increase models

agent.py

@@ -70,41 +70,66 @@ gemini_model = ChatGoogleGenerativeAI(
     temperature=0,
 )
 
-def smart_invoke(msgs, use_tools=False):
+def smart_invoke(msgs, use_tools=False, start_tier=0):
     """
-    Tiered fallback: Groq -> OpenRouter -> Google AI Studio.
-    Retries next tier if a 429 (rate limit) or server-side error occurs.
+    Tiered fallback: OpenRouter -> Gemini -> Groq.
+    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
     
+    # Adaptive Gemini names to try if 1.5 flash is 404
+    gemini_alternatives = ["gemini-2.5-flash", "gemini-2.5-flash-lite", "gemini-3.1-flash-lite", "gemini-3-flash"]
+    
     tiers = [
         {"name": "OpenRouter", "model": secondary, "key": "OPENROUTER_API_KEY"},
-        {"name": "Gemini", "model": tertiary, "key": "GOOGLE_API_KEY"},
+        {"name": "Gemini", "model": tertiary, "key": "GOOGLE_API_KEY", "alternatives": gemini_alternatives},
         {"name": "Groq", "model": primary, "key": "GROQ_API_KEY"},
     ]
     
     last_exception = None
-    for tier in tiers:
+    for i in range(start_tier, len(tiers)):
+        tier = tiers[i]
         if not os.getenv(tier["key"]):
-            continue # Skip if no API key
+            continue 
             
-        try:
-            return tier["model"].invoke(msgs)
-        except Exception as e:
-            err_str = str(e).lower()
-            # Catch rate limits, generic temporary server failures, or missing models
-            if any(x in err_str for x in ["rate_limit", "429", "500", "503", "overloaded", "not_found", "404"]):
-                print(f"--- {tier['name']} Error: {e}. Falling back... ---")
-                last_exception = e
-                continue
-            raise e
+        # For tiers with alternatives (like Gemini), try each if 404 occurs
+        models_to_try = [tier["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)
+                models_to_try.append(alt_model)
+
+        for current_model in models_to_try:
+            try:
+                model_name = getattr(current_model, "model", tier["name"])
+                print(f"--- Calling {tier['name']} ({model_name}) ---")
+                return current_model.invoke(msgs), i
+            except Exception as e:
+                err_str = str(e).lower()
+                # If it's a 404 (not found) and we have more alternatives, continue to the next alternative
+                if any(x in err_str for x in ["not_found", "404"]) and current_model != models_to_try[-1]:
+                    print(f"--- {tier['name']} model {model_name} not found. Trying alternative... ---")
+                    continue
+                
+                # Catch other fallback triggers
+                if any(x in err_str for x in ["rate_limit", "429", "500", "503", "overloaded", "not_found", "404", "402", "credits"]):
+                    print(f"--- {tier['name']} Error: {e}. Falling back... ---")
+                    last_exception = e
+                    break # Move to next tier
+                raise e
+            
+    if last_exception:
+        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
+    return None, 0
 
 @tool
 def web_search(keywords: str) -> str:
@@ -424,6 +449,7 @@ def answer_message(state: AgentState) -> AgentState:
     # Multi-step ReAct Loop (Up to 8 reasoning steps)
     max_steps = 8
     draft_response = None
+    current_tier = 0
     
     for step in range(max_steps):
         if step > 0:
@@ -431,7 +457,7 @@ def answer_message(state: AgentState) -> AgentState:
             time.sleep(4)
             
         print(f"--- ReAct Step {step + 1} ---")
-        ai_msg = smart_invoke(messages, use_tools=True)
+        ai_msg, current_tier = smart_invoke(messages, use_tools=True, start_tier=current_tier)
         messages.append(ai_msg)
 
         # Check if the model requested tools
@@ -475,7 +501,7 @@ def answer_message(state: AgentState) -> AgentState:
             "If it is a name or word, just return the exact string. If a list, return only the comma-separated list."
         )
     )
-    final_response = smart_invoke([formatting_sys, HumanMessage(content=draft_response.content)], use_tools=False)
+    final_response, _ = smart_invoke([formatting_sys, HumanMessage(content=draft_response.content)], use_tools=False, start_tier=current_tier)
     print(f"Draft response: {draft_response.content}")
     print(f"Strict Final response: {final_response.content}")