Spaces:
Sleeping
Sleeping
| import os | |
| from langchain.agents import AgentExecutor, create_structured_chat_agent | |
| from langchain.memory import ConversationBufferMemory | |
| from langchain_core.messages import AIMessage | |
| from langchain_core.runnables.config import RunnableConfig | |
| from langchain_openai import AzureChatOpenAI | |
| from src import tools | |
| from src.prompts import solver_prompt | |
| from src.state import AgentState | |
| # This node uses the REACT framework to solve subqueries | |
| tools_list = [getattr(tools, name) for name in tools.__all__] | |
| async def solver_node(state: AgentState): | |
| """Executes a single step of the agent's problem-solving process using | |
| a ReAct agent. | |
| This function retrieves the current subquery from the agent's state, | |
| constructs a prompt including the context of previous results, and invokes | |
| the agent to generate a response. The agent's output is then appended to | |
| the results, the current step is incremented, and the new message is added | |
| to the conversation history. | |
| Args: | |
| state (AgentState): The current state of the agent, containing | |
| subqueries, results, current step index, and | |
| message history. | |
| Returns: | |
| dict: An updated state dictionary with the new result, incremented | |
| step, and updated messages. | |
| """ | |
| # Create the agent with the REACT framework | |
| agent = create_structured_chat_agent( | |
| AzureChatOpenAI(deployment_name=os.getenv( | |
| "AZURE_OPENAI_DEPLOYMENT_NAME")), | |
| tools_list, | |
| solver_prompt | |
| ) | |
| # Set up memory to retain conversation context | |
| memory = ConversationBufferMemory( | |
| memory_key="chat_history", return_messages=True) | |
| # Create the agent executor | |
| agent_executor = AgentExecutor( | |
| agent=agent, tools=tools_list, memory=memory, | |
| verbose=True, handle_parsing_errors=True) | |
| # Set the recursion limit for the agent | |
| config = RunnableConfig( | |
| recursion_limit=32 | |
| ) | |
| # Get the current subquery | |
| current_subquery = state["subqueries"][state["current_step"]] | |
| # Add previous results to context | |
| context = "\n".join( | |
| [f"Step {i+1}: {res}" for i, res in enumerate(state["results"])]) | |
| full_query = ( | |
| f"Original Query: {state['original_query']}\n\n" | |
| f"Context:\n{context}\n\n" | |
| f"Current Task: {current_subquery}" | |
| ) | |
| # Invoke the agent with the full query | |
| response = await agent_executor.ainvoke( | |
| {"input": full_query}, config) | |
| output = response['output'] | |
| return { | |
| **state, | |
| "results": state["results"] + [output], | |
| "current_step": state["current_step"] + 1, | |
| "messages": state["messages"] + [AIMessage(content=output)] | |
| } | |