Update README.md

2026-06-03 12:44:40 +00:00
parent 1d27b02481
commit df8fc357f0
1 changed files with 43 additions and 46 deletions
@@ -1,64 +1,61 @@
 # Self‑Correcting LangGraph Agent

-This repository contains a minimal example of a **self‑correcting agent** built with
-[LangGraph](https://langchain-ai.github.io/langgraph/) and
-[LangChain](https://langchain.com/).  The agent:
+## Overview

-1. **Receives a natural‑language task** from the user.
-2. **Executes the task** via an *unreliable* tool that fails 30 % of the time.
-3. **Asks an LLM** (OpenAI GPT‑4o‑mini) to judge whether the result is correct.
-4. **Retries automatically** until the judge says *success* or the maximum number
-   of attempts is reached.
+This project implements a **self‑correcting agent** using LangGraph. The agent
+performs a task, asks an LLM to judge the result, and retries automatically until
+the result is judged **success** or the maximum number of attempts is reached.

-The code demonstrates how to build a small state machine with LangGraph, how to
-use a LLM as a *judge*, and how to implement retry logic.
+The key components are:

-## Setup
+| Component | Purpose |
+|-----------|---------|
+| `AgentState` | Typed state that tracks the task, result, attempts, status, error and max_attempts |
+| `unreliable_tool` | Simulates a tool that fails 30 % of the time (used to demonstrate retry logic) |
+| `verify_result` | LLM judge that must reply with the single word `success` or `failed` |
+| `handle_error` | Resets the error and sets the status back to `pending` for a retry |
+| `execute_task` | Runs the unreliable tool and updates the state |
+| `create_agent` | Builds the LangGraph with the above nodes and a retry loop |
+| `create_agent_executor` | Compiles the graph into a runnable executor |
+| CLI | Run the agent from the command line: `python agent.py "2+2" --max 5` |
+
+## How It Works
+
+1. **Start** – The graph begins at `execute_task`.
+2. **Execute** – The tool runs. If it throws an exception, the state status becomes `failed`.
+3. **Check attempts** – If the number of attempts is >= `max_attempts`, the graph ends with status `max_attempts`.
+4. **Verify** – The LLM judges the result. If the verdict is `success`, the graph ends. If `failed`, it goes to `handle_error`.
+5. **Retry** – `handle_error` clears the error and sets status to `pending`, then the graph loops back to `execute_task`.
+
+## Usage

 ```bash
-# Optional: create a virtual environment
-python -m venv venv
-source venv/bin/activate   # Windows: venv\Scripts\activate
-
 # Install dependencies
 pip install -r requirements.txt

-# Set your OpenAI API key
-export OPENAI_API_KEY=YOUR_KEY   # Windows: set OPENAI_API_KEY=YOUR_KEY
+# Run the agent
+python agent.py "2+2" --max 5
 ```

-## Running the agent
-
-```bash
-python agent.py "Вычисли 2+2"
-```
-
-You can also run the script without arguments – it will prompt you for a task.
-
-## Example output
+The console will show the final state, e.g.:

 ```
-Введите задачу: 2+2
-Попытка 1: результат Result of 2+2
-Попытка 2: результат Result of 2+2
+--- Final State ---
+result: 4
+attempts: 2
+status: success
+error: None
+max_attempts: 5
+``` 

-Итог:
-Успех за 2 попыток. Результат: Result of 2+2
+## Requirements
+
+```text
+langchain>=1.0.0
+langgraph>=1.0.0
+langchain-openai>=1.0.0
 ```

-The exact number of attempts may vary because the tool fails randomly.
+## Note

---
-
-### How it works
-
- **State** – `AgentState` tracks the task, result, number of attempts, status and
-  any error.
- **Nodes** – `execute_task`, `verify_result`, `handle_error`.
- **LLM judge** – a simple prompt that forces the model to answer only
-  "success" or "failed".
- **Graph** – a conditional router that loops back to `execute_task` on failure
-  until the maximum attempts are reached.
-
-Feel free to adapt the tool, the judge prompt, or the retry policy to fit your
-needs.
+The agent uses the OpenAI API. Make sure the environment variable `OPENAI_API_KEY` is set.