Add model selection and lab updates

backend/agent/graph.py

@@ -48,7 +48,7 @@ class CANIFAGraph:
         self.collection_tools = get_collection_tools()  # Vẫn lấy list name để routing
         self.retrieval_tools = self.all_tools
 
-        self.llm_with_tools = self.llm.bind_tools(self.all_tools, strict=True)
+        self.llm_with_tools = self.llm.bind_tools(self.all_tools)  # No strict: compat with both OpenAI & Gemini schemas
         self.cache = InMemoryCache()
 
         # Chain caching: avoid rebuilding ChatPromptTemplate every turn
@@ -185,13 +185,12 @@ class CANIFAGraph:
         return self.build()
 
 
-# --- Singleton & Public API ---
-_instance: list[CANIFAGraph | None] = [None]
+# --- Per-model Instance Cache & Public API ---
+_instances: dict[str, CANIFAGraph] = {}
 
 
 def build_graph(config: AgentConfig | None = None, llm: BaseChatModel | None = None, tools: list | None = None) -> Any:
-    """Get compiled graph (Singleton usage)."""
-    # Use singleton to avoid rebuilding graph on every request
+    """Get compiled graph (cached per model)."""
     manager = get_graph_manager(config, llm, tools)
     return manager.build()
 
@@ -199,38 +198,33 @@ def build_graph(config: AgentConfig | None = None, llm: BaseChatModel | None = N
 def get_graph_manager(
     config: AgentConfig | None = None, llm: BaseChatModel | None = None, tools: list | None = None
 ) -> CANIFAGraph:
-    """Get CANIFAGraph instance (Auto-rebuild if model config changes).
+    """Get CANIFAGraph instance per model_name.
     
-    Prompt is now fetched dynamically per request from Langfuse,
-    so no need to rebuild graph when prompt changes.
+    Each model gets its own cached graph instance, enabling true parallel
+    execution when Lab sends requests to different models simultaneously.
     """
-    # 1. New Instance if Empty
-    if _instance[0] is None:
-        _instance[0] = CANIFAGraph(config, llm, tools)
-        logger.info(f"✨ Graph Created: {_instance[0].config.model_name} (prompts from Langfuse)")
-        return _instance[0]
+    effective_config = config or get_config()
+    model_key = effective_config.model_name
 
-    # 2. Check for Model Config Changes only
-    is_model_changed = config and config.model_name != _instance[0].config.model_name
+    if model_key not in _instances:
+        _instances[model_key] = CANIFAGraph(effective_config, llm, tools)
+        logger.info(f"✨ Graph Created: {model_key} (prompts from Langfuse) | Total cached: {len(_instances)}")
     
-    if is_model_changed:
-        logger.info(f"🔄 Rebuilding Graph: Model ({_instance[0].config.model_name}->{config.model_name})")
-        _instance[0] = CANIFAGraph(config, llm, tools)
-        return _instance[0]
-
-    return _instance[0]
+    return _instances[model_key]
 
 
 def reset_graph() -> None:
-    """Reset singleton for testing."""
-    _instance[0] = None
+    """Reset all cached instances (for testing)."""
+    _instances.clear()
 
 
 def reset_chain_cache() -> None:
-    """Reset only the cached chain (when prompt changes).
+    """Reset only the cached chain for all instances (when prompt changes).
     Keeps the graph/LLM/tools intact, only forces chain rebuild on next request.
     """
-    if _instance[0] is not None:
-        _instance[0]._cached_chain = None
-        _instance[0]._cached_prompt_hash = None
-        logger.info("🔄 Chain cache cleared — will rebuild on next request")
+    for model_key, inst in _instances.items():
+        inst._cached_chain = None
+        inst._cached_prompt_hash = None
+    if _instances:
+        logger.info(f"🔄 Chain cache cleared for {len(_instances)} model(s) — will rebuild on next request")
+

backend/agent/models.py

@@ -14,6 +14,7 @@ class QueryRequest(BaseModel):
     user_query: str
     images: list[str] | None = None
     image_analysis: dict[str, Any] | None = None
+    model_name: str | None = None  # Override model per-request (Lab mode)
 
 
 class AgentState(TypedDict):

backend/agent/tools/data_retrieval_tool.py

@@ -31,7 +31,7 @@ from agent.prompt_utils import read_tool_prompt
 
 
 class SearchItem(BaseModel):
-    model_config = {"extra": "ignore"}  # Gemini may send extra fields
+    model_config = {"extra": "ignore", "json_schema_extra": {"additionalProperties": False}}  # ignore for Gemini compat; additionalProperties for OpenAI strict mode
 
     # ====== SEARCH TEXT (optional fallback) ======
     description: str | None = Field(
@@ -121,7 +121,7 @@ class SearchItem(BaseModel):
 
 
 class MultiSearchParams(BaseModel):
-    model_config = {"extra": "ignore"}  # Gemini may send extra fields
+    model_config = {"extra": "ignore", "json_schema_extra": {"additionalProperties": False}}  # ignore for Gemini compat; additionalProperties for OpenAI strict mode
     searches: list[SearchItem] = Field(description="Danh sách các truy vấn tìm kiếm")
 
 

backend/api/chatbot_route.py

@@ -106,10 +106,12 @@ async def fashion_qa_chat_dev(request: Request, req: QueryRequest, background_ta
 
     try:
         # DEV MODE: Return ai_response + products immediately
+        # Lab mode: allow model override from request body
+        effective_model = req.model_name or DEFAULT_MODEL
         result = await chat_controller(
             query=req.user_query,
             background_tasks=background_tasks,
-            model_name=DEFAULT_MODEL,
+            model_name=effective_model,
             images=req.images,
             identity_key=str(identity_id),
             return_user_insight=False,