Fair use
Limits and fair use.
The terms of service, the acceptable use policy and the Founding Lifetime checkout all point here for the current figures. This is that page. Every number on it is the number enforced in the code, not a softer one written for marketing.
The ceilings
Daily limits by plan.
Every ceiling is counted in model tokens per UTC day, input and output added together, across the whole workspace.
| Plan | Ceiling per UTC day | Why it is set there |
|---|---|---|
| Trial | 20,000 tokens a day | Deliberately the lowest ceiling on the price list. A trial has no card behind it, so it must not be able to run up a bill. |
| Pro | 200,000 tokens a day | One person, one seat. This is the highest per-person ceiling and it is well above what ordinary use of a few busy channels consumes. |
| Team | 120,000 tokens a day, plus 80,000 for each seat | At the three seat minimum that is 360,000 tokens a day for the workspace, shared across everyone in it. Adding a fourth seat raises it to 440,000. |
| Founding Lifetime | 120,000 tokens a day | The fair-use ceiling stated at checkout. A one-time payment cannot fund unbounded processing forever, so this one is lower than Pro and it is written down before you buy rather than afterwards. |
| Lapsed trial or cancelled plan | No extraction | The ceiling is zero, so nothing new is extracted. Your existing list stays readable and exportable, and nothing is deleted because a paid plan ended. A trial that never converts is the one exception: that account is deleted 90 days after the trial ends, with an email 7 days before. |
The ceiling belongs to the workspace, not to a person or a device. On Team it is shared: the whole workspace draws from one pool that grows with each paid seat, so a quiet week for one person leaves more room for everybody else.
Reaching one
Over the limit, work waits.
Hitting a ceiling pauses new extraction. It does not delete anything, it does not lock you out, and it does not silently drop the messages that were waiting.
- 01
Extraction stops making model calls
The check runs before the call, not after it, so the token that would have taken the workspace over the line is never spent. That is what makes the ceiling a real stop rather than a warning you get once the money is gone.
- 02
The pending batch goes back on the queue
Messages that were waiting to be read are re-queued with a delay that runs to the next midnight UTC. When the day rolls over the allowance resets and the queued work runs, so the commitments in those messages arrive late rather than never.
- 03
The app says so, in the app
Your plan, the ceiling, how much of it is used, how many batches are waiting and the exact time processing resumes are all readable while you are throttled. A paused list that looks like an empty list is a bug, not a design.
Everything else keeps working
The list you already have stays there and stays searchable. You can mark things done, snooze them, drop them, edit dates, chase people, read the brief and export the lot. The pause is on reading new messages, and nothing else.
Where the queue stops waiting
A batch is re-queued across at most three consecutive resets. A workspace that is still over its ceiling on the fourth day is using far more than the plan covers, and holding those messages indefinitely would just build a backlog nobody can clear, so at that point the batch is dropped and recorded. If that is happening to you, the plan is wrong for the workload and we would rather fix that than keep queueing.
What is measured
What counts.
Draws on the ceiling
- Extracting commitments from a window of messages in a channel you opted in to.
- Re-checking an open commitment against later messages to see whether the thread closed it.
- Classifying a message as relevant or not before the more expensive pass runs.
Does not draw on it
- Reading, searching, sorting, filtering or exporting the list you already have.
- Marking something done, snoozing it, dropping it, or editing a due date by hand.
- The morning brief and the evening close, which are assembled from commitments that were already extracted.
- Sending a nudge, which is a note on your own list rather than a message to anybody.
- Anything in a channel you have not opted in to. Those are never read.
The other limits
Other limits.
Retention, 30 days by default
Raw message text is deleted on your retention schedule, which is 30 days and which a Team admin can set anywhere from 1 day to 12 months. Search reaches back exactly that far. The commitments extracted from it outlive the text and stay until you delete them. Turning a channel off stops it being read; it does not delete what was already stored, which runs out on the same window.
Team seats, minimum 3
Team is sold in blocks from three seats up. Below that, Pro is the better and cheaper product, and every extra Team seat also raises the workspace token ceiling.
Platform limits we do not control
Discord gates message content behind a privileged intent, a Telegram bot in a group sees only messages addressed to it unless an admin turns privacy mode off, and Slack rate limits are applied per method. Those are their limits, and each connector page states the ones that apply to it.
Changes
When limits change.
The figures above are not promises fixed for all time. Model pricing moves, extraction gets cheaper or more thorough, and a ceiling set today may be the wrong one in a year. So Lupee reserves the right to change them, under three rules.
- A ceiling that goes up takes effect when we publish it. Nobody needs warning about more room.
- A ceiling that comes down on a plan you are already paying for is announced at least 30 days before it applies, by email to the workspace owner and on this page. If the new figure does not work for you, the terms let you cancel and be refunded for the unused part of the period you paid for.
- The Founding Lifetime ceiling is the one stated at your checkout. We can raise it. We will not quietly lower it later to make a one-time payment look like a mistake.
The page carries the current figures at all times, because the terms of service and the acceptable use policy both send people here instead of restating numbers that would go stale in a legal document.
Read the surrounding commitments in the terms of service, the acceptable use policy and the refund policy. If a number on this page and a number in one of those documents ever disagree, this page is the current one and the mismatch is a bug worth reporting.
Questions
Questions.
Why is the ceiling counted in tokens rather than messages?
Because tokens are what the work actually costs. A hundred one-line messages in a quiet channel cost a fraction of what ten long threads with pasted documents cost, and a message count would charge both the same. Counting tokens means a light user is measured lightly.
Does reading a channel use tokens even when nothing is committed to?
Yes, a small amount. Extraction has to read a window of messages before it can tell you there is nothing in it. Only channels you opted in to are read at all, so turning a noisy channel off is the fastest way to bring usage down.
Can I buy more?
Not today. There is no top-up product, and we would rather say that plainly than sell one that does not exist yet. If your workspace keeps hitting the ceiling, write to support and we will look at the account before anything else. Team seats raise the workspace ceiling, so on a growing team that is usually the answer.
Can my workspace ceiling be lower than the plan figure?
Only if an admin sets it lower on purpose. A workspace-level budget can bring the ceiling down, for example to keep spend predictable, and it can never raise it above what the plan allows.
Do these limits apply to the mobile app?
They apply to the workspace, not to a device, because the processing happens in one place regardless of where you opened the list. That will stay true of the phone apps when they ship: one ceiling, however many surfaces.
Lupee opens by invitation.
Every account starts with a 14 day trial when its invitation arrives.