Fair usage

Applies to all paid plans and trials.

Your plan is designed for one person using ChatRouter.ai as much as they like. There is no message counter, no credit balance and nothing to top up. Ask as much as you want, on whichever model you want.

What fair usage means

A very small number of accounts use many times more computing power than everyone else — usually automated tools, resold access, or many people sharing one login. To keep the service fast and the price the same for everybody else, those accounts can be slowed down temporarily.

What you would actually see

• Nothing at all, in normal use. Most people never reach any of this, on any plan.
• Under sustained heavy use, replies may come from a faster model for a while. You are told when that is happening.
• In the most extreme cases, new messages pause for a short period. We tell you roughly when you can carry on, and it lifts on its own — you never have to ask us, and you never have to buy anything to end it.

What is not allowed

• Sharing an account with other people, or reselling access to it.
• Automated or scripted use, including scraping and bulk generation.
• Anything that breaks the law or the underlying model providers’ rules.

These are the only cases where an account is suspended rather than slowed. If that happens to a paid plan, the unused part of what you paid is refunded.

Improving Auto

Auto reads your message before it chooses a model for it. To keep improving that choice, we keep a shortened copy of the message it read — with email addresses, phone numbers, payment card numbers, IP addresses and anything shaped like a password or key masked out automatically.

Masked is not anonymous: the copy stays linked to your account. It is kept to improve model routing, and it is not deleted when you delete the conversation.

If you rate an answer — a thumbs up or down, asking for a different model, or picking one by hand straight afterwards — that rating is kept in the same way, against the routing choice it was about. It is a score on the choice and nothing more: it holds none of your message and none of the answer, and deleting the conversation still deletes both of those.

Why we do not publish a limit

The thresholds are based on the real computing cost of the models you use, which differs enormously between a short question and a long piece of analysis — so a single published number would be misleading rather than helpful. We tune the thresholds so that ordinary use, including heavy ordinary use, is never affected.