GitHub Releases

Desktop builds

The installer is a thin Electron shell. llama.cpp and MLX are not bundled. On first use the app downloads a runtime, checks its SHA-256, and unpacks it into the data directory.

GitHub Releases

Windows

Revolver-0.2.0-win-x64.exe — NSIS, per-user, no administrator rights.

  1. Run Setup. The build is unsigned; SmartScreen may ask for More info → Run anyway.
  2. Config, models, and runtimes live in %APPDATA%\Revolver.
  3. An NVIDIA driver is required for CUDA. A vendor Vulkan ICD is required for Vulkan. Neither is included in the installer.
Download

Linux

Revolver-0.2.0-linux-x64.AppImage. A .deb is also produced by pack:native.

  1. Make the AppImage executable (chmod +x), then run it.
  2. Config: ~/.config/Revolver. Models default to ~/.revolver/models.
  3. NVIDIA driver for CUDA. Mesa, membership in video/render, and /dev/dri for Vulkan.
Download

macOS

Revolver-0.2.0-arm64.dmg — Apple Silicon only.

  1. Drag Revolver to Applications. The build is ad-hoc signed; use Privacy & Security → Open Anyway on first launch.
  2. Setup waits until both llama.cpp (Metal) and MLX are installed.
  3. Data lives in ~/Library/Application Support/Revolver.
Download

Build from source

Clone the repository, install dependencies, then pack or run for your platform. Node.js 22+ and npm are required.

Clone

git clone https://github.com/TownsendBrown/revolverllm.git
cd revolverllm
npm install

Linux

Native default. Output: release-native/ (AppImage and .deb).

npm run pack:native
# or run without packaging:
npm run start:native
npm run dev:native

Windows

Native default. Output: release-win/ (NSIS Setup.exe).

npm run pack:windows
# or run without packaging:
npm run start:windows
npm run dev:windows

macOS

Apple Silicon. Pack writes a DMG to release-mac/. Run uses the Metal host agent.

./scripts/pack-electron-macos.sh
# or run without packaging:
npm run start:macos

Docker

Browser UI on port 8080. Set MODELS_DIR to an absolute host path. Native spawn is disabled in Compose.

cp .env.example .env
npm run docker:up          # CPU
npm run docker:up:gpu      # NVIDIA
# macOS Compose + Metal:
# brew install llama.cpp
npm run docker:up:mac

First launch

Packaged builds include the UI and the runtime catalog. If no SKU is installed, the Server tab offers Install recommended: NVIDIA → CUDA, AMD or Intel → Vulkan, otherwise CPU.

Config: manage runtimes — CUDA, Vulkan, CPU SKUs
Runtimes — install CUDA, Vulkan, or CPU from Config.
Models tab after a local GGUF is on disk
Models — place GGUF files in the models directory, or download from the hub.