Per-Seat or Usage-Based: AI Agent Pricing

Per-seat pricing charges for the right to have an account. Usage-based pricing charges for work actually done. The difference shows up in who on your team feels free to try something.

Max Ma
Max Ma
Sep 6, 2026 ·

Per-seat AI pricing charges for the right to have an account. Usage-based AI pricing charges for the work actually done. The difference rarely shows up first in the invoice total — it shows up in who on your team feels free to try something.

Updated November 18, 2026, for a pricing change: on your own provider key, Fours now charges a platform fee of 10% of the supplier’s price for those tokens. AI helpers that work for you, such as knowledge-base indexing and search, are now billed like any other AI. The monthly free credit is $100 for each month through November 2026 and $50 from the December 2026 grant on. This post also now says that at a zero balance new sandboxes and uploads stop only past the account’s debt limit.

Corrected September 27, 2026: an earlier version said each hosted model’s rate is shown in the console before you run it. The model a turn runs on can change, so there is no single rate to show before a run.


Most software a business buys is priced per seat, so that is the shape buyers expect when they evaluate an AI platform. It is a familiar, predictable model, and for tools where every user does roughly the same amount of work, it is a fair one.

AI agents are not that kind of tool. The person who runs one report a month and the person whose agents process every inbound support case are doing wildly different amounts of work behind the same login. This post covers what each pricing model actually charges for, what Insulin meters and what it deliberately does not, and how the choice changes adoption behaviour on a team.

What per-seat pricing charges for

Per-seat pricing is a licence fee for access — a fixed amount per named user per month, whether that person runs one query or ten thousand. The bill is a function of headcount, not of work.

That predictability is the model’s real strength: finance can forecast it a year out from a headcount plan. But it has a consistent side effect on adoption, and it is worth naming before you compare price sheets. When access costs money, every new user becomes a purchase decision. Three things tend to follow:

  • Occasional users never get added. The analyst who would benefit twice a quarter cannot justify a full seat, so the tool never reaches the edges of the organization where a lot of the repetitive work actually sits.
  • Logins get shared. When a seat is scarce, people route requests through whoever has one. That breaks attribution, and it breaks permissions.
  • Renewals become seat audits. The annual conversation is about trimming inactive users rather than about what the tool produced.

None of that is an argument that per-seat pricing is wrong. It is an argument that the pricing model is also an adoption policy, and it is easy to buy one without noticing you bought the other.

What usage-based pricing charges for

Usage-based pricing is a charge for resources consumed, metered in the unit each resource is naturally measured in. Insulin’s documentation is explicit that “there is no seat count, no tier and no subscription” and that “an organization that uses nothing pays nothing.”

Insulin meters six resources:

ResourceMeasured in
AI input tokenstokens
AI output tokenstokens
Sandbox processorprocessor-seconds
Sandbox memoryGB-seconds
Sandbox storageGB-seconds
Object storageMiB-months

Two ways to pay for the models themselves. You can use Insulin’s hosted models, billed per token at the supplier’s published price for each model plus a 50% margin, or connect your own provider account, in which case that provider bills you for the tokens under your own contract and Fours charges a platform fee of 10% of the supplier’s price for those tokens, with no margin. Sandboxes and stored files are billed the same way with either. The pricing page explains how the rates are set; Fours gives the current rates on request.

The adoption consequence is the mirror image of per-seat. Adding a colleague costs nothing until they do something, so the tool spreads to the people with occasional work instead of stopping at the people who could justify a licence.

Does metering punish exploration?

This is the fair objection to usage-based pricing, and it deserves a direct answer rather than reassurance. If every click costs, people click less — and a platform nobody explores produces nothing.

What is not metered matters as much as what is. Insulin’s documentation lists three things that are never billed to you:

  • AI Fours runs on its own behalf. A model call Fours makes for itself rather than for you, such as rewriting a marketplace’s error message on a page you merely opened, is Fours’ cost, not yours.
  • Reading what you already have. Opening, listing or searching your stored files, and starting, stopping or connecting to an existing sandbox.
  • Database storage. It is not metered.

The line is drawn at who the AI works for. The helpers that work for you are billed like any other AI, each under its own name on the Agent tab of the usage breakdown: knowledge-base indexing and each search, thread titles, AI text generation, skill drafting, Inbox, PRM and deal registration. So are the AI features elsewhere in the Fours console, such as co-sell scoring and AI field mapping.

So reopening past work and opening or searching your stored files are free actions. Searching a knowledge base is not, because each search makes an embedding request. What costs money is generating tokens, for you or for a helper working for you, and running new compute: the work itself.

On top of that, every organization gets free credit each month: $100 for each month through November 2026, and $50 from the December 2026 grant on. An organization with a different amount arranged with Fours keeps it, and the Credit table in Settings → Billing shows your organization’s figure. No payment method is needed to spend it, so a team can try agents on a real workflow before anyone adds a card.

What happens when the credit runs out

A spend limit should pause work, not destroy it — and this is where usage-based platforms differ most from one another, so it is worth checking on any vendor you evaluate.

On pay as you go, at a zero balance in Insulin, AI requests pause, requests on your own provider keys included, because the platform fee is still owed. What keeps working is everything that reads what you already have: opening, listing and searching stored files, and any sandbox already running. Creating new sandboxes and uploading new files keep working too, until the balance passes the debt limit Fours sets for the account. The documentation is unambiguous about the guarantee — “No data is deleted in any of these states. A running sandbox is never killed to collect a debt.” Adding credit resumes service about a minute later, with no manual step.

That distinction is the one to test in a trial. “What stops” is a pricing question; “what is destroyed” is a data question, and only one of them is reversible.

Choosing between them

Neither model is universally better. They optimize for different things, and the right question is which failure mode you would rather have.

Per-seatUsage-based
Bill tracksheadcountwork done
Easy to forecastyes, from a headcount planneeds a spend limit and a review cadence
Cost of adding an occasional usera full seatnothing until they use it
Heavy automationeffectively free once seats are boughtpriced proportionally
Main riskpaying for inactive seatsan unwatched job running up consumption

Per-seat suits a team where usage is uniform and budget predictability outranks reach. Usage-based suits a team that wants the tool available to everyone and is willing to manage consumption — which is a real obligation, not a footnote. Set a monthly spend limit, and read enterprise AI spend governance for how to attribute cost by user, app and model once more than a couple of jobs are running.

Frequently asked questions

Does Insulin charge per seat? No. The billing documentation states there is no seat count, no tier and no subscription. Billing is consumption-based, so an organization that uses nothing pays nothing, and adding a colleague costs nothing until they do work.

What exactly does Insulin meter? Six resources: AI input tokens and AI output tokens, measured in tokens; sandbox processor, in processor-seconds; sandbox memory and sandbox storage, in GB-seconds; and object storage, in MiB-months.

Is searching my own documents billed? Searching your stored files is not: opening, listing or searching them, and starting, stopping or connecting to an existing sandbox, are not metered. A knowledge-base search is different: each search, like indexing a document, makes an embedding request, billed like any other AI.

What happens if the credit runs out? AI requests pause, your own keys included. Stored files stay readable and searchable, running sandboxes keep running, and new sandboxes and uploads stop only past the account’s debt limit. No data is deleted, and service resumes about a minute after credit is added.

Can I use my own model provider? Yes. Your provider bills you for the tokens under your own contract, and Fours charges a platform fee of 10% of the supplier’s price for those tokens, with no margin. On hosted models, each model is priced at its supplier’s published price plus a 50% margin.

Takeaways

  • A pricing model is also an adoption policy. Per-seat pricing makes every additional user a purchase decision; usage-based pricing makes the work itself the thing you pay for.
  • Check what is not metered. In Insulin, reads of existing work, database storage and AI Fours runs on its own behalf are not billed; AI helpers that work for you, such as knowledge-base indexing and search, are. Knowing which is which is what lets a team explore without surprise costs.
  • Test what happens at zero. Pausing AI requests is a pricing behaviour; deleting data would be a different thing entirely. Insulin deletes nothing and never kills a running sandbox to collect a debt.
  • Usage pricing comes with an obligation. Set a spend limit and review consumption on a cadence, or proportional cost becomes surprising cost.

Insulin is priced for the work, not the roster. See the pricing page for how the rates are set, or book a demo to walk through what a month of your workload would actually meter.

Sources

Primary sources for the platform rules cited above. Last verified November 18, 2026. Providers change their products, prices and terms without notice — check the source before relying on a detail.

  • Fours Insulin documentation: Billing — Establishes that billing is consumption-based with no seat count, no tier and no subscription, so an organization that uses nothing pays nothing, and that no payment method is needed to spend the monthly free credit.
  • Fours Insulin documentation: How billing works — Credit: free credit granted on the 1st of each month, $100 for each month through November 2026 and $50 from the December 2026 grant on; an account with a different recurring amount arranged with Fours keeps it, and the Credit table shows the real figure. What is not billed: AI Fours runs on its own behalf, and database storage.
  • Fours Insulin documentation: What is metered — The six metered resources and their units; How prices are set (the supplier's cost plus a 50% margin on model calls Fours supplies the key for); Using your own provider keys (BYOK): a platform fee of 10% of the supplier's price for those tokens, with no margin, and a zero balance still pausing BYOK requests; What is never billed: the helpers and console AI features billed under their own names on the Agent tab, and the not-billed list (AI Fours runs on its own behalf, reading what you already have, database storage).
  • Fours Insulin documentation: When AI pauses — What keeps working with credit exhausted and past the debt limit (new sandboxes and uploads stop only past it), that no data is deleted, and that service resumes about a minute after credit lands.
  • Fours Insulin documentation: Knowledge Bases — Embedding is billed like any other AI: each embedding request a knowledge base makes, indexing a document and running a search, is charged to the organization's AI credit.

Browse every post on the Insulin Blog

Stay Updated

New posts, product updates and marketplace strategy are shared on LinkedIn as they publish.

Follow Fours on LinkedIn