What happens when the daily cap is reached?

The bot stops answering for the rest of the day rather than continuing to spend. That is deliberate — a cap that can be exceeded is not a cap.

Will rate limiting frustrate real customers?

It should not if the limit is set sensibly. Genuine support conversations are short; a limit well above a normal exchange bounds abuse without ever being hit by a real customer.

Do I need separate caps per client?

If you run bots for several clients, yes — the cap is set per bot, so one client's spike cannot consume another's allowance or your margin.

Configure

Set spend caps, rate limits, and domain locks

Usage-based tooling without a ceiling is an open-ended liability. These three settings close it.

Start freeNo credit card. 2 minute setup.

The failure mode is a bill, not an outage

With usage-based AI the thing that goes wrong is rarely downtime. It is a scraper, a bored visitor, or a copied snippet on someone else's site quietly consuming your allowance until the invoice arrives.

No ceiling on what a single day can cost
A public key that would work on any origin
One visitor able to send unlimited messages
Costs discovered on the invoice rather than at the time

Lock the widget to your domains

List the origins allowed to serve the widget. The public key is visible in your page source by necessity, so this is what makes it safe: on any other domain, it will not answer.

Include every subdomain you use

www and apex are different origins.

Leave staging out of production

Use a separate bot for staging rather than widening the lock.

Rate-limit each visitor

Set the maximum messages one visitor can send in a window. This bounds a single bad actor without affecting a genuine customer, who almost never sends more than a handful of messages in a session.

The per-visitor rate setting with four options — relaxed, balanced, strict and custom — explaining that relaxed allows twenty messages a minute and that a bot-wide ceiling also applies. — There is a bot-wide ceiling behind the per-visitor one, so a burst from many visitors at once cannot slip past it.

Set the hard daily cap

The ceiling for the whole day, across all visitors. Once it is reached the bot stops answering rather than continuing to spend. Choose a number that comfortably exceeds a normal day but that you would be content to pay on the worst day.

The daily message cap setting, a slider between 25 and 1000 messages per day, described as a safety limit rather than a plan limit, showing yesterday's usage at zero. — The product calls it a safety limit rather than a plan limit, which is the right way to read it.

Revisit after two weeks of real traffic

The right numbers are the ones your actual usage suggests, not the ones you guessed before launch. Look at a fortnight of real data and set the cap somewhere above your busiest day rather than at an arbitrary round number.

Common questions

What happens when the daily cap is reached?: The bot stops answering for the rest of the day rather than continuing to spend. That is deliberate — a cap that can be exceeded is not a cap.
Will rate limiting frustrate real customers?: It should not if the limit is set sensibly. Genuine support conversations are short; a limit well above a normal exchange bounds abuse without ever being hit by a real customer.
Do I need separate caps per client?: If you run bots for several clients, yes — the cap is set per bot, so one client's spike cannot consume another's allowance or your margin.