How it works
Three steps: one API, GPUs we operate or vet, idle time put to use.
1
One OpenAI-compatible API
Point your existing OpenAI client at our base URL and API key. No other code changes.
2
The request runs on a GPU
Your request runs on a GPU we operate, or — for models we do not host ourselves — a provider required not to retain request content. See docs for which model runs where.
3
Idle time lowers our cost
Between requests the same GPUs mine PRL, which lowers what it costs us to keep that capacity ready for you. Outside GPU owners can connect their own cards at linkrra.com/providers; no outside card has served a paid request yet.