Skip to content

Installation

warpdrv works with any standard llama-server binary, so hardware compatibility tracks llama.cpp’s own support matrix:

Backend Hardware
CUDA NVIDIA GPUs
ROCm AMD GPUs
Vulkan Any GPU (Intel, AMD, NVIDIA)
CPU Any machine (no GPU required)

Note: warpdrv does not bundle a llama.cpp binary. You’ll download or compile one separately — covered in the Downloading & running your first AI model guide.

Head to the releases page and grab the latest build for your platform.

.deb (Debian, Ubuntu, Mint):

Terminal window
sudo dpkg -i warpdrv_*.deb

RPM (Fedora, openSUSE, RHEL):

Terminal window
sudo rpm -i warpdrv_*.rpm

.AppImage (any distro):

Terminal window
chmod +x warpdrv-*.AppImage
./warpdrv-*.AppImage

Download the .msi installer and run it through the standard Windows installer wizard.

Note: The Recipes feature (automated llama.cpp builds) requires Bash. On Windows, install Git for Windows or use WSL to get a Bash environment.

No prebuilt binary is available yet. You’ll need to build from source — see the Compiling guide. Untested on Apple Silicon; contributions welcome.

On first launch, warpdrv shows an onboarding welcome screen that walks you through:

  1. Adding a folder where your GGUF models live (or where to download them).
  2. Optionally installing a llama.cpp backend.
  3. A short slideshow of next steps.

See the Onboarding guide for a detailed walkthrough.

  • Config & data: ~/.config/warpcore/ (Linux) — chat database, settings, MCP config, recipes.
  • Models: Stay in whatever folder you point warpdrv at. warpdrv only indexes them; it never moves or modifies your model files.

Your data is preserved across updates and reinstalls.

warpdrv checks for updates on startup. If a new version is available, a banner appears at the top of the app. Click it to open the releases page, download the new version, and install. Your data is preserved.