Allowances, rate limits, and the ceiling

What counts

One message is one thing a visitor sends. Replies are not counted, and neither is opening the widget without saying anything. The count resets at the start of each UTC calendar month.

What happens when you reach a usage ceiling

The agent stops replying until the relevant window resets. Visitors see a short note that the assistant is unavailable rather than a broken widget.

Every bot also has a daily safety ceiling, about three times its plan's average daily allowance, so one exceptional day cannot consume the whole month. We do not keep going and bill for the excess. If usage is going to be tight, the portal shows this month's count — move up before the ceiling rather than discovering it on a Monday.

Rate limits are a different thing

Separately from the monthly total, there is a cap on how fast messages can arrive. It exists so one visitor — or one script pretending to be one — cannot degrade the service for everyone else sharing the hardware, including your other visitors.

The limit applies per conversation and per visitor address, with a larger burst allowance shared by the bot. Normal typing stays well below it; a script or unusually fast burst is told to slow down and can carry on a moment later. It is a rate limit, not a ban.

Conversation length

A single conversation ends after 60 visitor messages, after which the visitor is invited to start a fresh one. Long threads cost more per message because the history is re-read each time, and a fresh conversation is usually clearer by then.

If you need more

Move up a plan, which takes effect immediately. If none of the plans fit — very high volume, or a spiky seasonal pattern — say so. The numbers exist to separate a small site from a large one, not to ration capacity, and there is usually room to talk.