feat: Add support for Conversations in Responses API (#3743)

# What does this PR do?
This PR adds support for Conversations in Responses.

<!-- If resolving an issue, uncomment and update the line below -->
<!-- Closes #[issue-number] -->

## Test Plan
Unit tests
Integration tests

<Details>
<Summary>Manual testing with this script: (click to expand)</Summary>

```python
from openai import OpenAI

client = OpenAI()
client = OpenAI(base_url="http://localhost:8321/v1/", api_key="none")

def test_conversation_create():
    print("Testing conversation create...")
    conversation = client.conversations.create(
        metadata={"topic": "demo"},
        items=[
            {"type": "message", "role": "user", "content": "Hello!"}
        ]
    )
    print(f"Created: {conversation}")
    return conversation

def test_conversation_retrieve(conv_id):
    print(f"Testing conversation retrieve for {conv_id}...")
    retrieved = client.conversations.retrieve(conv_id)
    print(f"Retrieved: {retrieved}")
    return retrieved

def test_conversation_update(conv_id):
    print(f"Testing conversation update for {conv_id}...")
    updated = client.conversations.update(
        conv_id,
        metadata={"topic": "project-x"}
    )
    print(f"Updated: {updated}")
    return updated

def test_conversation_delete(conv_id):
    print(f"Testing conversation delete for {conv_id}...")
    deleted = client.conversations.delete(conv_id)
    print(f"Deleted: {deleted}")
    return deleted

def test_conversation_items_create(conv_id):
    print(f"Testing conversation items create for {conv_id}...")
    items = client.conversations.items.create(
        conv_id,
        items=[
            {
                "type": "message",
                "role": "user",
                "content": [{"type": "input_text", "text": "Hello!"}]
            },
            {
                "type": "message",
                "role": "user",
                "content": [{"type": "input_text", "text": "How are you?"}]
            }
        ]
    )
    print(f"Items created: {items}")
    return items

def test_conversation_items_list(conv_id):
    print(f"Testing conversation items list for {conv_id}...")
    items = client.conversations.items.list(conv_id, limit=10)
    print(f"Items list: {items}")
    return items

def test_conversation_item_retrieve(conv_id, item_id):
    print(f"Testing conversation item retrieve for {conv_id}/{item_id}...")
    item = client.conversations.items.retrieve(conversation_id=conv_id, item_id=item_id)
    print(f"Item retrieved: {item}")
    return item

def test_conversation_item_delete(conv_id, item_id):
    print(f"Testing conversation item delete for {conv_id}/{item_id}...")
    deleted = client.conversations.items.delete(conversation_id=conv_id, item_id=item_id)
    print(f"Item deleted: {deleted}")
    return deleted

def test_conversation_responses_create():
    print("\nTesting conversation create for a responses example...")
    conversation = client.conversations.create()
    print(f"Created: {conversation}")

    response = client.responses.create(
      model="gpt-4.1",
      input=[{"role": "user", "content": "What are the 5 Ds of dodgeball?"}],
      conversation=conversation.id,
    )
    print(f"Created response: {response} for conversation {conversation.id}")

    return response, conversation

def test_conversations_responses_create_followup(
        conversation,
        content="Repeat what you just said but add 'this is my second time saying this'",
    ):
    print(f"Using: {conversation.id}")

    response = client.responses.create(
      model="gpt-4.1",
      input=[{"role": "user", "content": content}],
      conversation=conversation.id,
    )
    print(f"Created response: {response} for conversation {conversation.id}")

    conv_items = client.conversations.items.list(conversation.id)
    print(f"\nRetrieving list of items for conversation {conversation.id}:")
    print(conv_items.model_dump_json(indent=2))

def test_response_with_fake_conv_id():
    fake_conv_id = "conv_zzzzzzzzz5dc81908289d62779d2ac510a2b0b602ef00a44"
    print(f"Using {fake_conv_id}")
    try:
        response = client.responses.create(
          model="gpt-4.1",
          input=[{"role": "user", "content": "say hello"}],
          conversation=fake_conv_id,
        )
        print(f"Created response: {response} for conversation {fake_conv_id}")
    except Exception as e:
        print(f"failed to create response for conversation {fake_conv_id} with error {e}")


def main():
    print("Testing OpenAI Conversations API...")

    # Create conversation
    conversation = test_conversation_create()
    conv_id = conversation.id

    # Retrieve conversation
    test_conversation_retrieve(conv_id)

    # Update conversation
    test_conversation_update(conv_id)

    # Create items
    items = test_conversation_items_create(conv_id)

    # List items
    items_list = test_conversation_items_list(conv_id)

    # Retrieve specific item
    if items_list.data:
        item_id = items_list.data[0].id
        test_conversation_item_retrieve(conv_id, item_id)

        # Delete item
        test_conversation_item_delete(conv_id, item_id)

    # Delete conversation
    test_conversation_delete(conv_id)

    response, conversation2 = test_conversation_responses_create()
    print('\ntesting reseponse retrieval')
    test_conversation_retrieve(conversation2.id)

    print('\ntesting responses follow up')
    test_conversations_responses_create_followup(conversation2)

    print('\ntesting responses follow up x2!')

    test_conversations_responses_create_followup(
        conversation2,
        content="Repeat what you just said but add 'this is my third time saying this'",
    )

    test_response_with_fake_conv_id()

    print("All tests completed!")


if __name__ == "__main__":
    main()
```
</Details>

---------

Signed-off-by: Francisco Javier Arceo <farceo@redhat.com>
Co-authored-by: Ashwin Bharambe <ashwin.bharambe@gmail.com>

llama_stack/apis/agents/agents.py

@@ -812,6 +812,7 @@ class Agents(Protocol):
         model: str,
         instructions: str | None = None,
         previous_response_id: str | None = None,
+        conversation: str | None = None,
         store: bool | None = True,
         stream: bool | None = False,
         temperature: float | None = None,
@@ -831,6 +832,7 @@ class Agents(Protocol):
         :param input: Input message(s) to create the response.
         :param model: The underlying LLM used for completions.
         :param previous_response_id: (Optional) if specified, the new response will be a continuation of the previous response. This can be used to easily fork-off new responses from existing responses.
+        :param conversation: (Optional) The ID of a conversation to add the response to. Must begin with 'conv_'. Input and output messages will be automatically added to the conversation.
         :param include: (Optional) Additional fields to include in the response.
         :param shields: (Optional) List of shields to apply during response generation. Can be shield IDs (strings) or shield specifications.
         :returns: An OpenAIResponseObject.

llama_stack/apis/common/errors.py

@@ -86,3 +86,18 @@ class TokenValidationError(ValueError):
 
     def __init__(self, message: str) -> None:
         super().__init__(message)
+
+
+class ConversationNotFoundError(ResourceNotFoundError):
+    """raised when Llama Stack cannot find a referenced conversation"""
+
+    def __init__(self, conversation_id: str) -> None:
+        super().__init__(conversation_id, "Conversation", "client.conversations.list()")
+
+
+class InvalidConversationIdError(ValueError):
+    """raised when a conversation ID has an invalid format"""
+
+    def __init__(self, conversation_id: str) -> None:
+        message = f"Invalid conversation ID '{conversation_id}'. Expected an ID that begins with 'conv_'."
+        super().__init__(message)

llama_stack/core/conversations/conversations.py

@@ -193,12 +193,15 @@ class ConversationServiceImpl(Conversations):
         await self._get_validated_conversation(conversation_id)
 
         created_items = []
-        created_at = int(time.time())
+        base_time = int(time.time())
 
-        for item in items:
+        for i, item in enumerate(items):
             item_dict = item.model_dump()
             item_id = self._get_or_generate_item_id(item, item_dict)
 
+            # make each timestamp unique to maintain order
+            created_at = base_time + i
+
             item_record = {
                 "id": item_id,
                 "conversation_id": conversation_id,

llama_stack/core/resolver.py

@@ -150,6 +150,7 @@ async def resolve_impls(
     provider_registry: ProviderRegistry,
     dist_registry: DistributionRegistry,
     policy: list[AccessRule],
+    internal_impls: dict[Api, Any] | None = None,
 ) -> dict[Api, Any]:
     """
     Resolves provider implementations by:
@@ -172,7 +173,7 @@ async def resolve_impls(
 
     sorted_providers = sort_providers_by_deps(providers_with_specs, run_config)
 
-    return await instantiate_providers(sorted_providers, router_apis, dist_registry, run_config, policy)
+    return await instantiate_providers(sorted_providers, router_apis, dist_registry, run_config, policy, internal_impls)
 
 
 def specs_for_autorouted_apis(apis_to_serve: list[str] | set[str]) -> dict[str, dict[str, ProviderWithSpec]]:
@@ -280,9 +281,10 @@ async def instantiate_providers(
     dist_registry: DistributionRegistry,
     run_config: StackRunConfig,
     policy: list[AccessRule],
+    internal_impls: dict[Api, Any] | None = None,
 ) -> dict[Api, Any]:
     """Instantiates providers asynchronously while managing dependencies."""
-    impls: dict[Api, Any] = {}
+    impls: dict[Api, Any] = internal_impls.copy() if internal_impls else {}
     inner_impls_by_provider_id: dict[str, dict[str, Any]] = {f"inner-{x.value}": {} for x in router_apis}
     for api_str, provider in sorted_providers:
         # Skip providers that are not enabled

llama_stack/core/stack.py

@@ -326,12 +326,17 @@ class Stack:
 
         dist_registry, _ = await create_dist_registry(self.run_config.metadata_store, self.run_config.image_name)
         policy = self.run_config.server.auth.access_policy if self.run_config.server.auth else []
-        impls = await resolve_impls(
-            self.run_config, self.provider_registry or get_provider_registry(self.run_config), dist_registry, policy
-        )
 
-        # Add internal implementations after all other providers are resolved
-        add_internal_implementations(impls, self.run_config)
+        internal_impls = {}
+        add_internal_implementations(internal_impls, self.run_config)
+
+        impls = await resolve_impls(
+            self.run_config,
+            self.provider_registry or get_provider_registry(self.run_config),
+            dist_registry,
+            policy,
+            internal_impls,
+        )
 
         if Api.prompts in impls:
             await impls[Api.prompts].initialize()

llama_stack/distributions/ci-tests/run.yaml

@@ -224,6 +224,9 @@ metadata_store:
 inference_store:
   type: sqlite
   db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/ci-tests}/inference_store.db
+conversations_store:
+  type: sqlite
+  db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/ci-tests}/conversations.db
 models: []
 shields:
 - shield_id: llama-guard

llama_stack/distributions/dell/run-with-safety.yaml

@@ -101,6 +101,9 @@ metadata_store:
 inference_store:
   type: sqlite
   db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/dell}/inference_store.db
+conversations_store:
+  type: sqlite
+  db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/dell}/conversations.db
 models:
 - metadata: {}
   model_id: ${env.INFERENCE_MODEL}

llama_stack/distributions/dell/run.yaml

@@ -97,6 +97,9 @@ metadata_store:
 inference_store:
   type: sqlite
   db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/dell}/inference_store.db
+conversations_store:
+  type: sqlite
+  db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/dell}/conversations.db
 models:
 - metadata: {}
   model_id: ${env.INFERENCE_MODEL}

llama_stack/distributions/meta-reference-gpu/run-with-safety.yaml

@@ -114,6 +114,9 @@ metadata_store:
 inference_store:
   type: sqlite
   db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/meta-reference-gpu}/inference_store.db
+conversations_store:
+  type: sqlite
+  db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/meta-reference-gpu}/conversations.db
 models:
 - metadata: {}
   model_id: ${env.INFERENCE_MODEL}

llama_stack/distributions/meta-reference-gpu/run.yaml

@@ -104,6 +104,9 @@ metadata_store:
 inference_store:
   type: sqlite
   db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/meta-reference-gpu}/inference_store.db
+conversations_store:
+  type: sqlite
+  db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/meta-reference-gpu}/conversations.db
 models:
 - metadata: {}
   model_id: ${env.INFERENCE_MODEL}

llama_stack/distributions/nvidia/run-with-safety.yaml

@@ -103,6 +103,9 @@ metadata_store:
 inference_store:
   type: sqlite
   db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/nvidia}/inference_store.db
+conversations_store:
+  type: sqlite
+  db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/nvidia}/conversations.db
 models:
 - metadata: {}
   model_id: ${env.INFERENCE_MODEL}

llama_stack/distributions/nvidia/run.yaml

@@ -92,6 +92,9 @@ metadata_store:
 inference_store:
   type: sqlite
   db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/nvidia}/inference_store.db
+conversations_store:
+  type: sqlite
+  db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/nvidia}/conversations.db
 models: []
 shields: []
 vector_dbs: []

llama_stack/distributions/open-benchmark/run.yaml

@@ -134,6 +134,9 @@ metadata_store:
 inference_store:
   type: sqlite
   db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/open-benchmark}/inference_store.db
+conversations_store:
+  type: sqlite
+  db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/open-benchmark}/conversations.db
 models:
 - metadata: {}
   model_id: gpt-4o

llama_stack/distributions/postgres-demo/run.yaml

@@ -86,6 +86,9 @@ inference_store:
   db: ${env.POSTGRES_DB:=llamastack}
   user: ${env.POSTGRES_USER:=llamastack}
   password: ${env.POSTGRES_PASSWORD:=llamastack}
+conversations_store:
+  type: sqlite
+  db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/postgres-demo}/conversations.db
 models:
 - metadata: {}
   model_id: ${env.INFERENCE_MODEL}

llama_stack/distributions/starter-gpu/run.yaml

@@ -227,6 +227,9 @@ metadata_store:
 inference_store:
   type: sqlite
   db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter-gpu}/inference_store.db
+conversations_store:
+  type: sqlite
+  db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter-gpu}/conversations.db
 models: []
 shields:
 - shield_id: llama-guard

llama_stack/distributions/starter/run.yaml

@@ -224,6 +224,9 @@ metadata_store:
 inference_store:
   type: sqlite
   db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/inference_store.db
+conversations_store:
+  type: sqlite
+  db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/starter}/conversations.db
 models: []
 shields:
 - shield_id: llama-guard

llama_stack/distributions/template.py

@@ -181,6 +181,7 @@ class RunConfigSettings(BaseModel):
     default_benchmarks: list[BenchmarkInput] | None = None
     metadata_store: dict | None = None
     inference_store: dict | None = None
+    conversations_store: dict | None = None
 
     def run_config(
         self,
@@ -240,6 +241,11 @@ class RunConfigSettings(BaseModel):
                 __distro_dir__=f"~/.llama/distributions/{name}",
                 db_name="inference_store.db",
             ),
+            "conversations_store": self.conversations_store
+            or SqliteSqlStoreConfig.sample_run_config(
+                __distro_dir__=f"~/.llama/distributions/{name}",
+                db_name="conversations.db",
+            ),
             "models": [m.model_dump(exclude_none=True) for m in (self.default_models or [])],
             "shields": [s.model_dump(exclude_none=True) for s in (self.default_shields or [])],
             "vector_dbs": [],

llama_stack/distributions/watsonx/run.yaml

@@ -107,6 +107,9 @@ metadata_store:
 inference_store:
   type: sqlite
   db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/watsonx}/inference_store.db
+conversations_store:
+  type: sqlite
+  db_path: ${env.SQLITE_STORE_DIR:=~/.llama/distributions/watsonx}/conversations.db
 models: []
 shields: []
 vector_dbs: []

llama_stack/log.py

@@ -30,6 +30,7 @@ CATEGORIES = [
     "tools",
     "client",
     "telemetry",
+    "openai",
     "openai_responses",
     "openai_conversations",
     "testing",

llama_stack/providers/inline/agents/meta_reference/__init__.py

@@ -21,6 +21,7 @@ async def get_provider_impl(config: MetaReferenceAgentsImplConfig, deps: dict[Ap
         deps[Api.safety],
         deps[Api.tool_runtime],
         deps[Api.tool_groups],
+        deps[Api.conversations],
         policy,
         Api.telemetry in deps,
     )

llama_stack/providers/inline/agents/meta_reference/agents.py

@@ -30,6 +30,7 @@ from llama_stack.apis.agents import (
 )
 from llama_stack.apis.agents.openai_responses import OpenAIResponseText
 from llama_stack.apis.common.responses import PaginatedResponse
+from llama_stack.apis.conversations import Conversations
 from llama_stack.apis.inference import (
     Inference,
     ToolConfig,
@@ -63,6 +64,7 @@ class MetaReferenceAgentsImpl(Agents):
         safety_api: Safety,
         tool_runtime_api: ToolRuntime,
         tool_groups_api: ToolGroups,
+        conversations_api: Conversations,
         policy: list[AccessRule],
         telemetry_enabled: bool = False,
     ):
@@ -72,6 +74,7 @@ class MetaReferenceAgentsImpl(Agents):
         self.safety_api = safety_api
         self.tool_runtime_api = tool_runtime_api
         self.tool_groups_api = tool_groups_api
+        self.conversations_api = conversations_api
         self.telemetry_enabled = telemetry_enabled
 
         self.in_memory_store = InmemoryKVStoreImpl()
@@ -88,6 +91,7 @@ class MetaReferenceAgentsImpl(Agents):
             tool_runtime_api=self.tool_runtime_api,
             responses_store=self.responses_store,
             vector_io_api=self.vector_io_api,
+            conversations_api=self.conversations_api,
         )
 
     async def create_agent(
@@ -325,6 +329,7 @@ class MetaReferenceAgentsImpl(Agents):
         model: str,
         instructions: str | None = None,
         previous_response_id: str | None = None,
+        conversation: str | None = None,
         store: bool | None = True,
         stream: bool | None = False,
         temperature: float | None = None,
@@ -339,6 +344,7 @@ class MetaReferenceAgentsImpl(Agents):
             model,
             instructions,
             previous_response_id,
+            conversation,
             store,
             stream,
             temperature,

llama_stack/providers/inline/agents/meta_reference/responses/openai_responses.py

@@ -24,6 +24,11 @@ from llama_stack.apis.agents.openai_responses import (
     OpenAIResponseText,
     OpenAIResponseTextFormat,
 )
+from llama_stack.apis.common.errors import (
+    InvalidConversationIdError,
+)
+from llama_stack.apis.conversations import Conversations
+from llama_stack.apis.conversations.conversations import ConversationItem
 from llama_stack.apis.inference import (
     Inference,
     OpenAIMessageParam,
@@ -61,12 +66,14 @@ class OpenAIResponsesImpl:
         tool_runtime_api: ToolRuntime,
         responses_store: ResponsesStore,
         vector_io_api: VectorIO,  # VectorIO
+        conversations_api: Conversations,
     ):
         self.inference_api = inference_api
         self.tool_groups_api = tool_groups_api
         self.tool_runtime_api = tool_runtime_api
         self.responses_store = responses_store
         self.vector_io_api = vector_io_api
+        self.conversations_api = conversations_api
         self.tool_executor = ToolExecutor(
             tool_groups_api=tool_groups_api,
             tool_runtime_api=tool_runtime_api,
@@ -205,6 +212,7 @@ class OpenAIResponsesImpl:
         model: str,
         instructions: str | None = None,
         previous_response_id: str | None = None,
+        conversation: str | None = None,
         store: bool | None = True,
         stream: bool | None = False,
         temperature: float | None = None,
@@ -221,11 +229,27 @@ class OpenAIResponsesImpl:
         if shields is not None:
             raise NotImplementedError("Shields parameter is not yet implemented in the meta-reference provider")
 
+        if conversation is not None and previous_response_id is not None:
+            raise ValueError(
+                "Mutually exclusive parameters: 'previous_response_id' and 'conversation'. Ensure you are only providing one of these parameters."
+            )
+
+        original_input = input  # needed for syncing to Conversations
+        if conversation is not None:
+            if not conversation.startswith("conv_"):
+                raise InvalidConversationIdError(conversation)
+
+            # Check conversation exists (raises ConversationNotFoundError if not)
+            _ = await self.conversations_api.get_conversation(conversation)
+            input = await self._load_conversation_context(conversation, input)
+
         stream_gen = self._create_streaming_response(
             input=input,
+            original_input=original_input,
             model=model,
             instructions=instructions,
             previous_response_id=previous_response_id,
+            conversation=conversation,
             store=store,
             temperature=temperature,
             text=text,
@@ -268,8 +292,10 @@ class OpenAIResponsesImpl:
         self,
         input: str | list[OpenAIResponseInput],
         model: str,
+        original_input: str | list[OpenAIResponseInput] | None = None,
         instructions: str | None = None,
         previous_response_id: str | None = None,
+        conversation: str | None = None,
         store: bool | None = True,
         temperature: float | None = None,
         text: OpenAIResponseText | None = None,
@@ -296,7 +322,7 @@ class OpenAIResponsesImpl:
         )
 
         # Create orchestrator and delegate streaming logic
-        response_id = f"resp-{uuid.uuid4()}"
+        response_id = f"resp_{uuid.uuid4()}"
         created_at = int(time.time())
 
         orchestrator = StreamingResponseOrchestrator(
@@ -319,13 +345,102 @@ class OpenAIResponsesImpl:
                 failed_response = stream_chunk.response
             yield stream_chunk
 
-        # Store the response if requested
-        if store and final_response and failed_response is None:
-            await self._store_response(
-                response=final_response,
-                input=all_input,
-                messages=orchestrator.final_messages,
-            )
+            # Store and sync immediately after yielding terminal events
+            # This ensures the storage/syncing happens even if the consumer breaks early
+            if (
+                stream_chunk.type in {"response.completed", "response.incomplete"}
+                and store
+                and final_response
+                and failed_response is None
+            ):
+                await self._store_response(
+                    response=final_response,
+                    input=all_input,
+                    messages=orchestrator.final_messages,
+                )
+
+            if stream_chunk.type in {"response.completed", "response.incomplete"} and conversation and final_response:
+                # for Conversations, we need to use the original_input if it's available, otherwise use input
+                sync_input = original_input if original_input is not None else input
+                await self._sync_response_to_conversation(conversation, sync_input, final_response)
 
     async def delete_openai_response(self, response_id: str) -> OpenAIDeleteResponseObject:
         return await self.responses_store.delete_response_object(response_id)
+
+    async def _load_conversation_context(
+        self, conversation_id: str, content: str | list[OpenAIResponseInput]
+    ) -> list[OpenAIResponseInput]:
+        """Load conversation history and merge with provided content."""
+        conversation_items = await self.conversations_api.list(conversation_id, order="asc")
+
+        context_messages = []
+        for item in conversation_items.data:
+            if isinstance(item, OpenAIResponseMessage):
+                if item.role == "user":
+                    context_messages.append(
+                        OpenAIResponseMessage(
+                            role="user", content=item.content, id=item.id if hasattr(item, "id") else None
+                        )
+                    )
+                elif item.role == "assistant":
+                    context_messages.append(
+                        OpenAIResponseMessage(
+                            role="assistant", content=item.content, id=item.id if hasattr(item, "id") else None
+                        )
+                    )
+
+        # add new content to context
+        if isinstance(content, str):
+            context_messages.append(OpenAIResponseMessage(role="user", content=content))
+        elif isinstance(content, list):
+            context_messages.extend(content)
+
+        return context_messages
+
+    async def _sync_response_to_conversation(
+        self, conversation_id: str, content: str | list[OpenAIResponseInput], response: OpenAIResponseObject
+    ) -> None:
+        """Sync content and response messages to the conversation."""
+        conversation_items = []
+
+        # add user content message(s)
+        if isinstance(content, str):
+            conversation_items.append(
+                {"type": "message", "role": "user", "content": [{"type": "input_text", "text": content}]}
+            )
+        elif isinstance(content, list):
+            for item in content:
+                if not isinstance(item, OpenAIResponseMessage):
+                    raise NotImplementedError(f"Unsupported input item type: {type(item)}")
+
+                if item.role == "user":
+                    if isinstance(item.content, str):
+                        conversation_items.append(
+                            {
+                                "type": "message",
+                                "role": "user",
+                                "content": [{"type": "input_text", "text": item.content}],
+                            }
+                        )
+                    elif isinstance(item.content, list):
+                        conversation_items.append({"type": "message", "role": "user", "content": item.content})
+                    else:
+                        raise NotImplementedError(f"Unsupported user message content type: {type(item.content)}")
+                elif item.role == "assistant":
+                    if isinstance(item.content, list):
+                        conversation_items.append({"type": "message", "role": "assistant", "content": item.content})
+                    else:
+                        raise NotImplementedError(f"Unsupported assistant message content type: {type(item.content)}")
+                else:
+                    raise NotImplementedError(f"Unsupported message role: {item.role}")
+
+        # add assistant response message
+        for output_item in response.output:
+            if isinstance(output_item, OpenAIResponseMessage) and output_item.role == "assistant":
+                if hasattr(output_item, "content") and isinstance(output_item.content, list):
+                    conversation_items.append({"type": "message", "role": "assistant", "content": output_item.content})
+
+        if conversation_items:
+            adapter = TypeAdapter(list[ConversationItem])
+            validated_items = adapter.validate_python(conversation_items)
+            await self.conversations_api.add_items(conversation_id, validated_items)

llama_stack/providers/registry/agents.py

@@ -35,6 +35,7 @@ def available_providers() -> list[ProviderSpec]:
                 Api.vector_dbs,
                 Api.tool_runtime,
                 Api.tool_groups,
+                Api.conversations,
             ],
             optional_api_dependencies=[
                 Api.telemetry,