Skip to main content
A native worker is a small program you run on a computer to dedicate it to bunnyCompute full-time. It serves jobs from the moment you start it until you stop it, so it earns Carrots more steadily than the browser worker. Use it when you have a desktop or a dedicated GPU box you want to leave running.
This is the most technical page in the docs: it involves typing one command into a terminal. Everything else happens in the app. If you would rather not touch a terminal, supply compute from your browser instead.

​
Before you start

Your machine needs a few things in place:
  • A computer with a real GPU or Apple Silicon.
  • Node.js 18.17 or newer (version 22 or newer is recommended).
  • About 6 GB of free disk space for the model.
  • Ollama installed, from ollama.com/download. The worker starts it and downloads the model for you.

​
Get a worker key and run it

1

Register the worker

In the app, open Settings › Workers and select Register a native worker. Copy the key it shows you.
2

Run the command

Open a terminal on the machine you want to dedicate, then run the command below, pasting your key in place of <your key>.
3

Let it warm up

On the first run it installs the model, then warms up and prints its measured speed. Once it connects, it starts serving jobs. Press Ctrl-C to stop it at any time.
Add --save-creds to the command if you want the worker to remember the key so you do not have to paste it next time.
The worker key is shown only once. Copy it before you leave the screen. It cannot be retrieved later. If you lose it, register a new worker.

​
Manage your workers

Everything you run shows up under Settings › Workers, including browser workers, which appear automatically when you supply compute from a device. Each worker shows its current status: From this screen you can:
  • Rename a worker to label the machine it runs on, so you can tell your devices apart.
  • Revoke a worker to shut it out immediately. Revoking permanently disables that key, so the worker cannot reconnect. You would register a fresh one to bring the machine back.
If a worker is ever turned away as “too slow”, it is falling below the network’s speed floor. Close other graphics-heavy apps on that machine, or move the worker to a faster GPU. See how contributing works for the details.

How contributing works

Worker types, per-job earnings, the speed floor, and privacy.

Setting up a browser worker

The one-button browser worker, with no terminal needed.

Your Carrots

Check your balance, history, and top-up options.