# Human‑in‑the‑loop (interrupt / resume) – LangGraph example

## What this project does

This repository contains a minimal, self‑contained example that demonstrates how to use **LangGraph**’s custom interrupt mechanism for *Human‑in‑the‑loop* (HITL).  The graph pauses at a node, asks the user for confirmation via an interactive console prompt, and then resumes execution with the chosen answer.

The key concepts shown are:

1. **State definition** – a `TypedDict` that holds data produced by the user.
2. **Interrupt node** – uses `interrupt(payload)` to pause the graph.
3. **Resume handling** – the main loop detects `__interrupt__`, shows a question, collects an answer and sends it back with `Command(resume=payload)`.
4. **Checkpointing** – `InMemorySaver` keeps the graph state so that resumption works correctly.

## File structure

| File | Purpose |
|------|---------|
| `requirements.txt` | Python dependencies (langgraph, questionary, python‑dotenv) |
| `main.py` | Entry point – builds and runs the graph, handles interrupts |
| `README.md` | Project description and usage instructions |

## Installation

```bash
# Create a virtual environment (optional but recommended)
python -m venv .venv
source .venv/bin/activate   # On Windows use `.venv\Scripts\activate`

# Install dependencies
pip install -r requirements.txt
```

## Running the example

```bash
python main.py
```

You will see a prompt asking whether you want to continue.  Choose an option and the graph will resume, printing the final state.

## How it works (high‑level)

1. **Graph construction** – two nodes: `ask_user` (interrupt) and `process_answer` (continues after resume).
2. **Interrupt payload** – a dictionary with `type`, `question`, and `options`.
3. **Main loop** – streams the graph, intercepts `__interrupt__`, uses `questionary.select` to get user input, then resumes the graph with the answer attached.
4. **Final state** – after completion, the final state is printed showing the chosen value.

## Extending the example

- Replace the console prompt with a web UI or chatbot interface.
- Add more nodes before/after the interrupt to build richer workflows.
- Persist checkpoints to disk using `FileSaver` if you need long‑term state.

---

**Author:** Kirill Kutlakhmetov