update agent

agent.py

@@ -47,81 +47,79 @@ load_dotenv()
 #     huggingfacehub_api_token=os.getenv("HUGGINGFACEHUB_API_TOKEN"),
 # )
 
-# Groq Model (Primary)
-model = ChatGroq(
-    model="meta-llama/llama-4-scout-17b-16e-instruct",
-    temperature=0,
-    max_tokens=None,
-    timeout=None,
-    max_retries=2,
-)
-
-# OpenRouter Model (Primary Fallback)
-openrouter_model = ChatOpenAI(
-    model="meta-llama/llama-3.3-70b-instruct",
-    openai_api_key=os.getenv("OPENROUTER_API_KEY"),
-    openai_api_base="https://openrouter.ai/api/v1",
-    temperature=0,
-)
-
-# Google AI Studio Fallback Model (Gemini)
-gemini_model = ChatGoogleGenerativeAI(
-    model="gemini-2.5-flash",
-    # google_api_key is automatically picked up from GOOGLE_API_KEY environment variable
-    temperature=0,
-)
-
-# NVIDIA Model (Secondary Fallback)
-nvidia_model = ChatOpenAI(
-    model="meta/llama-3.1-405b-instruct",
-    openai_api_key=os.getenv("NVIDIA_API_KEY"),
-    openai_api_base="https://integrate.api.nvidia.com/v1",
-    temperature=0,
-)
-
-# Vercel Model (Tertiary Fallback)
-# Note: Adjust model and base_url if using a specific Vercel AI Gateway setup
-vercel_model = ChatOpenAI(
-    model="meta-llama/llama-3.3-70b-instruct",
-    openai_api_key=os.getenv("VERCEL_API_KEY"),
-    openai_api_base="https://gateway.ai.vercel.com/v1",
-    temperature=0,
-)
+# Model initializations moved to smart_invoke for lazy loading to prevent import errors if keys are missing.
 
 def smart_invoke(msgs, use_tools=False, start_tier=0):
     """
-    Tiered fallback: OpenRouter -> Gemini -> Groq.
+    Tiered fallback: OpenRouter -> Gemini -> Groq -> NVIDIA -> Vercel.
     Retries next tier if a 429 (rate limit), 402 (credits), or 404 (model found) error occurs.
     """
-    primary = model_with_tools if use_tools else model
-    secondary = openrouter_with_tools if use_tools else openrouter_model
-    tertiary = gemini_with_tools if use_tools else gemini_model
-    quaternary = nvidia_with_tools if use_tools else nvidia_model
-    quinary = vercel_with_tools if use_tools else vercel_model
     
     # Adaptive Gemini names to try if 1.5 flash is 404
     gemini_alternatives = ["gemini-2.5-flash-lite", "gemma-3-1b", "gemini-3-flash", "gemini-3.1-flash-lite"]
     
-    tiers = [
-        {"name": "OpenRouter", "model": secondary, "key": "OPENROUTER_API_KEY"},
-        {"name": "Gemini", "model": tertiary, "key": "GOOGLE_API_KEY", "alternatives": gemini_alternatives},
-        {"name": "Groq", "model": primary, "key": "GROQ_API_KEY"},
-        {"name": "NVIDIA", "model": quaternary, "key": "NVIDIA_API_KEY"},
-        {"name": "Vercel", "model": quinary, "key": "VERCEL_API_KEY"},
+    tiers_config = [
+        {"name": "OpenRouter", "key": "OPENROUTER_API_KEY", "provider": "openai", "model_name": "meta-llama/llama-3.3-70b-instruct", "base_url": "https://openrouter.ai/api/v1"},
+        {"name": "Gemini", "key": "GOOGLE_API_KEY", "provider": "google", "model_name": "gemini-2.5-flash", "alternatives": gemini_alternatives},
+        {"name": "Groq", "key": "GROQ_API_KEY", "provider": "groq", "model_name": "meta-llama/llama-4-scout-17b-16e-instruct"},
+        {"name": "NVIDIA", "key": "NVIDIA_API_KEY", "provider": "openai", "model_name": "meta/llama-3.1-405b-instruct", "base_url": "https://integrate.api.nvidia.com/v1"},
+        {"name": "Vercel", "key": "VERCEL_API_KEY", "provider": "openai", "model_name": "meta-llama/llama-3.3-70b-instruct", "base_url": "https://gateway.ai.vercel.com/v1"},
     ]
     
     last_exception = None
-    for i in range(start_tier, len(tiers)):
-        tier = tiers[i]
-        if not os.getenv(tier["key"]):
+    for i in range(start_tier, len(tiers_config)):
+        tier = tiers_config[i]
+        api_key = os.getenv(tier["key"])
+        if not api_key:
             continue 
             
-        # For tiers with alternatives (like Gemini), try each if 404 occurs
-        models_to_try = [tier["model"]]
+        def create_model_instance(m_name, provider, b_url=None):
+            if provider == "openai":
+                return ChatOpenAI(model=m_name, openai_api_key=api_key, openai_api_base=b_url, temperature=0)
+            elif provider == "google":
+                return ChatGoogleGenerativeAI(model=m_name, temperature=0)
+            elif provider == "groq":
+                return ChatGroq(model=m_name, temperature=0, max_retries=2)
+            return None
+
+        primary_model = create_model_instance(tier["model_name"], tier["provider"], tier.get("base_url"))
+        if use_tools:
+            primary_model = primary_model.bind_tools(tools)
+            
+        models_to_try = [primary_model]
         if "alternatives" in tier:
             for alt_name in tier["alternatives"]:
-                # Create a new model instance if the default one fails
-                alt_model = ChatGoogleGenerativeAI(model=alt_name, temperature=0).bind_tools(tools) if use_tools else ChatGoogleGenerativeAI(model=alt_name, temperature=0)
+                alt_model = create_model_instance(alt_name, tier["provider"], tier.get("base_url"))
+                if use_tools:
+                    alt_model = alt_model.bind_tools(tools)
+                models_to_try.append(alt_model)
+    
+    last_exception = None
+    for i in range(start_tier, len(tiers_config)):
+        tier = tiers_config[i]
+        api_key = os.getenv(tier["key"])
+        if not api_key:
+            continue 
+            
+        def create_model_instance(m_name, provider, b_url=None):
+            if provider == "openai":
+                return ChatOpenAI(model=m_name, openai_api_key=api_key, openai_api_base=b_url, temperature=0)
+            elif provider == "google":
+                return ChatGoogleGenerativeAI(model=m_name, temperature=0)
+            elif provider == "groq":
+                return ChatGroq(model=m_name, temperature=0, max_retries=2)
+            return None
+
+        primary_model = create_model_instance(tier["model_name"], tier["provider"], tier.get("base_url"))
+        if use_tools:
+            primary_model = primary_model.bind_tools(tools)
+            
+        models_to_try = [primary_model]
+        if "alternatives" in tier:
+            for alt_name in tier["alternatives"]:
+                alt_model = create_model_instance(alt_name, tier["provider"], tier.get("base_url"))
+                if use_tools:
+                    alt_model = alt_model.bind_tools(tools)
                 models_to_try.append(alt_model)
 
         for current_model in models_to_try:
@@ -150,11 +148,6 @@ def smart_invoke(msgs, use_tools=False, start_tier=0):
         print("CRITICAL: All fallback tiers failed.")
         raise last_exception
     return None, 0
-            
-    if last_exception:
-        print("CRITICAL: All fallback tiers failed.")
-        raise last_exception
-    return None, 0
 
 @tool
 def web_search(keywords: str) -> str:
@@ -400,7 +393,6 @@ Your answer should only start with "FINAL ANSWER: ", then follows with the answe
 class AgentState(TypedDict):
     messages: List[Union[HumanMessage, AIMessage, SystemMessage]]
 
-
 def read_message(state: AgentState) -> AgentState:
     messages = state["messages"]
     print(f"Processing question: {messages[-1].content if messages else ''}")
@@ -432,11 +424,7 @@ def restart_required(state: AgentState) -> AgentState:
 # Augment the LLM with tools
 tools = [web_search, wiki_search, analyze_image, analyze_audio, analyze_video, read_url, run_python_script, read_document]
 tools_by_name = {tool.name: tool for tool in tools}
-model_with_tools = model.bind_tools(tools)
-openrouter_with_tools = openrouter_model.bind_tools(tools)
-gemini_with_tools = gemini_model.bind_tools(tools)
-nvidia_with_tools = nvidia_model.bind_tools(tools)
-vercel_with_tools = vercel_model.bind_tools(tools)
+# model_with_tools etc. removed, replaced by lazy initialization in smart_invoke
 
 def answer_message(state: AgentState) -> AgentState:
     messages = state["messages"]