The token allowance

Madebook meters nothing itself. Every model call it makes goes through the Bookbag account service, which writes down what the call cost as the tokens are spent — and refuses one the platform is no longer willing to pay for.

That is the same allowance Openbook and Bookbag spend, because it is the same account.

It caps what the platform pays for, not what you may use

This sentence is the whole page. Everything else is a consequence of it.

Your organization is in one of two modes, set on the organization's AI self hosted page at Bookbag account (see AI configuration):

	Managed by us	Self hosted
Whose key runs the model	The platform's	Your organization's own
Who pays the provider	The platform	You, directly
Counted against the allowance	Yes	Recorded, but as zero
Can be refused for being over it	Yes	Never

A self-hosted organization costs the platform nothing, so there is nothing for the platform to cap. Its runs are still recorded — you can see exactly what they used — but no figure they reach will stop them. A managed organization runs on the platform's key, and the allowance is how much of that the platform will carry.

Neither mode falls back to the other, deliberately. Running out in managed mode does not quietly start spending your key, and a broken self-hosted key does not quietly start spending the platform's — see AI configuration.

One allowance per account, shared across every product

The allowance belongs to an account — one person's Bookbag login — and it is charged to the owner of the organization, not to whoever started the run. This is exactly how storage works, for exactly the same reason: one account service sits behind all three products.

• Every organization that account owns draws on the same number.
• Every Bookbag product draws on the same number. A mission run in Madebook moves the usage shown on the organization's AI page at Bookbag account, and the meter in every product's account menu.
• A member running a mission spends the owner's allowance, not their own.

The default is 1,000,000 tokens per window, unless a platform admin has changed it or set a different figure for that one account.

What counts as a token

For a managed run, everything it actually consumed:

• fresh input — the part of the prompt the provider processed from scratch,
• output — what the model wrote back,
• cached input — counted in full, not at a discount.

Cached input counting in full is a decision, not an oversight. The allowance is a budget of work, and somebody watching the bar should not have to understand a provider's cache pricing to predict it. The three figures are stored separately on every run, so the rule can change later without rewriting what already happened.

A generated image counts as a flat number of tokens — 10,000 by default, set by a platform admin. An image is not made of tokens, so that is a conversion rate rather than a measurement.

A self-hosted run is recorded with zero billed tokens; its real counts are kept and reported separately.

This is not the same measurement as the token counts on an AI run. Those are Madebook's own receipt of what a run did — input, output, cache read and cache write, per run, next to the context items and skills that made it up. The allowance is the account service's record of what the platform paid for. They count the same work; only one of them decides anything.

Madebook does not measure money anywhere. AI runs used to carry a cost_cents column; it was dropped, because the figure it was reaching for does not exist. A self-hosted organization is billed by its provider directly and the platform is not in that invoice, and a managed one is metered in tokens against this allowance. Tokens are the unit; there is no second one.

Windows, not a monthly reset

Nothing resets. Usage is the sum of what was recorded since the current window began, and the window is computed from the period every time anyone asks.

• Monthly means the UTC calendar month — the 1st, because that is what "monthly" means to a person reading an invoice.
• Weekly means seven-day blocks from a fixed anchor, set to the most recent UTC Monday when an admin chooses weekly.

Because nothing was zeroed, last window's figure is still readable, and changing the period changes only what the window means from then on — every run keeps the window it actually fell in.

Where to read it

Madebook has no AI page of its own any more; the one it had, with the usage bar under the mode switch, is gone. There are two places to read the allowance.

The organization's AI self hosted page at Bookbag account has a Usage card:

• "{used} of {limit} tokens used this month" (or this week), the date it resets, and any carried-over credit, over a bar;
• when it is spent: "The allowance is used up. Managed runs are refused until the window resets or a platform admin raises it.";
• This organization, by product — tokens and run count for each product, so a bar that moved without anyone running a mission in Madebook is explainable;
• On the organization's own keys — what ran on the organization's own keys, and on members' personal keys, shown beside the allowance and never added into it.

In self-hosted mode the card says the organization's runs are recorded but never metered against the platform allowance.

The meter in Madebook's account menu, below.

The meter in the account menu

You do not have to leave Madebook to know where you stand. The account menu (your picture, top right) carries a thin bar for the organization you are in: AI tokens this month, used against the limit, with the turnover date and what is left. It turns amber at 85% and red when the month is spent, where it reads "Used up — turns over 1 October. AI actions are paused until then, or add your own key." add your own key opens the organization's page at Bookbag account in a new tab. In self-hosted mode the bar is replaced by what your own key has run, because no platform limit applies.

The same meter sits at the foot of the account menu in every product and at the Bookbag account service itself, where it shows your own account's allowance. The product pages no longer draw their own usage bars: one number, one place it is metered, one meter drawn from it.

When the allowance runs out

Every AI button disables itself first. Suggest, Different ones, Reindex & regenerate, Read my repository, Find decisions, Derive from the repository and Build/Rebuild knowledge all read the same allowance the meter draws, and once it is spent they cannot be pressed: hovering one says "AI allowance used up — 1.03M of 1M tokens this month. It turns over on 1 October 2026. Add your own provider key on the organization's page at Bookbag account to keep going." The Decisions and Architecture Rules pages say the same in a notice under their heading, with a link that opens the organization at Bookbag account. A person learns they are out before pressing anything, not from a failed run's error text afterwards. The gate never closes for an organization on its own provider key.

A managed run that starts over the allowance is refused before any provider is called — the model is resolved from the account service's own tables, so nothing has been spent and nothing will be.

The refusal is one sentence, and Madebook shows it exactly as it arrives:

Token allowance reached (1,020,500 of 1,000,000 tokens used this month). It turns over on 1 October 2026. An organization running on its own provider key (self-hosted mode) is not metered at all, or a platform admin can raise this account's allowance.

It names the figures, the date it turns over, and the two ways out. Nothing rewrites it, because nothing else knows those numbers — a refusal replaced by "the AI service could not be reached" is how somebody ends up filing a bug against a working system.

A refused run is not retried. The queue marks this failure terminal: three attempts at thirty seconds, two minutes and eight minutes against an exhausted allowance would spend ten minutes proving what the first attempt already said, and would bury the sentence under an attempt count. It fails once, with the sentence, and stops.

Nothing that is not a model call is affected. Workspaces, repositories, the brain, missions already written, changes, capsules, the attention queue, the audit log and everything a person does by hand all work normally.

The two ways out

1. Move the organization to its own provider key. On the organization's AI self hosted page at Bookbag account, switch to self hosted, add a key and a model, and nothing in that organization is metered or capped again. An owner or an organization admin can do this without asking anyone.
2. Ask a platform admin to raise the account's allowance. They set it per account, at the account service, not in Madebook.

Waiting also works: the window turns over on the date the sentence names.

One run can cross the line. The check asks whether the allowance is already spent, not whether the run you are about to start will fit — nobody knows what a completion will cost before it runs, and guessing would refuse work that would have fitted. So one run may finish slightly over, and the row recording it is exactly what refuses the next one.

A retried request is charged once

Every run is recorded under a key. A retry of the same logical call is charged once, even though it really does ask the model again. A genuinely new step gets a new key, so this is not a way to make repeated work free.

Where a platform admin changes it

At the Bookbag account service, on its AI tokens page (/admin/tokens) — beside the storage page it is modelled on. Getting in needs a platform administrator account at the account service; being a Madebook platform admin is not by itself enough.

There is no link to it from Madebook. Openbook's and Bookbag's admin sidebars both carry an AI tokens ↗ row under "The platform brain"; Madebook's carries Models & providers ↗ and Model catalog ↗ and stops there. A Madebook platform admin has to go to the account service's admin area and find the page themselves. That looks like an oversight rather than a decision, and it is worth raising.

A platform admin sets there:

• The default allowance — the most one account may use per window, across every organization it owns and every product it uses.
• Whether it turns over monthly or weekly. Monthly means the 1st, in UTC. Weekly means seven-day blocks from the most recent Monday, re-anchored the moment weekly is chosen. Changing it changes what the window means from then on; nothing already recorded moves.
• What a generated image costs in tokens.
• A per-account exception, found by the email the person signs in with, with a note saying why. Clearing it puts the account back on the default.

The same page shows the window's totals — billed, accounts spending, what ran on organizations' own keys, how many runs were refused — and the ledger itself: one row per run, never edited and never deleted, with a refusal kept as a void row so "why did this stop at 4pm" stays answerable.

Credits, if you have heard of them

Madebook never billed credits. There was nothing to retire here.

The other two products did. Openbook kept a per-organization credit ledger and Bookbag a per-person balance, and both are gone — tokens are metered once now, where they are spent. Any balance anyone held was carried across to the account service as a one-off opening adjustment: headroom in the current window, not a raised allowance, because an allowance is a recurring promise and a balance was granted once.

In the figures the account service reports, a carried balance appears as credit, used never goes below zero, and remaining is the allowance plus that credit. If you share an account with either of the other products, that is why your bar might sit at zero for a while.

One leftover: the model rows at the account service still carry a rate field from the credit era. Nothing charges against it any more.

Where to go next

• The mode switch and what each side asks for: AI configuration
• The same idea with bytes as the unit: Storage and files
• What Madebook records about each run: AI runs
• Which models are permitted at all: Providers, models and clients