Plans count answers, not chats, messages or seats.
An answer is one reply a bot writes on tufu's AI, in any channel: a chat bubble on your site, an Ask console in your docs, a WordPress plugin, email, Slack and Discord. A reply on Claude Haiku 5.5 with reasoning off is one answer. On Team, a bot can run on Claude Sonnet 5.5 or Claude Opus 5.5, or reason more, and each reply then counts a set number of answers, shown wherever you choose. Repeat Ask questions served from the cache, suggested replies, summaries and chat titles are free. On Team, answers on your own AI key are free and unlimited.
Team includes 4,000 answers a month on tufu's AI, and Free includes 200. See Plans, answers and packs for everything else in each plan.
About how many conversations#
A support conversation usually takes a few replies, so on Claude Haiku 5.5 with reasoning off, 4,000 answers are about 1,000 conversations, and 200 are about 50. A visitor with one quick question uses one answer; a long back and forth uses more, and so does a stronger model or more reasoning.
How each channel counts#
Every reply the bot writes counts, wherever it answers (one answer on Claude Haiku 5.5 with reasoning off; more on a stronger model, below):
| Where | What counts |
|---|---|
| Support chat | Each reply the bot writes. |
| Ask console | Each answered question, follow-ups included. A repeat question served from the cache is free. |
| Support email | Each reply that is emailed. A draft held back for a person never counts. |
| Slack thread | Each reply the bot writes in the thread. |
| Discord thread | Each reply the bot writes in the thread. |
Models and reasoning#
On Free, bots answer with Claude Haiku 5.5, reasoning off: one answer a reply. On Team, each bot can use Claude Haiku 5.5, Claude Sonnet 5.5 and Claude Opus 5.5 on tufu's AI, with reasoning off, low, medium or high, in the bot's AI settings. A stronger model or more reasoning costs more to run, so each reply counts more answers. The number comes from the model's price and is shown next to every choice:
| Model | Off | Low | Medium | High |
|---|---|---|---|---|
| Claude Haiku 5.5 | 1 | 2 | 3 | 6 |
| Claude Sonnet 5.5 | 20 | 28 | 51 | 111 |
| Claude Opus 5.5 | 56 | 56 | 101 | 222 |
Claude Opus 5.5 always reasons a little: off counts and runs as low. Reasoning also makes a reply slower to start: the chat shows "Writing" once the bot has found what it needs.
If your team moves to Free, bots answer with Claude Haiku 5.5, reasoning off, until you upgrade again; the choice you saved is kept and comes back.
What's free#
- Repeat Ask console questions served from the cache.
- Suggested replies, summaries and chat titles.
- Your team's own replies, from the inbox, email or Slack.
- Notifications.
Replies in Preview do count: it's the real bot, on the real meter.
On your own AI key#
Answers on your own AI key are counted apart from the ones on tufu's AI, one per reply whatever the model, since your key pays:
- on Team, they're free and unlimited (Plan and usage still shows how many);
- on Free, you get 1,000 a month, on top of the 200 on tufu's AI.
Seeing your count#
Admins see the month so far under Plan and usage: answers used on tufu's AI and on your own key, any left from packs, and when the count resets, with the replies and the answers they counted split by model.
We email your admins when you've used 80% of the month's answers. What happens after that is in When you run out. Visitors never see a message about plans: when the bot can't answer, they're offered a person instead.