Parta Runtime Lite / Guide / Install and run on Linux
Install and run on Linux
Parta Runtime Lite is a single portable executable. There is no installer, no background service, and no account. This page covers everything from download to your first request on Linux.
The Linux build is compiled under Ubuntu 26.04. It is a native Linux binary and runs the same way on bare-metal Linux, but if you hit anything distro-specific, let us know through the bug and feedback form, it helps us test against real environments.
1. Download
Get the current build from the download and verify section of the homepage. The download is a zip file containing the executable, a sample configuration file, and third-party licence notices.
2. Verify the checksum
Before running anything, check the file's SHA256 checksum against the value published next to the download:
sha256sum parta-runtime-lite-linux-cuda.zip
Compare the output to the published hash. If they don't match, do not run the file, download it again.
3. Extract
Unzip the archive anywhere on disk, a local folder, an external drive, or a network share. No root access is required and nothing is written outside the folder you extract to.
unzip parta-runtime-lite-linux-cuda.zip -d parta-runtime-lite
4. Configure
Open lite.toml in the extracted folder and set model.path to a GGUF model file you already have:
[model] path = "/home/you/models/your-model.gguf" gpu_layers = 99 context_size = 8192
Set gpu_layers = 0 to run on CPU only. Leave server.host as 127.0.0.1 to keep the server reachable only from your own machine, or set it to 0.0.0.0 to accept connections from other devices on your network.
5. Run
Mark the binary executable, then run it:
chmod +x parta-runtime-lite ./parta-runtime-lite
On startup it prints the model loaded and the address it's listening on. Leave the terminal open, or run it under tmux/screen/nohup if you want it to keep running after you disconnect. The server runs for as long as the process is running.
6. Send a request
The server speaks OpenAI-compatible and Anthropic-compatible chat completion formats. Anything already built against those APIs works here by pointing it at your local address instead:
curl http://127.0.0.1:8080/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{"model":"local","messages":[{"role":"user","content":"Hello"}]}'
Something not working?
Check the bug and feedback form on the homepage, or use it to tell us your distro, kernel version, GPU driver, and the model you're using, that's the fastest way to help us track down anything Linux-specific.