KRASISGitHub

Krasis 1.0.21 · Stable release

Your models.
Your machine.

Choose your operating system. The launcher takes you from model download to a running server.

You’ll need an NVIDIA GPU (Ampere or newer), a current NVIDIA driver, and enough RAM and disk space for your model. Model requirements vary.

Windows · 64-bit

Download. Install. Launch.

Native Windows installer with its own Python runtime.

Download for Windows

KrasisSetup-1.0.21-win64.exe

  1. Run the installer and let setup finish. The first install downloads CUDA dependencies, so allow a few minutes.
  2. Open Krasis from the Start menu.
  3. Download or select a model, choose your GPU and settings, then select Launch.

Open Krasis Manager from the Start menu for the browser dashboard.

Linux · x86-64 · Also WSL2

Three commands. You’re in.

Use Python 3.10–3.13. Krasis manages its own environment.

Download Linux installer

Or paste these commands into your terminal:

Terminal
curl -fsSL https://krasis.ai/downloads/install.sh | bash
krasis-setup
krasis

If krasis-setup isn’t found, reopen your terminal. In the launcher, download or select a model, choose your GPU and settings, then select Launch.

On Windows with WSL2, run these commands inside your Linux distribution.

Make your first connection.

After launch, point your OpenAI-compatible chat or coding app at http://127.0.0.1:8012/v1. Use the address shown by the launcher if you change the host or port.

Keep it current.

On Linux, run krasis update. On Windows, download and run the current installer above. Your models and saved profiles stay in your user’s .krasis folder.

NVIDIA GPUs only; macOS is not supported. Dense Qwen3.5 text support is experimental; vision remains unqualified. GLM’s published qualification applies to the tested scales-only profile, not every launcher configuration. Windows installer checks do not establish GPU inference quality on every card.