Installation

Installation Overview

Install the standalone refact binary, then optionally add IDE clients.

Refact is a single native binary that runs a resident daemon, an interactive TUI, an in-browser GUI, a per-project worker supervisor, and CLI controls. Install it directly — IDE extensions are optional clients of the same daemon, not the only way to run Refact.

Standalone binary (recommended)

Unix / macOS:

curl -fsSL https://raw.githubusercontent.com/JegernOUTT/refact/main/install.sh | sh

Windows (PowerShell):

irm https://raw.githubusercontent.com/JegernOUTT/refact/main/install.ps1 | iex

Then run refact from a workspace to launch the TUI and start the local daemon. Full details, daemon commands, and self-update: Standalone CLI & Daemon guide.

IDE clients (optional)

The plugins locate a system refact (or ~/.refact/bin/refact); if none is found or it is too old, they download the pinned binary themselves. Either way they attach to the shared daemon.

VS Code

Install from the Visual Studio Marketplace · VS Code guide

JetBrains IDEs

Install from JetBrains Marketplace · JetBrains guide

First-Run Checklist

1. Install the standalone binary (or an IDE extension).
2. Start Refact: run refact, or open the Refact sidebar/tool window in your IDE.
3. Confirm the local daemon starts in the background (refact status).
4. Add a hosted BYOK provider, OAuth-capable provider, local runtime, or custom endpoint in Provider Setup / BYOK.
5. Enable the models you want Refact to use and select defaults for chat, reasoning, agent work, and code completion.
6. Start a chat, switch to an agent mode, accept inline completions, or use refact / refact tui.

What Gets Installed

The standalone install places a single refact binary at ~/.refact/bin. It provides the daemon, TUI, in-browser GUI, worker supervisor, and CLI. An IDE extension reuses that same binary (locating or downloading it) and attaches to the daemon — it does not ship a separate engine. The daemon handles chat sessions, model-provider calls, local context, tool execution, code completion, trajectories, tasks, and knowledge, with one warm worker per project shared across all clients.

What You Still Need

Bring at least one model source:

• A provider key or OAuth flow for hosted providers.
• A running local runtime such as Ollama, LM Studio, or vLLM.
• A custom OpenAI-compatible endpoint with model details.

A separate hosted Refact backend is not part of the setup path.