Refact.ai

Refact.ai is an open-source autonomous AI coding agent that plans, writes, executes, and deploys code end-to-end, combining in-editor chat, accurate inline completion, and a full agent workflow into one developer tool. It claims the top spot among open-source agents on SWE-bench and pairs agentic autonomy with the ability to self-host the whole stack on customer infrastructure.

The platform is built on Qwen2.5-Coder with retrieval-augmented generation over the user's own repository, so completions reflect project-specific types, imports, and naming. Integrations with GitHub, databases, and CI/CD tools let the agent run real workflows — open PRs, run tests, query production databases — with supervised approval inside the IDE.

Key benefits:

• Open-source agent under a permissive licence allows full audit and forking
• Self-hosted deployment keeps source code, prompts, and completions on-premises
• Retrieval-augmented completion adapts to project-specific code style and APIs
• Broad IDE support covers VS Code, all JetBrains IDEs, Visual Studio, Neovim, and Sublime
• Autonomous agent mode executes multi-step plans across repos, CI, and databases
• 25+ languages from Python and TypeScript to Rust, C++, SQL, and YAML

Operated by Small Magellanic Cloud AI Ltd., headquartered in the United Kingdom, Refact.ai offers EU and UK hosting under GDPR, plus full on-premises deployment for teams with strict data sovereignty requirements. Self-hosting removes any external data sharing from the loop entirely.

Ideal for engineering teams, regulated enterprises, and open-source contributors that want a Copilot-class coding agent they can inspect, self-host, and run against private codebases without sending anything to US cloud providers.