Legal AI pricing in the UK usually comes in two parts, and ours is no exception. Seats are what you pay for the software. Credits are what the models cost. One credit is roughly one penny. A long question costs more than a short one, because the model reads more. An agent run costs more than a chat turn, because it does more work.
Why there is a meter at all
The models we call are billed by the token, roughly three-quarters of a word, counted on the way in and on the way out. A question with four documents attached costs the provider more than a one-line question, because the model reads all four before it answers.
We could hide that behind a flat monthly fee. Firms that use the product lightly would then subsidise firms that use it heavily, and we would have to price for the heavy end to stay solvent. That makes the product expensive for a 20-lawyer firm running a few matters, which is exactly the firm we built it for. So we pass the model cost through as credits and charge separately for the software. Prices are displayed in GBP excluding VAT and billed in USD.
What makes one question cost more than another
Three things move the number. How much the model reads is the biggest factor and the least intuitive: a short question against a matter holding thirty documents is not a short job, because the model reads a great deal before it writes anything.
Which tier you chose is the second: Fast, Balanced and Thorough run different models at different rates, and Thorough costs several times what Fast costs for the same question. Whether it is a chat turn or a run is the third: a chat turn is one exchange, while an agent run plans, searches, reads, drafts a memo, checks its own authorities and writes action items and open questions.
The plans
There are five. Free gives two seats and 500 welcome credits, which expire after thirty days, with the standard agents. Starter adds the standard agent catalogue without the review grid. Premium adds the six premium agents, Research, Contract, Due Diligence, Drafting, Audit & Risk and Matter Intelligence, along with the document review grid and custom horizon watches.
Business adds audit-log export, retention controls and priority support. Pro-BYOK lets the firm bring its own provider key, paying the provider directly for model usage. The audit and compliance register, the inspection bundle and client sharing are on every plan, not behind the top one.
What a seat buys, and what it does not
A seat is a named person with a login. It carries the software: the matters, the agents on your plan, the review grids, the audit and compliance register, client sharing. It does not carry an allowance of its own. Two firms on the same number of seats can have very different credit bills, because one runs agents daily and the other asks three questions a week.
Seats are the predictable line you can forecast a year out. Credits are the variable line you forecast from your own history, not from ours. Adding a paralegal who mostly runs review grids costs less to feed than adding a partner who runs the Contract Agent twice a day. Price the seats on headcount; price the credits on behaviour.
How to keep the bill predictable
Use the tiers deliberately: Fast for short lookups, Balanced as the default, Thorough for the hard question with many authorities. Running everything on Thorough is the single easiest way to triple a bill for no benefit. Attach the matter, so the cost lands against it and appears in cost by matter on the billing page.
Run an agent once with the right documents, because the expensive pattern is five runs with the wrong sources attached, each one nearly right; the run panel shows an estimate before you send. Know who can buy: credits are bought by an admin, and when the firm runs out, work stops until someone with the authority to spend acts. And watch the thirty-day clock on the free tier, because welcome credits expire.
What this does not do
Usage billing means a bad month is possible: a heavy due diligence exercise in one week will cost more than a quiet month, and the meter will show it. There is no per-fee-earner cap today; credits are held at firm level and drawn down by whoever is working.
The credit meter is a model-usage meter, not a time-recording tool, and it should not be presented to a client as a measure of effort. We do not do invoice-to-matter billing: cost by matter shows you where the spend fell, but it does not raise a disbursement or produce something you can put on a client bill. Prices are displayed in GBP excluding VAT and billed in USD, so the amount charged moves slightly with the exchange rate.