@ -165,6 +168,7 @@ async def honcho_llm_call(
     trace_name: str | None = None,
     iteration_callback: IterationCallback | None = None,
     telemetry: LLMTelemetryContext | None = None,
+    request_metadata: dict[str, str] | None = None,
 ) -> (
     HonchoLLMCallResponse[Any]
     | AsyncIterator[HonchoLLMCallStreamChunk]
@@ -243,6 +247,7 @@ async def honcho_llm_call(
                 selected_config=plan.selected_config,
                 plan=plan,
                 telemetry=telemetry,
+                request_metadata=request_metadata,
             )
         return await honcho_llm_call_inner(
             plan.provider,
@@ -263,6 +268,7 @@ async def honcho_llm_call(
             selected_config=plan.selected_config,
             plan=plan,
             telemetry=telemetry,
+            request_metadata=request_metadata,
         )
 
     decorated = _call_with_provider_selection
@@ -373,6 +379,7 @@ async def honcho_llm_call(
                         plan=plan,
                         telemetry=telemetry,
                         messages=captured_messages,
+                        request_metadata=request_metadata,
                     )
                 return await honcho_llm_call_inner(
                     plan.provider,
@@ -394,6 +401,7 @@ async def honcho_llm_call(
                     plan=plan,
                     telemetry=telemetry,
                     messages=captured_messages,
+                    request_metadata=request_metadata,
                 )
 
             wrapped = _toolless_call

src/llm/backends/openai.py

@@ -322,6 +322,8 @@ class OpenAIBackend:
             ):
                 if key in extra_params:
                     params[key] = extra_params[key]
+            if "metadata" in extra_params:
+                params["metadata"] = extra_params["metadata"]
         return params
 
     def _normalize_response(

src/llm/executor.py

@@ -210,6 +210,7 @@ async def honcho_llm_call_inner(
     selected_config: ModelConfig | None = None,
     plan: AttemptPlan | None = None,
     telemetry: LLMTelemetryContext | None = None,
+    request_metadata: dict[str, str] | None = None,
 ) -> HonchoLLMCallResponse[M]: ...
 
 
@@ -234,6 +235,7 @@ async def honcho_llm_call_inner(
     selected_config: ModelConfig | None = None,
     plan: AttemptPlan | None = None,
     telemetry: LLMTelemetryContext | None = None,
+    request_metadata: dict[str, str] | None = None,
 ) -> HonchoLLMCallResponse[str]: ...
 
 
@@ -258,6 +260,7 @@ async def honcho_llm_call_inner(
     selected_config: ModelConfig | None = None,
     plan: AttemptPlan | None = None,
     telemetry: LLMTelemetryContext | None = None,
+    request_metadata: dict[str, str] | None = None,
 ) -> AsyncIterator[HonchoLLMCallStreamChunk]: ...
 
 
@@ -281,6 +284,7 @@ async def honcho_llm_call_inner(
     selected_config: ModelConfig | None = None,
     plan: AttemptPlan | None = None,
     telemetry: LLMTelemetryContext | None = None,
+    request_metadata: dict[str, str] | None = None,
 ) -> HonchoLLMCallResponse[Any] | AsyncIterator[HonchoLLMCallStreamChunk]:
     """One backend call. No retry, no fallback, no tool loop.
 
@@ -318,6 +322,8 @@ async def honcho_llm_call_inner(
     # knobs — they pass through extra_params. execute_completion merges
     # build_config_extra_params(effective_config) on top for top_p/seed/etc.
     call_extras: dict[str, Any] = {"json_mode": json_mode, "verbosity": verbosity}
+    if request_metadata is not None:
+        call_extras["metadata"] = request_metadata
 
     if stream:
         # Stream path: setup must run inside the awaited coroutine so it

tests/deriver/test_deriver_processing.py

@@ -64,6 +64,7 @@ class TestDeriverProcessing:
             == expected_config.thinking_budget_tokens
         )
         assert kwargs["model_config"].stop_sequences == expected_config.stop_sequences
+        assert kwargs["request_metadata"] == {"namespace": settings.NAMESPACE}
         assert "llm_settings" not in kwargs
 
     async def test_process_representation_tasks_batch_passes_custom_instructions_into_prompt(

tests/llm/test_backends/test_openai.py

@@ -227,6 +227,44 @@ async def test_openai_backend_skips_extra_body_when_thinking_budget_zero() -> No
     assert "extra_body" not in call
 
 
+@pytest.mark.asyncio
+async def test_openai_backend_passes_request_metadata() -> None:
+    client = Mock()
+    client.chat.completions.create = AsyncMock(
+        return_value=SimpleNamespace(
+            choices=[
+                SimpleNamespace(
+                    finish_reason="stop",
+                    message=SimpleNamespace(
+                        content="ok",
+                        tool_calls=[],
+                        reasoning_details=[],
+                    ),
+                )
+            ],
+            usage=SimpleNamespace(
+                prompt_tokens=10,
+                completion_tokens=5,
+                prompt_tokens_details=None,
+            ),
+        )
+    )
+
+    backend = OpenAIBackend(client)
+    await backend.complete(
+        model="gpt-4.1",
+        messages=[{"role": "user", "content": "Hello"}],
+        max_tokens=100,
+        extra_params={"metadata": {"namespace": "honcho"}},
+    )
+
+    await_args = client.chat.completions.create.await_args
+    if await_args is None:
+        raise AssertionError("Expected OpenAI create call")
+    call = await_args.kwargs
+    assert call["metadata"] == {"namespace": "honcho"}
+
+
 @pytest.mark.asyncio
 async def test_openai_backend_converts_anthropic_style_tools() -> None:
     client = Mock()