Refact.ai positions itself as a fully autonomous coding agent rather than just an assistant that waits for instructions one step at a time: it can plan, execute, and iterate on engineering tasks using retrieval-augmented generation to stay grounded in a specific project's actual code, rather than relying purely on general training. It works inside VS Code, JetBrains, Neovim, and others, with models including Claude 4, GPT-4o, and Code Llama.
Being open source under the BSD-3-Clause license and available self-hosted with bring-your-own-key support means teams with strict data requirements can run it entirely on their own infrastructure. It has reported strong SWE-bench results among AI coding agents, and includes integrations with GitHub, GitLab, Docker, databases, and CI/CD, along with persistent memory across sessions.
Pricing spans Free, Pro, and Enterprise tiers, which, combined with the open-source option, gives it a wider range of adoption paths than most closed, subscription-only coding agents.