Update app.py

app.py

@@ -9,7 +9,7 @@ from llama_index.llms.huggingface_api import HuggingFaceInferenceAPI
 # 1. SETUP LLM
 hf_token = os.getenv("HF_TOKEN")
 
-# We use the 7B model because it's faster and more available on the free tier
+# We use the 7B model because it's much faster and reliable for the free tier
 llm = HuggingFaceInferenceAPI(
     model_name="Qwen/Qwen2.5-7B-Instruct", 
     token=hf_token,
@@ -32,22 +32,35 @@ tools = [
     FunctionTool.from_defaults(fn=multiply)
 ]
 
-# 3. CREATE THE AGENT
-# The system prompt ensures the AI follows the ReAct pattern
+# 3. THE "STABILITY" PROMPT
+# This prevents the AI from freezing by giving it a clear pattern to follow.
+RE_ACT_PROMPT = """You are a helpful assistant.
+For every query, you MUST follow this sequence:
+Thought: <your reasoning>
+Action: <tool_name>
+Action Input: {"arg1": value}
+Observation: <result from tool>
+... (repeat if needed)
+Thought: I have the final answer.
+Answer: <your final response to the user>
+"""
+
+# 4. CREATE THE AGENT
 agent = ReActAgent.from_tools(
     tools, 
     llm=llm, 
-    verbose=True
+    verbose=True,
+    context=RE_ACT_PROMPT
 )
 
-# 4. GRADIO INTERFACE
+# 5. GRADIO INTERFACE
 def chat(message, history):
     try:
-        # Use .chat() to maintain memory
+        # Use .chat() to maintain the conversation flow
         response = agent.chat(message)
         return str(response)
     except Exception as e:
-        # This will show you exactly what is failing in the Gradio UI
-        return f"Error: {str(e)}"
+        # If it still fails, this will tell us WHY in the chat window
+        return f"System Error: {str(e)}"
 
-gr.ChatInterface(chat, title="Unit 2: LlamaIndex Agent").launch()
+gr.ChatInterface(chat, title="Unit 2: LlamaIndex Agent (Fixed)").launch()