import os from dotenv import load_dotenv from langchain_google_genai import ChatGoogleGenerativeAI from langchain_core.tools import tool from langchain_core.messages import HumanMessage from langgraph.prebuilt import create_react_agent # Keep this import load_dotenv() @tool def get_fridge_inventory() -> str: """Returns a list of ingredients currently in the user's fridge.""" return "3 Eggs, 1 Spinach bunch, Feta cheese, 2 Tomatoes." llm = ChatGoogleGenerativeAI(model="gemini-2.5-flash") tools = [get_fridge_inventory] # CHANGE: 'state_modifier' is now 'prompt' in the latest prebuilt agent chef_agent = create_react_agent( llm, tools, prompt="You are a Michelin-star chef. Always check the fridge inventory before suggesting a meal." ) if __name__ == "__main__": print(" Running Agent...") # Standard input for ReAct agents inputs = {"messages": [HumanMessage(content="What high-protein snack can I make?")]} # Streaming the response to see the 'thought' process for event in chef_agent.stream(inputs, stream_mode="values"): message = event["messages"][-1] if hasattr(message, "content") and message.content: print(f"\n[CHEF]: {message.content}")