fix(nemo-agents): Support Studio multi-turn transcript replay for Fabric agents (#1015)

* interim workaround

Signed-off-by: Manjesh Mogallapalli <mmogallapall@nvidia.com>

* lint

Signed-off-by: Manjesh Mogallapalli <mmogallapall@nvidia.com>

* remove ticket number

Signed-off-by: Manjesh Mogallapalli <mmogallapall@nvidia.com>

---------

Signed-off-by: Manjesh Mogallapalli <mmogallapall@nvidia.com>

Author: mmogallapalli

plugins/nemo-agents/src/nemo_agents_plugin/fabric/server.py

@@ -74,9 +74,16 @@ def _to_fabric_invocation_request(
     *,
     session_id: str,
 ) -> FabricInvocationRequest:
-    """Translate the current chat turn into a Platform-owned Fabric request."""
+    """Translate the chat transcript into a Platform-owned Fabric request."""
+    messages = request.messages
+    # Interim behavior to get multi-turn in Studio chat.
+    input_text = (
+        messages[0].content
+        if len(messages) == 1
+        else "\n\n".join(f"{message.role}: {message.content}" for message in messages)
+    )
     return FabricInvocationRequest(
-        input=request.messages[-1].content,
+        input=input_text,
         caller_context={"session_id": session_id},
     )
 

plugins/nemo-agents/tests/unit/test_fabric_server.py

@@ -570,7 +570,22 @@ async def invoke_session(session: Any, request: Any) -> FabricRuntimeResult:
     assert response.json() == {"detail": "adapter failed"}
 
 
-def test_chat_completion_request_translates_final_user_turn() -> None:
+def test_chat_completion_request_preserves_single_user_turn() -> None:
+    request = ChatCompletionRequest.model_validate(
+        {
+            "messages": [{"role": "user", "content": "Say hello."}],
+            "model": "test-model",
+            "stream": False,
+        }
+    )
+
+    invocation_request = server._to_fabric_invocation_request(request, session_id="session-1")
+
+    assert invocation_request.input == "Say hello."
+    assert invocation_request.caller_context == {"session_id": "session-1"}
+
+
+def test_chat_completion_request_serializes_full_transcript() -> None:
     request = ChatCompletionRequest.model_validate(
         {
             "messages": [
@@ -585,7 +600,7 @@ def test_chat_completion_request_translates_final_user_turn() -> None:
 
     invocation_request = server._to_fabric_invocation_request(request, session_id="session-1")
 
-    assert invocation_request.input == "Say hello."
+    assert invocation_request.input == ("system: Be concise.\n\nassistant: How can I help?\n\nuser: Say hello.")
     assert invocation_request.caller_context == {"session_id": "session-1"}