Refact.ai
VerifiedOpen-source AI coding agent, local-first with your own model keys
Product preview
Refact.ai
What it is
Refact.ai combines openness with autonomy: its agent plans and executes multi-step coding tasks with tool use (shell, debugger, browser), and completions and chat work in VS Code and JetBrains. Refact Cloud, the hosted service, is being retired, so the project continues as open source, local-first, and bring-your-own-key, working with providers such as OpenAI, Gemini, Grok, and DeepSeek or local runtimes like Ollama and LM Studio, including fully air-gapped setups. It topped SWE-bench among open agents, showing open tools can compete.
Key features
- Autonomous agent with tools
- Local-first and self-hosted
- Bring-your-own-key models
- IDE completions and chat
What we liked
- + Full data control
- + Strong open benchmarks
- + Flexible models
Worth noting
- · You supply models and API keys
- · Hosted cloud being retired
- · Smaller community
Editor's notes
More context from AI BirdView
Refact.ai stands out as one of the few coding agents that keeps its whole stack open and lets you decide where it runs. It plans and carries out multi-step coding tasks inside your editor, searches your repository for context, and can reach tools such as GitHub, databases, Docker and CI pipelines. In 2026 the company announced it is retiring its hosted Refact Cloud and continuing the project as open source, local-first and bring-your-own-key, so self-hosting is now the main way to use it.
- Plugins cover VS Code and the JetBrains family (IntelliJ, PyCharm, WebStorm, GoLand and CLion), with Visual Studio, Neovim and Sublime Text support in beta.
- Bring-your-own-key support works with providers such as OpenAI, Gemini, Grok and DeepSeek, so teams can pick the models they trust.
- Code completion and chat both use retrieval over your project, which keeps suggestions grounded in your actual codebase.
- The company reported a 2025 SWE-bench Verified result that it described as the top score among open-source agents, and it published its benchmark pipeline on GitHub.
Refact is a strong fit for engineering teams with strict data rules, regulated industries and developers who prefer owning their tooling. Support and community discussion are moving to GitHub. Tip: if you used the hosted service, export any cloud-only data and move your provider keys into a local configuration, then pair the agent with a capable model of your choice.
Reader reviews
From people who've used Refact.ai. Emails are never shown publicly.
4.4 · 12 reviews
Alternatives to Refact.ai
Other tools we like in this space.