RepoMaster

RepoMaster is an open-source AI agent built by QuantaAlpha that transforms how developers solve complex coding tasks by automatically finding and leveraging GitHub repositories. Released in August 2025 and accepted as a NeurIPS 2025 Spotlight paper, it enables users to describe a task in natural language and watch as the agent discovers, configures, and orchestrates the right open-source tools to complete it autonomously.

What It Is

RepoMaster is a multi-agent system designed to make the 100M+ repositories on GitHub work as an intelligent, on-demand toolbox. Rather than writing code from scratch, users describe what they want and the agent handles repository discovery, environment setup, and execution end-to-end. The core pipeline follows a Discover → Understand → Execute loop: the agent analyzes the request, searches for relevant GitHub tools, auto-configures them, and delivers results.

How the Agent Works

RepoMaster uses hierarchical code trees, module-dependency views, and function-call graphs to understand large repositories with pruned context — reducing token usage while maintaining task accuracy. The system supports multiple specialized agent modes:

• Deep Search Agent: performs multi-step web and repository search to find the best tools for a task
• Repository Agent: focuses on understanding and executing within a specific codebase
• General Programming Assistant: handles broader coding questions and tasks
• Unified Mode: orchestrates all agents together for complex multi-step workflows

The web interface (built on Streamlit, accessible at localhost:8501) and CLI backend both expose these modes, making it accessible to beginners and power users alike.

Setup Path

Installation requires cloning the repository, installing Python dependencies via pip install -r requirements.txt, and configuring API keys in a .env file. Required keys include an OpenAI API key (with GPT-5 as the default model), a Serper key for Google search integration, and a Jina key for web content extraction. Optional support is available for Anthropic Claude, DeepSeek, and Google Gemini as alternative AI providers.

Update: NeurIPS 2025 Spotlight and Open-Source Release

The project was open-sourced on August 28, 2025. The accompanying paper, "RepoMaster: Autonomous Exploration and Understanding of GitHub Repositories for Complex Task Solving" (arXiv:2505.21577), was accepted to NeurIPS 2025 as a Spotlight paper — a distinction the team notes represents approximately 3.2% of submissions. A companion benchmark, GitTaskBench, was also open-sourced to evaluate repo-level code agents on real-world task delivery. The repository had accumulated 523 stars and 67 forks as of the last recorded update.

Ecosystem and Research Context

RepoMaster is part of a broader QuantaAlpha research ecosystem that includes SE-Agent (a self-evolution trajectory framework, NeurIPS 2025 Poster) and GitTaskBench (a repo-level benchmark). The project acknowledges foundational work from AutoGen, OpenHands, and SWE-Agent. QuantaAlpha was founded in April 2025 by researchers affiliated with Tsinghua University, Peking University, CAS, CMU, HKUST, and other institutions, with a stated focus on CodeAgent, DeepResearch, Agentic RL, and self-evolving multi-agent systems.