Add gen_ai.conversation.id instrumentation to OpenAI provider

## Subject
  Add `gen_ai.conversation.id` instrumentation to OpenAI provider

  ## Description
  Automatically extract and capture `gen_ai.conversation.id` from OpenAI's stateful APIs to comply with OTel GenAI semantic conventions.

  ### Changes
  - Extract conversation ID from `/responses` endpoint request (`conversation` parameter)
  - Add handler for `/conversations` endpoint
  - Extract conversation ID from Response objects (both sync and streaming)
  - Extract conversation ID from conversation creation responses

  ### Files Modified
  - `logfire/_internal/integrations/llm_providers/openai.py`
  - `tests/otel_integrations/test_openai.py`

Author: jimilp7

logfire/_internal/integrations/llm_providers/openai.py

@@ -10,6 +10,7 @@
 from openai.types.chat.chat_completion import ChatCompletion
 from openai.types.chat.chat_completion_chunk import ChatCompletionChunk
 from openai.types.completion import Completion
+from openai.types.conversations import Conversation
 from openai.types.create_embedding_response import CreateEmbeddingResponse
 from openai.types.images_response import ImagesResponse
 from openai.types.responses import Response
@@ -19,6 +20,7 @@
 
 from ...utils import handle_internal_errors, log_internal_error
 from .semconv import (
+    CONVERSATION_ID,
     OPERATION_NAME,
     PROVIDER_NAME,
     REQUEST_FREQUENCY_PENALTY,
@@ -122,6 +124,12 @@ def get_endpoint_config(options: FinalRequestOptions) -> EndpointConfig:
         }
         _extract_request_parameters(json_data, span_data)
 
+        if (conversation := json_data.get('conversation')) is not None:
+            if isinstance(conversation, str):
+                span_data[CONVERSATION_ID] = conversation
+            elif isinstance(conversation, dict) and 'id' in conversation:
+                span_data[CONVERSATION_ID] = conversation['id']
+
         return EndpointConfig(
             message_template='Responses API with {gen_ai.request.model!r}',
             span_data=span_data,
@@ -164,6 +172,17 @@ def get_endpoint_config(options: FinalRequestOptions) -> EndpointConfig:
             message_template='Image Generation with {request_data[model]!r}',
             span_data=span_data,
         )
+    elif url == '/conversations':
+        span_data = {
+            'request_data': json_data,
+            'url': url,
+            PROVIDER_NAME: 'openai',
+            OPERATION_NAME: 'conversation',
+        }
+        return EndpointConfig(
+            message_template='OpenAI Conversation Create',
+            span_data=span_data,
+        )
     else:
         span_data = {
             'request_data': json_data,
@@ -225,6 +244,11 @@ def get_response_data(self) -> Any:
     def get_attributes(self, span_data: dict[str, Any]) -> dict[str, Any]:
         response = self.get_response_data()
         span_data['events'] = span_data['events'] + responses_output_events(response)
+
+        if (conversation := getattr(response, 'conversation', None)) is not None:
+            if hasattr(conversation, 'id'):
+                span_data[CONVERSATION_ID] = conversation.id
+
         return span_data
 
 
@@ -311,6 +335,10 @@ def on_response(response: ResponseT, span: LogfireSpan) -> ResponseT:
     elif isinstance(response, ImagesResponse):
         span.set_attribute('response_data', {'images': response.data})
     elif isinstance(response, Response):  # pragma: no branch
+        if (conversation := getattr(response, 'conversation', None)) is not None:
+            if hasattr(conversation, 'id'):
+                span.set_attribute(CONVERSATION_ID, conversation.id)
+
         try:
             events = json.loads(span.attributes['events'])  # type: ignore
         except Exception:
@@ -319,6 +347,9 @@ def on_response(response: ResponseT, span: LogfireSpan) -> ResponseT:
             events += responses_output_events(response)
             span.set_attribute('events', events)
 
+    if isinstance(response, Conversation):
+        span.set_attribute(CONVERSATION_ID, response.id)
+
     return response
 
 

tests/otel_integrations/test_openai.py

@@ -360,6 +360,23 @@ def request_handler(request: httpx.Request) -> httpx.Response:
             200,
             json={'id': 'thread_abc123', 'object': 'thread', 'created_at': 1698107661, 'metadata': {}},
         )
+    elif request.url == 'https://api.openai.com/v1/responses':
+        json_body = json.loads(request.content)
+        conversation_data = {'id': json_body['conversation']} if json_body.get('conversation') else None
+        return httpx.Response(
+            200,
+            json={
+                'id': 'resp_test_123',
+                'created_at': 1698107661,
+                'model': json_body.get('model', 'gpt-4'),
+                'object': 'response',
+                'output': [
+                    {'type': 'message', 'role': 'assistant', 'content': [{'type': 'output_text', 'text': 'Nine'}]}
+                ],
+                'conversation': conversation_data,
+                'usage': {'input_tokens': 10, 'output_tokens': 5, 'total_tokens': 15},
+            },
+        )
     else:  # pragma: no cover
         raise ValueError(f'Unexpected request to {request.url!r}')
 
@@ -2803,3 +2820,117 @@ def test_openrouter_streaming_reasoning(exporter: TestExporter) -> None:
             },
         ]
     )
+
+
+def test_get_endpoint_config_conversations() -> None:
+    """Test that /conversations endpoint is properly configured."""
+    from unittest.mock import MagicMock
+
+    from logfire._internal.integrations.llm_providers.openai import (
+        get_endpoint_config,  # pyright: ignore[reportPrivateUsage]
+    )
+
+    options = MagicMock()
+    options.url = '/conversations'
+    options.json_data = {}
+
+    config = get_endpoint_config(options)
+    assert config.message_template == 'OpenAI Conversation Create'
+    assert config.span_data['gen_ai.provider.name'] == 'openai'
+    assert config.span_data['gen_ai.operation.name'] == 'conversation'
+
+
+def test_get_endpoint_config_responses_with_conversation_string() -> None:
+    """Test that /responses endpoint extracts conversation ID when provided as string."""
+    from unittest.mock import MagicMock
+
+    from logfire._internal.integrations.llm_providers.openai import (
+        get_endpoint_config,  # pyright: ignore[reportPrivateUsage]
+    )
+
+    options = MagicMock()
+    options.url = '/responses'
+    options.json_data = {
+        'model': 'gpt-4',
+        'input': 'Hello',
+        'conversation': 'conv_123456',
+    }
+
+    config = get_endpoint_config(options)
+    assert config.span_data['gen_ai.conversation.id'] == 'conv_123456'
+
+
+def test_get_endpoint_config_responses_with_conversation_dict() -> None:
+    """Test that /responses endpoint extracts conversation ID when provided as dict."""
+    from unittest.mock import MagicMock
+
+    from logfire._internal.integrations.llm_providers.openai import (
+        get_endpoint_config,  # pyright: ignore[reportPrivateUsage]
+    )
+
+    options = MagicMock()
+    options.url = '/responses'
+    options.json_data = {
+        'model': 'gpt-4',
+        'input': 'Hello',
+        'conversation': {'id': 'conv_789012'},
+    }
+
+    config = get_endpoint_config(options)
+    assert config.span_data['gen_ai.conversation.id'] == 'conv_789012'
+
+
+def test_on_response_handles_conversation_creation_response() -> None:
+    """Test that on_response extracts conversation ID from /conversations response."""
+    from unittest.mock import MagicMock
+
+    from openai.types.conversations import Conversation
+
+    from logfire._internal.integrations.llm_providers.openai import (
+        on_response,  # pyright: ignore[reportPrivateUsage]
+    )
+
+    response = Conversation.model_construct(id='conv_created_456')
+    mock_span = MagicMock()
+    mock_span.attributes = {}
+
+    on_response(response, mock_span)
+
+    mock_span.set_attribute.assert_any_call('gen_ai.conversation.id', 'conv_created_456')
+
+
+def test_streaming_handler_extracts_conversation_id() -> None:
+    """Test that OpenaiResponsesStreamState.get_attributes extracts conversation ID."""
+    from unittest.mock import MagicMock
+
+    from logfire._internal.integrations.llm_providers.openai import (
+        OpenaiResponsesStreamState,  # pyright: ignore[reportPrivateUsage]
+    )
+
+    mock_response = MagicMock()
+    mock_response.output = []
+    mock_response.conversation = MagicMock(id='conv_stream_789')
+
+    stream_state = OpenaiResponsesStreamState()
+    stream_state._state._completed_response = mock_response
+
+    span_data: dict[str, Any] = {'events': []}
+    result = stream_state.get_attributes(span_data)
+
+    assert result['gen_ai.conversation.id'] == 'conv_stream_789'
+
+
+def test_responses_api_with_conversation_id(instrumented_client: openai.Client, exporter: TestExporter) -> None:
+    """Integration test for conversation ID extraction from Responses API."""
+    response = instrumented_client.responses.create(
+        model='gpt-4',
+        input='What is four plus five?',
+        conversation='conv_integration_test_123',
+    )
+    assert response.id == 'resp_test_123'
+    assert response.conversation is not None
+    assert response.conversation.id == 'conv_integration_test_123'
+
+    spans = exporter.exported_spans_as_dict(parse_json_attributes=True)
+    assert len(spans) == 1
+    assert spans[0]['attributes']['gen_ai.conversation.id'] == 'conv_integration_test_123'