Implementation of POST v1/responses endpoint

Author: asimurka

docs/openapi.json

@@ -7,6 +7,7 @@
     APIConnectionError,
     APIStatusError,
 )
+from llama_stack_client.types.conversations import ItemListResponse
 from sqlalchemy.exc import SQLAlchemyError
 
 from app.database import get_session
@@ -45,7 +46,10 @@
     normalize_conversation_id,
     to_llama_stack_conversation_id,
 )
-from utils.conversations import build_conversation_turns_from_items
+from utils.conversations import (
+    build_conversation_turns_from_items,
+    get_all_conversation_items,
+)
 from log import get_logger
 
 logger = get_logger(__name__)
@@ -225,57 +229,54 @@ async def get_conversation_endpoint_handler(  # pylint: disable=too-many-locals,
     logger.info(
         "Retrieving conversation %s using Conversations API", normalized_conv_id
     )
-
+    client = AsyncLlamaStackClientHolder().get_client()
     try:
-        client = AsyncLlamaStackClientHolder().get_client()
-
         # Convert to llama-stack format (add 'conv_' prefix if needed)
         llama_stack_conv_id = to_llama_stack_conversation_id(normalized_conv_id)
         logger.debug(
             "Calling llama-stack list_items with conversation_id: %s",
             llama_stack_conv_id,
         )
 
+        # Retrieve turns metadata from database
+        db_turns: list[UserTurn] = []
+        items: list[ItemListResponse] = []
+
+        try:
+            with get_session() as session:
+                db_turns.extend(
+                    (
+                        session.query(UserTurn)
+                        .filter_by(conversation_id=normalized_conv_id)
+                        .order_by(UserTurn.turn_number)
+                        .all()
+                    )
+                )
+        except SQLAlchemyError as e:
+            logger.error(
+                "Database error occurred while retrieving conversation turns for %s.",
+                normalized_conv_id,
+            )
+            response = InternalServerErrorResponse.database_error()
+            raise HTTPException(**response.model_dump()) from e
+
         # Use Conversations API to retrieve conversation items
-        conversation_items_response = await client.conversations.items.list(
-            conversation_id=llama_stack_conv_id,
-            after=None,
-            include=None,
-            limit=None,
-            order="asc",  # oldest first
+        conversation_items = await get_all_conversation_items(
+            client, llama_stack_conv_id
         )
-
-        if not conversation_items_response.data:
+        if not conversation_items:
             logger.error("No items found for conversation %s", conversation_id)
             response = NotFoundResponse(
                 resource="conversation", resource_id=normalized_conv_id
             ).model_dump()
             raise HTTPException(**response)
 
-        items = conversation_items_response.data
-
+        items.extend(conversation_items)
         logger.info(
             "Successfully retrieved %d items for conversation %s",
             len(items),
             conversation_id,
         )
-        # Retrieve turns metadata from database
-        db_turns: list[UserTurn] = []
-        try:
-            with get_session() as session:
-                db_turns = (
-                    session.query(UserTurn)
-                    .filter_by(conversation_id=normalized_conv_id)
-                    .order_by(UserTurn.turn_number)
-                    .all()
-                )
-        except SQLAlchemyError as e:
-            logger.error(
-                "Database error occurred while retrieving conversation turns for %s.",
-                normalized_conv_id,
-            )
-            response = InternalServerErrorResponse.database_error()
-            raise HTTPException(**response.model_dump()) from e
 
         # Build conversation turns from items and populate turns metadata
         # Use conversation.created_at for legacy conversations without turn metadata
@@ -284,8 +285,7 @@ async def get_conversation_endpoint_handler(  # pylint: disable=too-many-locals,
         )
 
         return ConversationResponse(
-            conversation_id=normalized_conv_id,
-            chat_history=chat_history,
+            conversation_id=normalized_conv_id, chat_history=chat_history
         )
 
     except APIConnectionError as e:

docs/responses.md

@@ -1,5 +1,3 @@
-# pylint: disable=too-many-locals,too-many-branches,too-many-nested-blocks
-
 """Handler for REST API call to provide answer to query using Response API."""
 
 import datetime
@@ -264,7 +262,7 @@ async def query_endpoint_handler(
     )
 
 
-async def retrieve_response(  # pylint: disable=too-many-locals
+async def retrieve_response(
     client: AsyncLlamaStackClient,
     responses_params: ResponsesApiParams,
     moderation_result: ShieldModerationResult,
@@ -291,7 +289,9 @@ async def retrieve_response(  # pylint: disable=too-many-locals
             responses_params.input,
             [moderation_result.refusal_response],
         )
-        return TurnSummary(llm_response=moderation_result.message)
+        return TurnSummary(
+            id=moderation_result.moderation_id, llm_response=moderation_result.message
+        )
     try:
         response = await client.responses.create(
             **responses_params.model_dump(exclude_none=True)