An LLM-powered repository agent for wiki-style documentation and deep codebase understanding.

English · 简体中文

repo-agent is designed to help developers and teams build, maintain, and query repository documentation automatically. It is especially useful when you want to:

• Quickly understand a newly inherited project
• Build or improve high‑quality Wiki / developer documentation
• Continuously track repository changes and keep docs up to date

• Automatic Wiki generation
  • Generate comprehensive documentation for any GitHub / Gitee repository: project overview, commit history, PR history, release notes, code file documentation, and more.
• Repository QA
  • Ask natural language questions about the repository, powered by the generated Wiki and code understanding.
• Automatic documentation updates
  • Monitor repository changes (commits, merged PRs, new releases) and incrementally update the related documents.
• Multi‑agent orchestration
  • Use LangGraph / LangChain to orchestrate multiple agents that handle code parsing, documentation generation, vector store writing, QA, etc.

Modern repositories evolve quickly: frequent commits, complex dependency structures, multiple contributors, and sparse documentation. Keeping a trustworthy, human-readable wiki in sync with the codebase is hard and time‑consuming.

repo-agent combines:

• LLMs for high‑level summarization and natural language interfaces
• Code parsers for precise structural understanding
• Graph-based agent orchestration for robust, extensible workflows

With these, it can:

• Turn raw repositories into navigable documentation spaces
• Serve as a knowledge layer for downstream tasks (RAG QA, dashboards, review tools, etc.)

git clone https://github.com/sse-orz/repo-agent
cd repo-agent

# Install backend dependencies
uv sync

# Configure environment
cp .env.example .env
# then edit .env with your keys and settings

# Start core agents
make repo    # repo documentation & wiki pipeline
make rag     # retrieval-augmented QA pipeline
make server  # backend API service

For a web UI:

cd web
pnpm install
pnpm dev

Then open the printed URL in your browser to interact with repo-agent.

• OS: Linux / macOS (WSL2 is supported)
• Python: >= 3.10 recommended
• Dependency management: uv (recommended), or your own Python environment manager
• Node.js: >= 18 recommended for the frontend, using pnpm as the package manager

Using uv:

uv sync

If uv is not installed, refer to its official documentation or install the dependencies defined in pyproject.toml manually.

cp .env.example .env
# Edit .env to configure API keys, model settings, GitHub / Gitee tokens, etc.

Common configuration items (refer to the actual .env.example):

• LLMs: API keys for DeepSeek / Qwen / OpenAI / other models
• Code hosting: GitHub / Gitee access tokens
• Vector databases: Chroma / FAISS storage paths or connection settings

make repo

This triggers workflows that:

• Analyze the structure of a target repository
• Generate / update Wiki-style markdown documentation

make rag

This starts the retrieval pipeline to:

• Ingest docs and code snippets into the vector store
• Answer questions such as “Where is this function implemented?” or “What are the dependencies of this module?”

make server

This typically starts an HTTP API service (e.g. based on FastAPI) that powers the web UI and external integrations.

cd web
pnpm install
pnpm dev

Open the URL printed in the terminal to access the repo-agent web interface.

• Built with Python and LangGraph / LangChain to orchestrate multiple agents
• Integrates multiple LLM providers (DeepSeek, Qwen, GPT‑4, etc.)
• Provides unified abstractions to make it easy to plug in new models and data sources

• Built with Vue.js + Vite
• Provides UI for repository selection, task progress tracking, documentation preview, and interactive QA

• LangGraph / LangChain: Agent workflow orchestration
• Tree-sitter: Multi-language code parsing (functions, classes, AST, etc.)
• LLMs: DeepSeek, Qwen, GPT‑4, and others
• Vector databases: Chroma, FAISS for efficient semantic search
• APIs: GitHub / Gitee integrations for repository data
• Frontend: Vue.js + Vite
• Backend: Python (FastAPI or similar framework)

Contributions are very welcome! You can help in several ways:

• Issues: Report bugs, request features, or suggest improvements
• Pull Requests: Submit code changes (ideally with a short design explanation)
• Documentation: Improve usage docs, examples, and architecture descriptions

Before opening a PR, it is recommended to:

• Ensure basic tests / lint checks (if available) pass
• Keep the coding style consistent with the existing codebase

This project is released under the MIT License. See the LICENSE file for details.