Usage + quotas

Two numbers shape your bill: conversations and messages. The Usage page shows both, with their caps.

What counts as a conversation

A conversation is one visitor's session with the widget — opens, exchanges N messages, closes. Multiple turns within the same session are one conversation, not N. If the same visitor comes back tomorrow and sends another message, that's a new conversation.

Sessions are tied to the visitor's browser. Clearing cookies or switching browsers starts a fresh session, even from the same human.

What counts as a message

A message is one persisted assistant turn — every time the bot replies to a visitor, that's one. Tool calls and intermediate retrieval steps don't count; only the user-facing reply does. Image uploads count as one message each, same as text.

Conversation caps per workspace

Each workspace has a monthly conversation cap that we set when you sign up — different workspaces get different numbers depending on what you need. When you hit it, the next visitor gets a polite "we're at capacity" message instead of a bot reply. The cap resets at the start of each billing period. Talk to us if you need it raised.

Monthly message cap

Each workspace has a monthly message cap that's shown on Billing and on Usage. You can:

  • Stop the widget when exhausted — the bot returns the configured fallback message instead of replying past the cap. (Trial default.)
  • Keep serving past the cap — the widget keeps working; the overage shows on next month's invoice. (Paid workspace default.)

Pick the policy that fits your tolerance under Usage & Billing → Usage limits → When your monthly budget is exhausted.

Where usage lives

  • Admin → Usage — current period to-date, per-day chart, breakdown by API key, top-traffic pages.
  • Admin → Billing — current period dates and projected end-of-period usage.
  • Notifications — automatic email at 80% and 100% of your monthly message cap if you have billing alerts on.

Why your usage spiked

Common causes, in order of frequency:

  1. Viral page — one of your pages got 10× the traffic for a day. Compare day-on-day in the chart and check the top-traffic-pages breakdown.
  2. Loop in your app — a runaway script keeps re-opening the widget. Check whether one visitor (in Conversations) has hundreds of conversations in a window. If yes, an integration is misbehaving.
  3. Bot is being asked harder questions — a site change that pushed visitors to the widget for complex topics can drive more turns per session.

Reducing your usage

  • Prune the knowledge base. A leaner KB tends to produce more focused answers and shorter conversations.
  • Tighten the personality prompt. Shorter is cheaper and generally produces better-grounded replies.
  • Turn off image uploads if you don't actually need them. Most accounts never do — flip them off under Widget → Privacy.
  • Use suggested replies. Visitors who tap a suggested reply tend to ask focused questions; visitors who type unprompted ramble.

When you hit the cap

You'll get an email notification at 80% (one chance to react before it bites) and at 100%. After 100%:

  • If you set "stop when exhausted", the widget shows your fallback message; existing conversations in flight don't break, but new turns can't start. Wait for the next period or talk to us.
  • If you set "keep serving", the widget keeps working. You'll see overage on next month's invoice.

The cap resets at the start of the next billing period. There's no carry-over of unused capacity — buying more capacity now and shrinking later doesn't bank credits.

Trial accounts behave the same way; whether you have a conversation cap during trial is part of the setup we agreed when you signed up.