All posts
Workers

Running a worker: turn idle GPU into USDC

Your graphics card spends most of the day doing nothing. Here is exactly what happens when you point it at ashao, and what it earns you for the trouble.

Tomas IversenWorker experience
6 min read

There is a real GPU within a meter of you that is, statistically, doing nothing right now. It rendered this page in a few milliseconds and then went back to idle. Running an ashao worker is the act of pointing that idle capacity at the network's job queue and getting paid when it does work. No data center, no install, no command line. This is what the experience actually looks like.

Starting a worker takes one click

On the Earn page there is a single button. When you press it, the browser lazily downloads a small language model — the worker code is never loaded until you opt in, so visiting the page costs you nothing. You will watch a progress bar as the weights stream in and compile against your GPU through WebGPU. The first start takes a moment; after that the model is cached and subsequent starts are near-instant.

Requirements
A browser with WebGPU and a GPU with a little headroom. If WebGPU is unavailable, the worker tells you plainly rather than failing in some cryptic way — there is nothing to debug, just a browser that cannot do the job yet.

The work loop

Once your model is warm, the worker registers itself with the network and settles into a simple loop. It asks the queue for a Pro-tier job; if one is waiting, it claims it and starts generating. As the model produces text, the worker forwards each chunk to the network, which relays it to whoever is waiting on the other end. When the answer is complete, the worker reports the token count and goes back to asking for more.

  • Register — announce your worker and the model it runs.
  • Claim — pull the next queued job, if any.
  • Stream — generate and forward tokens as they arrive.
  • Complete — report the count, bank the earnings, repeat.

The dashboard makes all of this visible: live status, jobs completed, tokens per second, and a running total of what you have earned. It is oddly satisfying to watch your machine quietly answer strangers' questions and tick the counter up.

What it pays

Every job carries a value, and the worker that completes it keeps the lion's share. By default you earn 70% of each job's value, settled in USDC. The remainder funds the network that found you the work and routed the answer home.

The compute was already paid for and already idle. Everything it earns from here is found money.

The stake-to-work boost

If you stake $ASHAO, your cut rises from 70% to 80%. The network pays its committed workers more because committed workers are what make it reliable — a fleet you can count on at three in the morning is worth more than one that drifts in and out. Staking also earns you a share of network revenue independent of the jobs you personally run, so a serious worker is paid twice: once for the work, once for the commitment.

Stopping is just closing a tab

There is no contract and no penalty for leaving. Close the tab and your worker simply stops claiming jobs; the network notices it has gone quiet and stops routing to it. Come back whenever you have spare cycles. That is the entire point — a network made of people who show up because it is worth their while, and who are free to stop the second it is not. Idle silicon, turned into income, on your terms.

ashao is live. Connect a wallet and the network is yours to use.

Open the chat

Keep reading