How it works
- Create an account at app.garageai.eu and open Connect your GPU.
- Run one line on the machine with the GPU. It installs GarageAI Bridge, a small open-source program (
garageai), which looks at your GPU, memory, runtime and models. - Approve your computer in the browser. Bridge opens app.garageai.eu with a code; check that it is your computer, name your garage and approve. No token is copied anywhere.
- Go live. Your models pass an acceptance test with a real request, the page says Your garage is live, and you set your price per million tokens in the portal.
What you need
- macOS (Apple silicon), Linux or Windows, with a GPU.
- A runtime with a model loaded: Ollama, LM Studio, llama.cpp, vLLM, SGLang or Xyntetik Runner. No runtime yet?
garageai runtimechecks whether Xyntetik Runner, a one-file engine made in Sweden, fits your machine, andgarageai runtime installsets it up with a model. Bridge keeps it up to date from then on: it updates when the machine is quiet and rolls back if anything fails. - A normal home or office internet connection. You open no ports on your router.
What you get back
Access to more models. Other garages share models you may not find elsewhere: community fine-tunes, distills and models in other languages, all through the same API.
Credits. You set the price per million tokens, input and output. You earn it for every token your garage serves, whether a buyer picked your garage or the request came through the pool, and there is no platform fee during launch.
- Now: every token your garage serves is counted at your price and shown in My garages.
- Next: use your earnings as GarageAI credits on any model, for example larger models than your own machine runs. Everything earned from launch carries over.
- Later, not yet scheduled: payouts to your bank account through Stripe.
What Bridge does on your computer
- Its own encrypted tunnel. WireGuard, built into Bridge. Your machine dials out to the EU gateway, so it works behind any router and through NAT. No VPN app, no network interface, no ports opened, and it runs fine next to Tailscale.
- Your runtime stays on localhost. Bridge forwards one port into the tunnel, and only inference requests pass: downloading or deleting models and every other management call is refused. Nothing else on your computer or your home network can be reached.
- No secrets leave the machine. The tunnel key is made on your computer; only its public half is sent. Bridge never reports prompts, keys or command lines.
- It keeps itself up to date. A heartbeat reports your models and their context windows every 5 minutes, and Bridge updates itself to the current stable release, checked against its checksum.
- It tells you what is wrong.
garageai doctorshows what runs on the machine, what is wrong and exactly how to fix it. - GarageAI in your menu bar. On macOS (menu bar) and Windows (notification area), a small GarageAI icon shows at a glance whether your garage is live, which models it serves and how many requests it handled in the last hour, with Open My garage and Check this machine. It is read-only and optional;
garageai connectoffers it at the end. - It measures energy. On NVIDIA GPUs and Apple silicon, Bridge samples power every 5 seconds and reports the energy your garage used while serving: watt-hours and tokens per kWh, per model. The menu shows today’s figure, for example Energy today: 420 Wh · 1.2M tokens/kWh. Only the model name and the token counts are kept; prompts and answers are not stored.
Your machine stays yours first
Your own garage is available to you from your phone or any other computer, privately and free, whether or not it is on the market: see My machine. While you use it, buyers are sent to other garages. And other people’s prompts are confidential: the operator terms forbid logging, storing or reading them.