Plan, bTokens and limits
What counts as a bToken, what happens when the plan runs out, and how the AI allowance works.
Everything this guide describes lives in Account → Plan and usage, and almost all of it belongs to the owner, who is the one paying. The values in force (prices, limits, features of each plan) are always on that screen; this guide explains the mechanics.
To switch plans, see change plan. For payment methods and invoices, see payments and invoices.
What counts as a bToken
The plan is measured in bTokens per month, with rules designed not to penalise normal typing:
- A typed search counts as one bToken, counted when the typing settles and not on every keystroke. Correcting the term in the same burst is still one bToken.
- Indexing counts a token amount (10 bTokens per 10,000 indexed items) and is never billed: it doesn't touch the limit. Keeping the catalogue up to date costs nothing.
- The recommendation carousel on product pages counts at a token rate of its own: 10 bTokens per 1000 times it's shown. It appears on every visit to a product page, and its number follows the store's traffic and not the use of search; at one bToken per view, it was the piece that ate the most of the plan. The recommendations inside the search box do count one bToken per opening, like the rest of search.
- Each direct API call counts one.
- The playground doesn't count.
The window is the billing cycle, not the calendar month: a plan that started on the 20th counts up to the following 20th. The meter at the bottom of the sidebar shows the cycle's spend at all times, and the plan screen adds the projection for the end of the cycle, so that the warning arrives before the problem.
The same screen shows the detailed usage: a chart with each day's bTokens stacked by feature (searches, assistant conversations, comparisons, recommendations, and so on), and the cycle table with the weight of each one. It's the answer to "what did I spend the bTokens on?", and only what gets billed is counted there: indexing, which never counts against the limit, sits on a line of its own below the table, with the cycle total and a note saying it isn't billed. Live detail exists since 6 August 2026; earlier days were rebuilt from the logs and settled against the billed totals of each cycle, with the part the logs don't explain shown as "Not itemized".
When the limit runs out
The store never stops. Once the plan is used up, search, the assistant and everything else keep answering exactly as before; bTokens above the limit start being charged at the plan's overage price (shown on the screen and on the plan cards in euros per bToken, for example €0.0040/bToken on Pro) and the amount is added to the next cycle's charge.
So that none of this is a surprise, three emails go out to the account owner:
- at 75% of the plan: a simple notice, with nothing to do;
- at 90%: the warning that, once the plan is used up, the overage starts being charged, with the rate stated;
- at 100%: the confirmation that overage billing has started, and until when.
A larger plan usually works out cheaper than recurring overage: the plan screen shows the current cycle's overage, and changing plan is one click away.
The AI allowance
The AI features (the assistant, AI search, generative suggestions) have an allowance of their own, sold with the plan in tokens per cycle (for example, 10M on Pro; a conversation turn runs about 3,000). The plan screen shows the allowance bar next to the bTokens one, and the warnings go out by email at 90% and on running out. Once the allowance is used up, AI pauses until renewal: the assistant leaves the store and the deterministic paths carry on; search itself never stops. In "Store → AI in the store" you can also set a daily cap on tokens, to control the pace of the spend without touching the plan. Deleting conversations on the customer's screen doesn't give the allowance back: the spend is measured by what was consumed, not by what stayed in view.
When a feature isn't in the plan
The panel doesn't hide the screen: it shows what the feature does and states from which plan on it exists, so that the upgrade decision is made with the feature in plain sight, and not with a menu entry that vanished without explanation.