AutoFixAgentCore

The AutoFixAgent is a RAG and multi-agents based APR tool using MetaGPT and LightRAG.

Prerequisites

This proj is a part of the whole project. You must do those before running this project.

Important

This project is running under app directory.

Install

Config the LLM backend by ollama

Note

Still on work

Copy the file app/core/config/config2.example.yaml to app/core/config/config2.yaml, then modify it to include your own LLM source. Refer to MetaGPT's documentation for proper configuration within the MetaGPT framework.

Then initialize the independent configuration service by dynaconf:

cp .secret.example.toml .secrets.toml 

Store sensitive credentials like api_key in .secrets.toml and configuration settings in settings.toml. Maintain separate files for security.

Important

Always provide a value for the api_key field, even if the backend (e.g., Ollama) does not require a valid API key. This is because the instructor library uses this field to populate the bearer token. Omitting it or leaving it empty may result in a 400 Bad Request error when requesting a structured response.

Tip

For detailed examples and instructions on implementing a custom client (other than OllamaClient) and using it, refer to app/llm_test.py and app/core/utils/provider/ollama.py.

Install the dependencies

Before installing, check:

apt-get update && apt-get install build-essential python3-dev

If on x86 platform, you can install the dependencies by:

conda env create -f environment_x86.yaml

For macOS, you can install the dependencies by:

conda env create -f environment_osx-arm64.yaml

Usage

Run the example to test the reachability of metagpt framework

cd app
python3 example.py

Run the example to test the custom LLM client

cd app
python3 llm_test.py

Run core client by cli

cd app
python3  cli_core.py --help

NOTE

Ref https://www.cnblogs.com/wuhuacong/p/18380808 to organize