import os from dotenv import load_dotenv from typing import TypedDict, Optional, Annotated from langchain_google_genai import ChatGoogleGenerativeAI from langchain_mistralai import ChatMistralAI from langchain_groq import ChatGroq from langgraph.graph import StateGraph, START, END from langchain_core.messages import AnyMessage, HumanMessage from langgraph.graph.message import add_messages from langgraph.prebuilt import ToolNode, tools_condition from custom_tools import custom_tools class QuestionState(TypedDict): input_file: Optional[str] messages: Annotated[list[AnyMessage], add_messages] class NodesReActAgent: def __init__(self, provider: str="Google", model: str="gemini-2.5-pro"): print('Initializing ReActAgent...') load_dotenv() # Set up the LLM based on provider if provider == "Google": os.environ["GOOGLE_API_KEY"] = os.getenv("GOOGLE") llm = ChatGoogleGenerativeAI(model=model, temperature=0, max_retries=5) elif provider == "Mistral": os.environ["MISTRAL_API_KEY"] = os.getenv("MISTRAL") llm = ChatMistralAI(model=model, temperature=0, max_retries=5) elif provider == "Groq": os.environ["GROQ_API_KEY"] = os.getenv("GROQ") llm = ChatGroq(model=model, temperature=0, max_retries=5) else: raise ValueError(f"Unknown provider: {provider}") self.llm_with_tools = llm.bind_tools(custom_tools) def assistant(state: QuestionState): input_file = state["input_file"] sys_prompt = f""" You are a general AI assistant. I will ask you a question. Report your thoughts, and finish your answer with the following template: FINAL ANSWER: [YOUR FINAL ANSWER].\n \n YOUR FINAL ANSWER should be a number OR as few words as possible OR a comma separated list of numbers and/or strings. If you are asked for a number, DON'T use comma to write your number NEITHER use units such as $ or percent sign unless specified otherwise. If you are asked for a string, DON'T use articles, NEITHER abbreviations (e.g. for cities) capitalize the first letter, and write the digits in plain text unless specified otherwise. If you are asked for a comma separated list, apply the above rules depending, unless the first letter capitalization, whether the element to be put in the list is a number or a string.\n \n EXAMPLES:\n - What is US President Obama's first name? FINAL ANSWER: Barack\n - What are the 3 mandatory ingredients for pancakes? FINAL ANSWER: eggs, flour, milk\n - What is the final cost of an invoice comprising a $345.00 product and a $355.00 product? Provide the answer with two decimals. FINAL ANSWER: 700.00\n - How many pairs of chromosomes does a human cell contain? FINAL ANSWER : 23\n \n \n You will be provided with tools to help you answer questions.\n If you are asked to make a calculation, absolutely use the tools provided to you. You should AVOID calculating by yourself and ABSOLUTELY use appropriate tools.\n If you are asked to find something in a list of things or people, prefer using the wiki_search tool. Else, prefer to use the web_search tool. After using the web_search tool, look for the first URL provided with the url_search tool and ask yourself if the answer is in the tool response. If it is, answer the question. If not, search on other links.\n \n If needed, use one tool first, then use the output of that tool as an input to another thinking then to the use of another tool.\n \n \n You have access to some optional files. Currently the loaded file is: {input_file}" """ return { "messages": [self.llm_with_tools.invoke([sys_prompt] + state["messages"])], "input_file": state["input_file"] } # The graph builder = StateGraph(QuestionState) # Define nodes: these do the work builder.add_node("assistant", assistant) builder.add_node("tools", ToolNode(custom_tools)) builder.add_edge(START, "assistant") builder.add_conditional_edges( "assistant", # If the latest message requires a tool, route to "tools" # Otherwise, route to "END" and provide a direct response tools_condition, ) builder.add_edge("tools", "assistant") self.react_graph = builder.compile() print(f"ReActAgent initialized with {provider} - {model}.") def __call__(self, question: str, input_file: str = "") -> str: input_msg = [HumanMessage(content=question)] out = self.react_graph.invoke({"messages": input_msg, "input_file": input_file}) for o in out["messages"]: o.pretty_print() # The last message contains the agent's reply reply = out["messages"][-1].content # Optionally, strip out “Final Answer:” headers if "FINAL ANSWER: " in reply: reply = reply.split("FINAL ANSWER: ")[-1].strip() return reply