The first public release is being prepared
There is nothing to download yet, and we would rather say so than hand out a link that is not ready. The engine and STRADA are built and measured on Windows. This page says what the release will contain and how installing it works, and it becomes the download page the day the release is out.
The VEYMEER engine
One self-contained program, veymeer, that installs itself: the server, the command line, and the kernels for NVIDIA GPUs from the GTX 10-series to the RTX 50-series.
Version 0.1.0. Free on the Individual tier.
STRADA
The open-source editor and coding agent. Its installer installs the engine first if the machine does not have it, then STRADA.
Free for everyone. Source published with the release.
Other systems
The engine is written to run on Linux and on Apple silicon. Those builds have not yet been verified on real machines, so they are not promised for the first release.
Follows the Windows release.
How it will go
Two ways in: STRADA's installer, which brings the engine with it, or the engine on its own.
Run one installer
It installs the engine if needed, then STRADA, for your user account. On first launch STRADA helps you pick a model that fits your GPU.
What the installer does →Unpack, then one command
# in the unpacked folder: install for this user, add to the PATH
veymeer setup
# check the machine, get a model, serve it
veymeer doctor
veymeer pull bartowski/Qwen2.5-7B-Instruct-GGUF:Q4_K_M
veymeer serve --model <the path pull printed>What the machine needs
A current NVIDIA driver. No CUDA toolkit, no Python, nothing else.
Direct3D 12 or Vulkan, which a current graphics driver provides.
A 64-bit processor and enough memory for the model. It runs; a GPU is much faster.
Any supported GGUF file. Models Ollama already downloaded are found and used where they are.
Until then
The docs describe the engine as it will ship: the command line, the API and the settings.