Skip to content
← Tools / Skills
ToolSmall AI Labs/10 trustVerified

Refact.ai

Open-source, self-hostable AI coding assistant with IDE agent for VS Code and JetBrains.

Visit official site →

About

Refact.ai is an open-source AI coding assistant by Small AI Labs (SmallCloud) built around a local-first Rust engine (refact-lsp) with VS Code and JetBrains plugins. It combines real-time code autocompletion, in-IDE chat with Quick, Explore, and Agent modes, and an autonomous agent that plans and executes multi-file tasks with project-aware retrieval-augmented generation, shell execution, and MCP tool integrations.

Its defining trait is data sovereignty: the engine runs locally or on your own infrastructure, models can be self-hosted open-weight models (or your own API keys via BYOK), and team-specific fine-tuning on your codebase is a first-class feature. The project is BSD-3-Clause licensed with around 3.5k GitHub stars, and Refact markets strong SWE-bench Verified results — though those numbers depend on the model and harness together rather than Refact alone.

Limitation: as a smaller open-source project it lacks the polish and ecosystem of Copilot or Cursor, and enterprise features like the self-hosted server need your own GPU infrastructure. The marketed benchmark leadership should be treated as vendor-claimed until independently reproduced.

What it can do

  • Local-first Rust engine with real-time autocompletion
  • Agentic mode for multi-file planning and editing with MCP tools
  • Bring-your-own-model and team codebase fine-tuning
  • Self-hostable server for full data sovereignty

Setup

Install the VS Code or JetBrains plugin, or self-host the engine via Docker with your own models.

Prerequisites

VS Code or a JetBrains IDE; optional GPU infrastructure for self-hosting.

Use cases

code-completionai-coding-agentself-hosted-airefactoring

Works with

Trust score — 8/10

/10 trustVerified

Scored 8/10 (verified tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 59; Trending score 75/100 from researcher assessment.

★ 3,533 GitHub stars

Sources

Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.