Install & start the worker
Download the small shareai-worker script, run it with your contributor token. Only Python + a running Ollama server are required.
Contributor program
Your machine can work the night shift. You shouldn't have to.
Already running open models locally with Ollama? Install the ShareAI worker. It notices which model you're running, serves real chat requests while your machine is on, and pays you in USDC on Base. No API keys, no cloud account, and stopping is as easy as closing the terminal.
Connect wallet to start earning
5 minutes if Ollama already runs
while you were out
Request served by your local model
Per-token earningscredited automatically
How earning works
Three steps, then it runs without you
It advertises your model automatically
The worker detects which model your Ollama is serving and its capacity, then reports it to the pool. While your machine is on, your offer is online.
Get paid automatically
Requests are routed to your machine when it's your turn. You earn per processed token at the price you set, paid out in USDC on your Base wallet.
Important before you join
The fine print, in plain words
Generation only, never an agent on your host
The worker only sends prompts to your local model and returns the generated text. It never runs tools, commands, or file/network actions on your machine. If a request uses the model as an agent, that execution happens only on the user's side.
You can see the prompts you serve
Because inference runs on your hardware, your machine receives the prompt text in plaintext. The worker never logs or stores it, and you must not inspect or retain it. Users who need privacy can exclude local workers (Confidential mode).
Your machine, your hours
You only earn while your computer is on and the worker is running. Stop it any time; your offer disappears from the pool within seconds.
Payout in USDC
Earnings are paid out to your specified Base wallet once they reach the minimum payout threshold.
Model licences are your responsibility
Some open-weight licences restrict commercial or paid serving. Make sure the model you run is licensed for answering paid requests.