AI points and limits

Hosted AI in flowss is paid for with AI points. Plus, Ultra and Enterprise include a monthly allowance of points, you can top it up with packs that never expire, and every hosted AI request has a published price in points that is reserved when the request starts and charged only when its work reaches you. This page is the complete reference: the allowance on each plan, the price of every AI feature, exactly when you are and are not charged, how points are held while a request runs, where to watch your balance, how to buy more, and every daily limit and size limit that can stop a request, with the message you will see and what to do about it. If you use your own AI provider key instead, points are not involved at all; see Bring your own AI key, and for the bigger picture AI in flowss.

At a glance

How the allowance works

Each plan includes a fixed number of hosted AI points a month:

How the balance behaves:

• One point is one ordinary request. The price of everything else is worked out from how much AI work it does compared with that ordinary request, so a run that does eight requests' worth of work costs 8.
• The month is a calendar month in UTC. The allowance refills on the 1st at 00:00 UTC, whatever day you subscribed. Your balance shows the new allowance the first time it is read in the new month.
• Monthly on every plan. An annual subscription still refills monthly; it is not a year of points delivered at once.
• Allowance first, then bought points. A request is paid from what is left of this month's allowance, and only the remainder comes out of bought points. That way bought points are touched only once the included points are gone.
• Bought points never expire. They carry over from month to month. If your subscription ends or you move to Free or Starter, bought points stay on your balance and hosted AI keeps spending them until they run out; you cannot buy more on a plan without an allowance.
• Your balance never goes below zero. If two requests finish at the same moment against your last points, the balance stops at zero.

What each AI feature costs

Every hosted AI request is priced before it runs. On your own key, every row costs 0 points.

Diagrams in the Studio and the other studios

Images

Agents

How a run that does not finish is charged is described in When you are charged.

Study, the Academy and Evidence

Projects, integrations and everything else

Note: The same short summary appears on your account page and the pricing page: "1 point per standard AI call; an Architect run 4, an agent run 8, a deep-specialist run 47; a flow costs the sum of its steps; an attached image adds 2."

Why the prices differ

A point stands for one ordinary request's worth of AI work. A request that carries an image costs twice as much to serve; a course, a multi-figure project or a TikZ document asks the AI for many times the output of one diagram, so it is priced by the size of output it is allowed; an agent run makes about eight requests; and a deep specialist reasons on the most capable model across about twelve steps. The price is fixed in advance, so a run that happens to finish early still costs its listed price, and one that runs long costs no more.

When you are charged

The rule is the same everywhere: you pay for AI work that reaches you, and nothing else.

You are never charged when:

• the request is refused before it reaches the AI: not signed in, no points, a plan that does not include it, a daily limit, a request that is too long, a file that is too large, an invalid URL;
• the AI provider fails, times out or is unavailable before any answer reaches you;
• the request is cancelled before anything was delivered;
• the AI declines the request ("The model declined this request. Rephrase it rather than resending it as is.") or its answer is cut off at the output limit ("output too large for this request — try a smaller diagram") on diagram generation and refinement, conversions and study cards. The exception is an answer that streams in as it is written (Describe it, the Academy tutor, Evidence Ask and Weave Generate on a branch): once its first part has reached your screen it is charged, and stopping it, closing the tab or hitting the output or time limit after that does not undo the charge.

When the work is delivered, the moment of charging depends on the feature:

Agent runs

An agent run's price is held from your balance when it starts and settled when it ends:

Your account page lists your last 20 runs under Agent runs ("Your last 20 · times in UTC") with When, Agent, Outcome and Points. See The flowss Studio Agent and its modes.

Holds: points reserved while a request runs

When a hosted request starts, its full price is reserved from your balance as a hold. When the request ends, the hold is turned into the actual charge (which may be smaller, or zero) and the rest is released straight away.

What this means for you:

• Points held by a running request cannot be spent twice. If you have 10 points and start an 8-point agent run, a second request costing more than 2 points is refused as out of points until the run finishes, even though your balance still reads 10.
• Holds are released as soon as a request ends on almost every feature, whether it succeeded, failed or was cancelled. Where a failure skips that release (a failed TikZ export, for example), the hold simply expires, as below.
• A hold cannot get stuck. If a request is cut off in a way that stops it from settling, its hold expires by itself, within ten minutes at the latest, and nothing is charged for it.
• Holds are not listed in your points history; only actual charges and credits are.

Where you see your balance

The AI reading

flowss sums up your AI situation in a few words, the AI reading, and shows it in several places (listed below). It says one of these:

The reading refreshes when a studio opens and after AI requests. It reflects a key kept in this browser; a key saved to your account is used just the same but does not change the reading.

Where the reading appears

The chip on the Studio prompt is described on The flowss Studio Agent: prompt and quick help. (The prompt's AI settings button holds Dictate a change, not the reading.)

The account page

The AI points card on your account page shows:

• your total spendable balance as a large number;
• the breakdown, for example "150 of 200 monthly + 0 purchased"; on a plan without an allowance but with bought points, "40 purchased (no monthly allowance on Starter)"; otherwise "Free includes no hosted AI" (with your plan's name);
• a bar of the allowance used this month, green, then amber from 80% used and red at 100%;
• the buy buttons (or, on Free and Starter, a No hosted allowance card with Get Plus — 200 pts / mo, which opens the pricing page, and the line "Hosted points start on Plus, and point packs top up that allowance.");
• the price summary and the reset rules;
• Points history: the last 50 movements of your balance, newest first, dated in UTC.

Under each entry, the line "Monthly allowance", "Purchased points" or "Monthly + purchased" says which part of the balance it touched. When nothing has happened yet: "Nothing yet — top-ups and hosted AI calls will be listed here." More on the account page itself: Account settings and your data.

Buying more points

Top-up packs add points to your balance with a one-time payment.

• Who can buy: plans with a monthly allowance, which are Plus, Ultra and Enterprise. On Free and Starter, checkout refuses with "AI point packs top up a hosted-AI allowance, and your plan doesn't include one. Upgrade to Plus (200 points / month) or Ultra (600)…"
• Where: the AI points card on your account page (100 pts — $5, 500 pts — $20), the pricing page, and the out-of-points message in Sketch's wireframe generator, which brings you back to the sketch after paying.
• After paying: you come back to the page you bought from with the points on your balance. When that is your account page, it says "100 AI points added to your balance. Thanks for the top-up." Points history lists the pack.
• Rules: "One-time payment. Purchased points never expire and are spent after your monthly allowance." Packs carry no discount codes.
• Refunds and disputes: a fully refunded pack, or a dispute decided against the payment, removes that pack's points (never taking your balance below zero).

Step-by-step buying, receipts and refunds are on Managing your subscription.

Daily limits on hosted AI

Besides your own balance, hosted AI has two daily safety limits. They protect the service for everyone; most people never meet them.

On Free, the sentence about your own key is left out, because Free cannot use one. Requests on your own key are not affected by any of these limits.

Limits when you use your own key

Requests on your own key spend no points, but they are not unlimited. Each account has a daily fair-use ceiling, counted per account (or per network address when nobody is signed in) and reset at midnight UTC:

A request refused at the ceiling is never sent to your provider, so it costs you nothing there either. Details on Bring your own AI key.

Per-feature daily caps

A few features have a daily cap of their own, on top of points. All reset at midnight UTC.

The TikZ figures are per account; Free and Starter rarely meet theirs because hosted TikZ export needs points those plans do not include (bought points excepted).

Size limits on a request

These limits stop a request before it is sent, so they never cost points.

There is also a limit on how much the AI may write back. A generated diagram may be up to 16,000 output tokens; a refinement is sized to the diagram it rewrites. When an answer would not fit, you see "output too large for this request — try a smaller diagram" and are not charged.

The REST API and the MCP server

Requests made with a flowss API key are charged to the key's owner exactly like requests from the app, and they also count against per-account daily request counters. All of an account's API keys share the same counters.

Generation through the API needs points or, for direct REST calls, your own key in the request. Over MCP, AI tools always run on hosted points. The daily generation counter is counted before points are checked, so a call refused for lack of points still uses one of the day's generations. See The REST API and The flowss MCP server.

Tips

• Use Quick for single changes (1 point) and save agent runs (8) for multi-step work. Reserve the deep specialists (47) for structural judgement on finished work.
• A pipeline costs the sum of its steps. Build → Verify → Critique → Refine costs 71 because the Critic step is a deep specialist; run Build → Review (16) when you do not need the deep critique.
• If you buy points regularly, the 500-point pack works out at 4.0¢ a point against 5.0¢ for the 100-point pack.
• Bought points are spent only after your monthly allowance, so buying early never wastes included points.
• If you already pay an AI provider, adding your own key stops points being spent at all, on every plan from Starter up.
• Check Points history whenever your balance moves unexpectedly: every charge, top-up and reversal is listed.
• Automatic repair spends a point per attempt. If your diagrams often fail while you are still writing them, switch it off in Studio Settings.

Limits and known constraints

• Hosted points are a fixed monthly allowance, never unlimited, and unused monthly points do not roll over.
• Packs can only be bought on plans with a monthly allowance (Plus, Ultra and Enterprise).
• The balance you see does not subtract points held by requests still running, so a request can be refused for lack of points while your balance reads more.
• Some features charge once the AI has answered, even if the answer is then unusable (see When you are charged).
• A streamed answer is charged when its first part reaches you, even if you close the tab before it finishes.
• A finished agent run is charged its full listed price even if it used fewer steps than expected.
• The daily safety limits on hosted AI reset at 00:00 UTC; there is no way to raise them from your account.
• Hosted TikZ export is capped per day by plan, even when you have points left.
• A Quick change in the Studio can cost 2 points when the first answer could not be applied and the prompt asked again.
• Points history shows the last 50 movements and Agent runs the last 20.

Troubleshooting

Related pages

• AI in flowss
• Bring your own AI key
• The flowss Studio Agent: prompt and quick help
• The flowss Studio Agent and its modes
• Plans and what they include
• Managing your subscription
• Account settings and your data
• The flowss MCP server