Update app.py

app.py

@@ -259,8 +259,13 @@ from typing import List
 from langchain_core.messages import BaseMessage
 from langchain_core.tools import tool
 from langchain_cohere.chat_models import ChatCohere
-from langchain.agents import AgentExecutor, create_cohere_tools_agent
+from langchain.agents import AgentExecutor
 from langchain_core.prompts import ChatPromptTemplate
+# These are the fundamental components we need for a Cohere Tools agent
+from langchain.agents.format_scratchpad.cohere import format_cohere_tools
+from langchain.agents.output_parsers.cohere import CohereToolsAgentOutputParser
+
+
 from tavily import TavilyClient
 import pypdf
 
@@ -334,27 +339,39 @@ def python_interpreter(code: str) -> str:
 
 #
 # ================================================================================================
-#  ✅ 2. CONFIGURE AND BUILD THE AGENT (Standard, Robust Method)
+#  ✅ 2. CONFIGURE AND BUILD THE AGENT (Manual, Stable Method)
 # ================================================================================================
 #
 def build_agent_graph():
-    """Builds the agent using the standard create_cohere_tools_agent function."""
+    """Builds the agent using the most fundamental LangChain components."""
     tools = [tavily_search, read_file, python_interpreter]
 
-    # Instantiate the ChatCohere model
+    # 1. Create the ChatCohere model instance
     llm = ChatCohere(model="command-r-plus", temperature=0, cohere_api_key=os.getenv("COHERE_API_KEY"))
 
-    # Create the prompt template. The 'agent_scratchpad' is crucial.
+    # 2. Bind the tools to the LLM. This lets the LLM know about the tools.
+    llm_with_tools = llm.bind_tools(tools)
+
+    # 3. Create the prompt template. This is the core instruction for the agent.
     prompt = ChatPromptTemplate.from_messages([
         ("system", AGENT_SYSTEM_PROMPT),
-        ("human", "{input}"),
-        ("placeholder", "{agent_scratchpad}"),
+        ("user", "{input}"),
+        ("placeholder", "{agent_scratchpad}"), # This is where tool results will be injected.
     ])
 
-    # Use the standard agent creation function
-    agent = create_cohere_tools_agent(llm, tools, prompt)
-    
-    # Create the executor to run the agent loop
+    # 4. Define the agent runnable. This is a chain that pipes components together.
+    #    It formats the input, sends it to the LLM, and parses the output.
+    agent = (
+        {
+            "input": lambda x: x["input"],
+            "agent_scratchpad": lambda x: format_cohere_tools(x["intermediate_steps"]),
+        }
+        | prompt
+        | llm_with_tools
+        | CohereToolsAgentOutputParser()
+    )
+
+    # 5. Create the AgentExecutor to run the agent-tool loop.
     agent_executor = AgentExecutor(agent=agent, tools=tools, verbose=True)
 
     return agent_executor
@@ -366,17 +383,14 @@ def build_agent_graph():
 #
 class GaiaAgent:
     def __init__(self):
-        print("GaiaAgent initialized. Building agent with create_cohere_tools_agent...")
+        print("GaiaAgent initialized. Building agent with fundamental LangChain components...")
         self.agent_app = build_agent_graph()
 
     def __call__(self, question: str) -> str:
         print(f"\n{'='*60}\nAgent received question: {question[:100]}...\n{'='*60}")
         try:
-            # The standard agent executor expects 'input' and 'chat_history'.
-            response = self.agent_app.invoke({
-                "input": question,
-                "chat_history": [] # Each question is independent, so history is empty.
-            })
+            # The standard agent executor expects 'input'.
+            response = self.agent_app.invoke({"input": question})
             final_answer = str(response.get("output", "")).strip()
             print(f"\n--- Agent finished. Final Answer: {final_answer} ---\n")
             return final_answer
@@ -430,13 +444,13 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
             f"Message: {result_data.get('message', 'No message received.')}"
         )
         return final_status, pd.DataFrame(results_log)
-    except Exception as e: return f"An unexpected error occurred during submission: {e}", pd.DataFrame(results_log)
+    except Exception as e: return f"An unexpected error in submission: {e}", pd.DataFrame(results_log)
 
 with gr.Blocks() as demo:
     gr.Markdown("# GAIA Agent Final Assessment (Direct Cohere Integration)")
     gr.Markdown(
         """
-        **Instructor's Note:** This version uses the standard `create_cohere_tools_agent` function, which is the most robust way to create a tool-calling agent with Cohere.
+        **Instructor's Note:** This version uses a fundamental, manual agent construction. It is the most stable and recommended approach, avoiding any version-specific helper functions.
         1.  Ensure you have a **`COHERE_API_KEY`** and a **`TAVILY_API_KEY`** set in your Space secrets.
         2.  Ensure your `requirements.txt` includes `langchain-cohere` and `langchain`.
         """