feat: support multi-model chat

backend/onyx/chat/models.py

@@ -102,6 +102,14 @@ class MessageResponseIDInfo(BaseModel):
     reserved_assistant_message_id: int
 
 
+class MultiModelMessageResponseIDInfo(BaseModel):
+    """Response info for multi-model chat: one user message, multiple assistant messages."""
+
+    user_message_id: int | None
+    reserved_assistant_message_ids: list[int]  # One per model
+    model_names: list[str]  # Display names for UI
+
+
 class StreamingError(BaseModel):
     error: str
     stack_trace: str | None = None
@@ -200,6 +208,7 @@ class LLMMetricsContainer(BaseModel):
     Packet
     | StreamStopInfo
     | MessageResponseIDInfo
+    | MultiModelMessageResponseIDInfo
     | StreamingError
     | UserKnowledgeFilePacket
     | CreateChatSessionID

backend/onyx/server/query_and_chat/models.py

@@ -83,6 +83,8 @@ class SendMessageRequest(BaseModel):
     message: str
 
     llm_override: LLMOverride | None = None
+    # For multi-model chat: list of LLM overrides to compare (2-3 models supported)
+    llm_overrides: list[LLMOverride] | None = None
 
     allowed_tool_ids: list[int] | None = None
     forced_tool_id: int | None = None

backend/onyx/server/query_and_chat/placement.py

@@ -8,3 +8,5 @@ class Placement(BaseModel):
     tab_index: int = 0
     # Used for tools/agents that call other tools, this currently doesn't support nested agents but can be added later
     sub_turn_index: int | None = None
+    # For multi-model chat: identifies which model's response this packet belongs to (0, 1, or 2)
+    model_index: int | None = None