How it works
A lane of your own, not a place in the queue
Shared endpoints make you compete for throughput with everyone else. A reserved lane gives you a fixed share of a node for the hours you book, so your speed stops depending on who else is busy.
Yours
Your lane’s throughput does not change when the lanes beside it get busy. That is the whole difference from a shared endpoint.
- 01
Apply and get approved
Create an account and answer a few questions about how you work: time zone, hours of coding per day, your main use case and a rough cache-hit rate. We approve accounts by hand in small batches and email you when yours is ready. Approved accounts start with free credit for five lane-hours to test the lane; nothing is charged.
Beta · manual approval · first 5 hours free
- 02
Buy credits and book hours
Credits come in prepaid packs from $10. A lane is $0.20 per hour and carries two concurrent requests; book one lane for an interactive session or several for a fan-out job, any hour of the day, up to a week ahead. Cancel free until an hour before the slot.
$0.20 per lane-hour
- 03
Your key goes live for the hour
Your API key is permanent, so you set up your harness once. It is admitted the moment a booked hour starts and declined outside your hours. If the lane is ever disrupted during a booked hour, the time is credited back automatically.
Permanent key · automatic service credits
- 04
Run your workflow
Point any OpenAI-compatible client at the endpoint with the model id. Streaming, tool calls, images and optional reasoning all work on the same lane, and your conversation prefix stays cached between turns so follow-ups only read the new tokens.
262,144-token context · images · tools