Update agent.py

agent.py

@@ -1,7 +1,7 @@
 import os
 import re
 from groq import Groq
-from duckduckgo_search import DDGS
+from ddgs import DDGS
 from bs4 import BeautifulSoup
 import requests
 from utils import BaseAgent, SimpleRateLimiter
@@ -11,21 +11,28 @@ class GaiaAgent(BaseAgent):
     
     def __init__(self):
         self.client = Groq(api_key=os.environ.get("GROQ_API_KEY"))
-        self.model = "mistral/mistral-small-latest"
+        # Modèles Groq valides (en ordre de préférence)
+        self.models = [
+            "openai/gpt-oss-120b",      # ← en priorité
+            "qwen/qwen3.6-27b",          # fallback 1
+            "llama-3.3-70b-versatile",   # fallback 2
+            "llama-3.1-8b-instant"       # fallback 3
+        ]
         self.rate_limiter = SimpleRateLimiter()
+        self.current_model = self.models[0]
     
     def web_search(self, query, max_results=3):
-        """Search the web using DuckDuckGo"""
+        """Search the web using DDGS (DuckDuckGo)"""
         print(f"[web_search] query: {query[:80]}...")
         results = []
         try:
-            with DDGS() as ddgs:
-                for r in ddgs.text(query, max_results=max_results):
-                    results.append({
-                        "title": r.get("title"),
-                        "url": r.get("href"),
-                        "snippet": r.get("body")
-                    })
+            ddgs = DDGS()
+            for r in ddgs.text(query, max_results=max_results):
+                results.append({
+                    "title": r.get("title"),
+                    "url": r.get("href"),
+                    "snippet": r.get("body")
+                })
         except Exception as e:
             print(f"[web_search] error: {e}")
         return results
@@ -104,7 +111,7 @@ IMPORTANT: Your response MUST end with a line starting with "FINAL ANSWER:" foll
 - For numbers, use exact format requested
 - For names, use exact spelling
 
-Reasoning:
+Question:
 {question}
 
 Context:
@@ -114,34 +121,52 @@ Remember to end with:
 FINAL ANSWER: <answer>
 """
         
-        print(f"[agent] calling {self.model}...")
-        try:
-            response = self.client.chat.completions.create(
-                model=self.model,
-                messages=[{"role": "user", "content": prompt}],
-                temperature=0,
-                max_tokens=1000
-            )
-            output = response.choices[0].message.content
-            print(f"[agent] got response ({len(output)} chars)")
-        except Exception as e:
-            print(f"[agent] error: {e}")
-            return "I am unable to answer"
-        
-        answer = self.extract_answer(output)
+        # Try models in order with fallback
+        answer = ""
+        for model in self.models:
+            self.current_model = model
+            print(f"[agent] trying model: {model}...")
+            try:
+                response = self.client.chat.completions.create(
+                    model=model,
+                    messages=[{"role": "user", "content": prompt}],
+                    temperature=0,
+                    max_tokens=1000
+                )
+                output = response.choices[0].message.content
+                print(f"[agent] ✓ got response ({len(output)} chars)")
+                answer = self.extract_answer(output)
+                
+                if answer and len(answer) > 0:
+                    print(f"[agent] ✓ got answer from {model}")
+                    break
+                else:
+                    print(f"[agent] ✗ model {model} returned empty answer, trying next...")
+                    
+            except Exception as e:
+                error_msg = str(e)
+                if "does not exist" in error_msg or "not found" in error_msg:
+                    print(f"[agent] ✗ model {model} not found, trying next...")
+                elif "overload" in error_msg.lower() or "rate limit" in error_msg.lower():
+                    print(f"[agent] ✗ rate limit on {model}, trying next...")
+                else:
+                    print(f"[agent] ✗ error with {model}: {e}")
+                continue
         
-        # Retry without context if empty
         if not answer or len(answer) == 0:
-            print(f"[agent] retrying without context...")
+            print(f"[agent] retrying with shorter prompt...")
             self.rate_limiter.wait_if_needed()
-            response = self.client.chat.completions.create(
-                model=self.model,
-                messages=[{"role": "user", "content": f"Question: {question}\n\nFINAL ANSWER:"}],
-                temperature=0,
-                max_tokens=500
-            )
-            output = response.choices[0].message.content
-            answer = self.extract_answer(output)
+            try:
+                response = self.client.chat.completions.create(
+                    model=self.models[0],
+                    messages=[{"role": "user", "content": f"Answer briefly:\n{question}\n\nFINAL ANSWER:"}],
+                    temperature=0,
+                    max_tokens=200
+                )
+                output = response.choices[0].message.content
+                answer = self.extract_answer(output)
+            except Exception as e:
+                print(f"[agent] retry error: {e}")
         
         if not answer:
             answer = "I am unable to answer"
@@ -151,4 +176,4 @@ FINAL ANSWER: <answer>
         return answer
     
     def __call__(self, question: str, file_content: str = "") -> str:
-        return self.run(question, file_content)
+        return self.run(question, file_content)