Human‑in‑the‑Loop Agent

This repository implements a simple command‑line agent that pauses before each tool call and asks the user to approve or reject the action. It uses LangChain and LangGraph's HumanInTheLoopMiddleware.

Prerequisites

Python 3.11+ (tested with 3.11 and 3.12)
An OpenAI API key or a local LLM compatible with the OpenAI API (e.g., Ollama)

Setup

# Create a virtual environment
python -m venv .venv
source .venv/bin/activate

# Install dependencies
pip install -r requirements.txt

Configuration

OpenAI API key

export OPENAI_API_KEY="YOUR_KEY"

Local model (e.g., Ollama)

If you run a local model, edit agent.py and uncomment/adjust the base_url and api_key arguments in the ChatOpenAI constructor:

ChatOpenAI(
    base_url="http://localhost:11434/v1",
    api_key="ollama",
    model="llama3",
)

Usage

python agent.py

You will be prompted to enter a question. The agent will invoke the get_weather tool and pause for confirmation.

Enter your question: What is the weather in Kazan today?

Action 1: get_weather
Arguments: {'city': 'Kazan', 'date': 'today'}
Decision (a=approve, r=reject): a

Agent response:
Weather in Kazan on today

If you reject an action, you will be asked for a reason, which is sent back to the agent as feedback.

License

MIT

1.4 KiB Raw Blame History Unescape Escape