Download Ollaya

curl -fsSL https://ollaya.dev/install.sh | sh

The script detects your CPU and NVIDIA GPU, downloads the release from GitHub, checks its sha256 and, where systemd runs, sets up the ollaya service. With a GPU it also fetches the CUDA libraries (1 to 1.6 GB). It never installs drivers.

ollaya run winnow:e4b

The recommended model: close to Jev's accuracy, in about 90 ms on an NVIDIA GPU with 10 GB or more (an 8 GB download). Otherwise, ollaya run laya answers in a fraction of a second, on the CPU too.

• x86-64 or ARM64 with glibc 2.38 or newer: Ubuntu 24.04, Debian 13, Fedora 39, RHEL 10 or newer.
• Runs on the CPU. An NVIDIA GPU is optional: driver R525 or newer, on x86-64 (CUDA 13 from R580, CUDA 12 before and for GTX 10-series and Volta cards).

Start and stop the server, download models and try them, in one window. Your code talks to the same local API; for the ollaya command, install the command line too.

On its own the app runs models on the CPU. With an NVIDIA GPU, install the command line as well: the app then starts the server from it, on the GPU, and its status line says which one is running.

Start and stop the server, download models and try them, in one window. Your code talks to the same local API; for the ollaya command, install the command line too.

On its own the app runs models on the CPU. With an NVIDIA GPU, install the command line as well: the app then starts the server from it, on the GPU, and its status line says which one is running.

The Windows app is not code-signed yet. If SmartScreen stops the installer, choose More info, then Run anyway.

irm https://ollaya.dev/install.ps1 | iex

In PowerShell. It installs ollaya for your user, with no administrator rights, and puts it on your PATH. The archives are checked against the release's sha256. With an NVIDIA GPU it also fetches the CUDA libraries (about 1 GB). It never installs drivers.

ollaya run winnow:e4b

The recommended model: close to Jev's accuracy, in about 90 ms on an NVIDIA GPU with 10 GB or more (an 8 GB download). Otherwise, ollaya run laya answers in a fraction of a second, on the CPU too.

• Windows 10 or 11 on a 64-bit x86 PC. Runs on the CPU.
• An NVIDIA GPU is optional: driver R527 or newer (CUDA 13 from R580, CUDA 12 before and for GTX 10-series and Volta cards). The command line uses it; the desktop app uses it when the command line is installed too, and the CPU otherwise.
• WSL 2 with the Linux installer works too. The server in WSL answers Windows programs at localhost:11435.

Start and stop the server, download models and try them, in one window. Your code talks to the same local API; for the ollaya command, install the command line too.

curl -fsSL https://ollaya.dev/install.sh | sh

The script downloads the release from GitHub and checks its sha256. Start the server with ollaya serve, or let ollaya run start it for you.

• A Mac with Apple silicon (arm64) and macOS 14 or newer.

docker run -d --name ollaya --gpus=all -p 11435:11435 -v ollaya:/home/ollaya/.ollaya ghcr.io/ollaya-dev/ollaya:cuda

Needs the NVIDIA Container Toolkit and a host driver with CUDA 13 support (R580 or newer). For drivers with CUDA 12 (R525 to R575), or a GTX 10-series or Volta GPU, use :cuda12.