Tender pools the Macs people already have into one fabric. Tap it for compute cheaper than any provider can offer — or share yours and earn while it sits idle. Either way, your data and models never leave the machine.
Point any OpenAI-compatible client at the fabric. Your calls run on real consumer GPUs at a fraction of hyperscaler prices — because there's no data center, just hardware that already exists.
Flip one switch and the silicon you already paid for becomes income — metered per machine and paid for the work it actually does, not for sitting on.
A fabric of machines people already own and trust — with guarantees a data center can't make.
Jobs run in a sealed VM on the owner's machine. Only the prompt and the answer cross the wire — never your files, never the model.
OpenAI-compatible. Change one base URL and the code you already wrote runs on the fabric — same shapes, same clients.
Sleeping machines wake when there's work and sleep when there isn't — capacity that scales like the cloud, on hardware that rests like a person's.
Prepaid and metered per token. It hard-stops at zero — no overage, no collections, no surprise invoice.
One Google account. Two ways in.
One Google account across web, menu bar, and phone. Nothing to copy, no keys to manage.
Flip a switch to lend a Mac's GPU, or grab an API key to consume. Do both at once if you like.
Jobs find the right machine, wake it if it's asleep, run locally, and meter the tokens. You earn, or you build.
It's the API you already use — answered by someone's Mac, not a mega-cluster.
$ curl https://api.nomio.ws/v1/chat/completions \ -H "Authorization: Bearer $TENDER_KEY" \ -d '{"model":"qwen2.5:7b","messages":[{"role":"user","content":"Explain photosynthesis in one line."}]}' { "choices": [{ "message": { "content": "Plants turn sunlight, water, and CO₂ into sugar and oxygen." }}], "served_by": "Mac Studio", "tokens_per_sec": 131, "cost_usd": 0.0001 }