Changed to mini models of openai to make it faster

agents/llama_index_agent.py

@@ -281,6 +281,8 @@ def create_writer_agent(model_config: Dict[str, Any]) -> ReActAgent:
         Action Input: {"to_agent": "review_agent", "reason": "Need final verification and cleanup", "query": "<original question>", "formatted_answer": "<my formatted answer>", "format_requirements": "<explicit formatting requirements>"}
 
         CRITICAL: Even if the answer seems perfectly formatted already, you MUST use the handoff tool with the EXACT format above. NEVER respond with "Answer: ..." as this breaks the workflow.
+        Action: handoff
+        Action Input: {"to_agent": "review_agent", "reason": "Final answer to be sent", "query": "<original question>", "formatted_answer": "<my formatted answer>", "format_requirements": "<explicit formatting requirements>"}
         """,
         llm=llm,
         can_handoff_to=["review_agent"],
@@ -375,7 +377,11 @@ def create_review_agent(model_config: Dict[str, Any]) -> ReActAgent:
           Output: 5
 
         REMEMBER: Your ENTIRE response should be just the bare answer with NOTHING else.
+
+        Hand the answer off to `main_agent` generate the final answer
+        Action: handoff
+        Action Input: {"to_agent": "jefe", "reason": "Final answer to be sent", "query": "<original question>", "formatted_answer": "<my formatted answer>", "format_requirements": "<explicit formatting requirements>"}
         """,
         llm=llm,
-        can_handoff_to=[]  # Review agent is the final step, so it doesn't hand off to anyone
+        can_handoff_to=["jefe"],
     )