RepoPilot

Open-source AI-powered tool for smart repository maintainance

Features

1. Q&A service for documentation and code understanding with advanced RAG techniques
2. PR analysis is provided by pr-agent
3. Documentation generatinon using RepoAgent

Get started

0. I strongly recommend to use deepseek model. Package was tested only with this model.
1. Create new github account (or use exist)
2. Generate new personal token
3. Run qdrant: docker run -p 6333:6333 qdrant/qdrant:v1.12.1

Option 1 (Preferable. Using pre-build docker image)

4. Create folder with following structure:

   my_project
   --.env  # file with environment envs
   --config.yml  # config file for Q&A RAG
   --pr_agent.toml  # config for pr-agent
   --start.py  # startup script
   

   You can find config and .env file examples in examples folder.

   Note

   .env file structure as follows:

   HUGGINGFACE_ACCESS_TOKEN=<<hf token>>
   OPENAI_BASE_URL=https://api.deepseek.com
   OPENAI_API_KEY=<<deepseek key>>
   DEEPSEEK_API_KEY=<<deepseek key>>
   GITHUB_TOKEN=<<github token>>
   

5. If GPU is available:

   Run docker run --net=host --gpus all -it -v /path/to/folder/with/startup/script:/usr/local/app/example valer1435/repo_pilot bash

   Then in the container run cd /usr/local/app/example && python start.py

6. If GPU is NOT available:

   Run docker run --net=host -it -v /path/to/folder/with/startup/script:/usr/local/app/example valer1435/repo_pilot bash

   Then in the container run cd /usr/local/app/example && python start.py

Option 2 (Build from source)

Run

git clone https://github.com/valer1435/RepoPilot.git
cd RepoPilot
docker build -f docker/Dockerfile -t valer1435/repo_pilot:0.0.1 .

Option 3 (no docker)

4. Install locally

   git clone https://github.com/valer1435/RepoPilot.git
   cd RepoPilot
   pip3 install -r requirements.txt
   pip3 install -e .

5. Create folder with following structure:

   my_project
   --.env  # file with environment envs
   --config.yml  # config file for Q&A RAG
   --pr_agent.toml  # config for pr-agent
   --start.py  # startup script
   

   You can find config and .env file examples in examples folder.

   Note

   .env file structure as follows:

   HUGGINGFACE_ACCESS_TOKEN=<<hf token>>
   OPENAI_BASE_URL=https://api.deepseek.com
   OPENAI_API_KEY=<<deepseek key>>
   DEEPSEEK_API_KEY=<<deepseek key>>
   GITHUB_TOKEN=<<github token>>
   

6. Run python start.py

Usage

Please follow format are presented in the examples folder.

Pr reviewer

PR review capabilities have not been changed comparing with original framework. For more information please refer to pr-agent documentation. To disable pr reviewer just set pr_config_path=None in RepoPilot in startup script.

Issue automatic Q&A

You have to create .yml config file. Most things are common, but let's consider specific:

You have to setup 2 data sources (here example for vllm framework).

1.   docs:
    collection_name: "vllm" # should be the same with qdrant collection
    site: "https://docs.vllm.ai/en/latest/" # site with docs
    extensions: ["/", '.html'] # extension of files to use

   Doc parser will recursively parse all pages of documentation, starting from single one. To do this algorithm parses all links to documentation pages on a single page. Then collects links from the second page and so on.

2. code:
    collection_name: "vllm_code" # should be the same with qdrant collection
    repo_owner: "vllm-project"  # repo owner (from github)
    repo_name: "vllm" # repo name (from github)
    branch: "main" # repo branch (from github)
    extensions: [".py"] # files to include
    folders: ["vllm"] # folders to include

Documentation

You can follow AI-generated documentation genereted by RepoAgent. Please note that llm can hallucinate!