mariahberdnikova/dz
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Human-in-the-Loop через middleware: README.md

Browse Source
This commit is contained in:
Мария Бердникова
2026-05-27 10:14:50 +00:00
parent 359779e3a8
commit f469c8828f
1 changed files with 97 additions and 120 deletions
Download Patch File Download Diff File Expand all files Collapse all files
solutions/69a86305c46fd26feae6bcaa_Human-in-the-Loop_через_middleware/README.md
+91 -114
View File
@@ -1,52 +1,80 @@
# Human‑in‑the‑Loop Middleware Demo # Human‑in‑the‑Loop Middleware Demo
A minimal LangChain project that demonstrates how to pause an agent before every tool call and let a user approve or reject the action via the terminal.
This repository contains a minimal LangChain/​LangGraph example that demonstrates how to pause an agent whenever it wants to call a tool and ask the user for approval (`approve` / `reject`).
The pausing logic is handled by **`HumanInTheLoopMiddleware`**, which automatically generates a prompt, receives the user’s decision, and resumes execution via a `Command(resume={…})`.
> **Why this matters** – In many real‑world scenarios an LLM should not act autonomously. By inserting a human checkpoint you can keep control over every tool call, ensuring safety, compliance or simply giving the user a chance to correct mistakes.
--- ---
## 📦 Installation ## Table of Contents
- [Project Overview](#project-overview)
- [Prerequisites](#prerequisites)
- [Installation](#installation)
- [Running the Demo](#running-the-demo)
- [1️⃣ `main.py` – Interactive Agent](#1-mainpy---interactive-agent)
- [2️⃣ `tool_example.py` – Custom Tool](#2-tool_exempley---custom-tool)
- [Example Interaction](#example-interaction)
- [Project Structure](#project-structure)
---
## Project Overview
The agent is created with **HumanInTheLoopMiddleware**.
When the agent wants to use a tool (e.g., `get_weather`), it pauses, prints a prompt in the terminal, and waits for user input:
| Decision | Effect |
|----------|--------|
| `approve` | Tool call proceeds |
| `reject` | Tool call is skipped; the agent continues reasoning |
| `edit` | (Optional) User can modify the tool arguments before resuming |
The middleware automatically handles the pause/resume logic via `Command(resume={"decisions": [...]})`.
---
## Prerequisites
- Python 3.10+
- A working OpenAI API key (or any LLM provider supported by LangChain)
Set your key in an environment variable:
```bash ```bash
# 1️⃣ Clone the repo export OPENAI_API_KEY="sk-..."
git clone https://github.com/your‑org/human‑in‑the‑loop-demo.git ```
cd human‑in‑the‑loop-demo
# 2️⃣ Create a virtual environment (recommended) ---
## Installation
1. **Clone the repo**
```bash
git clone https://github.com/yourusername/human-in-loop-demo.git
cd human-in-loop-demo
```
2. **Create a virtual environment (optional but recommended)**
```bash
python -m venv .venv python -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate source .venv/bin/activate # Windows: .venv\Scripts\activate
```
# 3️⃣ Install dependencies 3. **Install dependencies**
```bash
pip install -r requirements.txt pip install -r requirements.txt
``` ```
> **Requirements** > `requirements.txt` contains:
> * Python ≥ 3.10 > ```
> * `langchain` (>=0.2) > langchain==0.2.*
> * `langgraph` (>=0.1) > langgraph==0.1.*
> * `openai` or any LLM provider you prefer > python-dotenv # optional, for .env support
> ```
The `requirements.txt` file contains the exact versions used for this demo.
--- ---
## 📁 Project Structure ## Running the Demo
``` ### 1️⃣ `main.py` – Interactive Agent
.
├── agent.py # Agent definition with HumanInTheLoopMiddleware
├── main.py # CLI entry point – runs the agent interactively
├── tools.py # Example tool(s) (e.g., get_weather)
└── README.md # This file
```
---
## 🚀 Running the Demo
### 1️⃣ Start the Agent
```bash ```bash
python main.py python main.py
@@ -55,106 +83,55 @@ python main.py
You will see a prompt like: You will see a prompt like:
``` ```
Assistant: What would you like to do? Agent: What would you like to know?
User: Tell me the weather in London. User: Tell me the weather in London.
Assistant: (Thinking...) Agent: (pausing) Would you like to call tool 'get_weather' with arguments {'location': 'London'}? [approve/reject/edit]
>
``` ```
When the agent decides to call `get_weather`, it pauses and prints: Type `approve`, `reject`, or `edit` and press **Enter**.
If you choose `edit`, you’ll be prompted to modify the JSON arguments.
``` ### 2️⃣ `tool_example.py` – Custom Tool
Human-in-the-loop checkpoint:
The assistant wants to execute tool 'get_weather' with arguments {'location': 'London'}.
Please type one of the following decisions: approve / reject
```
Type **`approve`** to let the tool run, or **`reject`** to cancel it. The demo includes a simple weather tool (`get_weather`).
After your decision, the agent continues: You can add more tools by editing `tools/__init__.py` or creating new modules.
```
Assistant: The weather in London is 18°C and sunny.
```
### 2️⃣ Using a Different Tool
If you want to test another tool (e.g., `get_time`), add it to `tools.py`, import it in `agent.py`, and adjust the middleware configuration accordingly.
--- ---
## 📚 Example Usage ## Example Interaction
```python ```
# main.py $ python main.py
from agent import agent Agent: Hi! How can I help you today?
User: What's the weather in Paris?
def run(): Agent: (pausing) Would you like to call tool 'get_weather' with arguments {'location': 'Paris'}? [approve/reject/edit]
print("Welcome! Type 'exit' to quit.") > approve
while True: Tool get_weather called. Result: "Sunny, 22°C"
user_input = input("\nUser: ")
if user_input.lower() == "exit":
break
response = agent.invoke({"input": user_input})
print(f"Assistant: {response['output']}")
if __name__ == "__main__": Agent: The current weather in Paris is Sunny, 22°C.
run() User: Thanks!
``` ```
```python If you type `reject`, the agent will continue without calling the tool:
# agent.py
from langchain.agents import create_agent
from langchain.agents.middleware import HumanInTheLoopMiddleware
from langgraph.checkpoint.memory import MemorySaver
from tools import get_weather # <-- your tool(s)
memory = MemorySaver()
agent = create_agent(
model="gpt-4o-mini", # or any LLM you have access to
tools=[get_weather],
system_prompt="You are a helpful assistant.",
middleware=[
HumanInTheLoopMiddleware(
interrupt_on={"get_weather": True},
description_prefix="Please confirm the tool call:",
),
],
checkpointer=memory,
)
``` ```
Agent: (pausing) Would you like to call tool 'get_weather' with arguments {'location': 'Paris'}? [approve/reject/edit]
```python > reject
# tools.py Agent: I couldn't retrieve the weather. Could you provide more details?
from langchain.tools import BaseTool
class GetWeather(BaseTool):
name = "get_weather"
description = "Returns current weather for a location."
args_schema = ... # define your schema here
def _run(self, location: str) -> str:
# Dummy implementation – replace with real API call
return f"The weather in {location} is 18°C and sunny."
get_weather = GetWeather()
``` ```
--- ---
## 🔧 Customization Tips ## Project Structure
| Feature | How to change | ```
|---------|---------------| human-in-loop-demo/
| **Allowed decisions** | `interrupt_on={"get_weather": {"allowed_decisions": ["approve", "reject"]}}` | ├── main.py # Entry point – runs the interactive agent
| **Prompt prefix** | Change `description_prefix` in the middleware. | ├── tool_example.py # Example custom tool (get_weather)
| **Tool list** | Add or remove tools in the `tools=[...]` array. | ├── requirements.txt # Dependencies
| **LLM model** | Pass a different model name or a custom LLM instance to `create_agent`. | └── README.md # This file
```
--- Feel free to extend the project by adding new tools, customizing prompts, or integrating a different LLM provider. Happy hacking!
## 📜 License
This project is licensed under the MIT License – see the [LICENSE](LICENSE) file for details.
---