Rent a machine that runs local LLMs

A whole machine to yourself · unlimited tokens · nothing kept · rent as many as you need for concurrency — by the hour, day or month

Rent now From ¥2.77/hour

Not a relay

Models run on the machine you rent — nothing is forwarded to a third-party API

Your data stays yours

The whole machine is yours; we only record who, which model and how much — never the content; your cache and results are wiped on switch or at the end

No interruptions

If a machine has trouble we switch you within 2 minutes, with shared compute free in between; downtime beyond 10 minutes is compensated by the minute

Scale out as far as you need

More machines means more concurrency — requests spread across them automatically; add another anytime

Zero setup

Claude Code, Codex, your API keys and Tutu suite apps automatically prefer your rented machines

How we make money

From the rent. When nobody's renting, the machines serve shared compute — so we never need to cap tokens or concurrency

Configurations and prices (live)

Rental rules

  • The whole machine is yours alone; priced by hour / day / month, always the cheapest combination for your duration
  • If a machine drops, we switch you to another within 2 minutes; meanwhile your requests use shared compute for free
  • End early anytime: we recompute using time used (rounded up to the hour) and refund the difference
  • Cancel before the start for a full refund; any machine that can’t arrive on time is refunded in full automatically
  • Refunds go back the way you paid (vouchers back to vouchers); no content is recorded, and your cache and results are wiped on handover and at the end

FAQ

How is this different from a pay-per-call API?
You pay for time and the whole machine is yours: unlimited tokens, no extra charge for long context or concurrency, and more machines give you more concurrency.
Is my data safe?
During your rental the machine runs only your requests; we record who, which model and how much — never the content of requests or replies — and wipe your cache and results on switch, handover and at the end. You can also tick “data-center machines only” when ordering.
What if a machine breaks?
We move you to another machine of the same configuration within 2 minutes; meanwhile your requests use shared compute for free, and any downtime beyond 10 minutes is refunded by the minute.
Which models can I use?
Any open model the machine can hold (text, images, video): tick it when ordering and it’s installed before you start; during the rental you can change parameters in Rental settings and they apply within seconds.
How do I connect?
No configuration changes: the Claude Code, Codex, API keys and Tutu suite apps you already use automatically prefer your rented machines during the rental; when they’re all busy you can choose to queue or overflow to the shared pool.
Can I end early or get a refund?
Yes. Anything not started is refunded in full; ending early recomputes using time used (rounded up to the hour) and refunds the difference; refunds go back the way you paid (vouchers to vouchers), and anything that can’t be refunded is shown to you plainly.

Try it for an hour — end early anytime and get the difference back