More doing.
Pick your pace.

Pro and Max are subscriptions, not one-off purchases, and the most useful thing a pricing page can do is say so before you buy rather than after you are charged.

• When. Every month, on the same day of the month you first subscribed. Your exact next date is shown in the app beside your plan, and on Stripe's billing page.
• How much. The same price you signed up at — $20 a month for Pro, $100 a month for Max — charged to the card on file. If we ever change what a plan costs, that is not something we would do to an existing subscription quietly.
• Until when. Until you cancel. There is no fixed term, no minimum number of months, and no cancellation fee.
• How to stop it. One button: Settings → Manage billing in the app, which opens Stripe's own page. It takes about as long as signing up did. The billing page walks through it.
• What cancelling does not do. It does not cut the month short. You keep the plan to the end of the period you have already paid for, then it drops to Free on its own.

Top-up credit is not a subscription. It is a single charge, it does not repeat, and there is nothing recurring to cancel.

A team is not a bigger plan, and there is no such thing as a team allowance. It is a number of Pro accounts paid for together — one subscription, one card, one invoice, at $20 per seat, per month, for however many seats you have. Ten seats is $200 a month; fifty is $1,000 a month. Those two are named on the card above because they are the sizes people ask for by name, not because they are the plans — 25 seats is $500 a month, by the same multiplication, and there is no size the price does not work at. Seats can be added or given up later and Stripe prorates the difference, though never below the number of people actually on the team: the seat has to go after the person, not before them.

Every seat is a full Pro seat, metered per person. The same 150 agent tasks a month, the same further ceiling for conversation behind them, the same daily spending limit — one copy of all of it each, counted against the account of the person using it. A team of ten is ten separate meters, not one meter ten times the size. Nothing is pooled, which is the question every buyer asks second: one person's heavy week cannot leave the other nine with nothing. It cuts the other way too, and we would rather say so — a quiet seat does not hand its unused tasks to anybody else.

What the owner sees, and what nobody sees. One owner manages the seats and the billing; everybody else just uses Fetchi. What the owner gets is two things: who holds a seat, and what each of those people has cost this month, in dollars — a list we run and send back by hand today, since the screen for it is exactly the thing that does not exist yet. What is on the list will not change when the screen arrives. What anyone asked Fetchi, said to it, or had it do is not on that list — not summarised, not sampled, not available on request. That is not a setting we left switched off: the proxy's database has no column a prompt, a transcript or a screenshot could go into, so there would be nothing for an owner's screen to read even if one wanted to. An assistant your employer can read over your shoulder is a different product, and it is not the one being sold here. The privacy page shows the schema.

How a team starts, today: with an email, not a button. The seats, the per-seat subscription and the invite codes are built and running on our server. What does not exist yet is the screen that opens them: no part of the Mac app and no page on this website touches any of it, so for now a team is set up by hand at our end. Write to riyaan.aldimassi+fetchi@gmail.com saying how many seats you want, and we will set the team up with you and hand the owner the first invite code. It renews the way Pro and Max do, at $20 a seat, until it is cancelled — everything in the renewal section above applies to a team unchanged. The trial does not: a seat starts paid. Free's $2.50 and its 14 days belong to a free individual account, and a seat has never been one.

It is the one number that decides which plan you need, so it is worth being exact about. An agent task is a job where Fetchi picks up a tool or takes the mouse — reading a page to answer from it, driving an app, filling a form, working through a list. Asking it something, dictating into a document, or just talking to it is not one, and never has been.

The count resets on the 1st on a paid plan. On Free it does not save you: the trial's $2.50 and its 14 days are one-time bounds sitting in front of the monthly counter, and reaching either one ends the account's free run whatever the counter says. Behind the count sit two spending ceilings on the same lane — one for the month and a smaller one for the day — so a run of unusually heavy jobs can stop the hands before the counter has reached its number, and an extremely heavy single day can stop them until tomorrow with plenty of month left. Whichever of the three runs out first is what stops. Fetchi says which one it was, and whether it resets tomorrow or on the 1st.

This is the part worth reading twice, because it is where a plan turns out to be more or less generous than it looked.

On Pro and Max there are two ceilings on the same meter. Agent work — screen control, a tool loop that has already acted, the realtime voice lane — stops at the first one. Plain conversation keeps going to the second, further one. So when your agent tasks are gone, Fetchi does not go silent: it stops doing, not talking. That was deliberate. Gating both on one number meant somebody who ran out got refused on "what time is it" as flatly as on "book me a table", and an assistant that has gone quiet is an assistant you stop reaching for.

On Free there is one ceiling covering both. Conversation and agent work draw on the same allowance, so when it is spent it is spent — the voice does not carry on past it. And on Free the allowance is not the only thing that can go: the trial's $2.50 of usage and its 14 days sit in front of it, and neither comes back on the 1st. We would rather write that here than let you find it out on the 20th. Talk that outlives the agent cap is most of what a paid plan buys, and giving it away on Free would be giving away the reason to pay for one.

One more thing that surprises people: the realtime voice lane is metered as agent work, not as conversation. It is billed by the minute rather than by the turn, and a minute of it costs about what thirty typed turns do. When the agent budget is running low Fetchi drops back to the ordinary voice lane instead of spending what is left on it — quieter, and still talking.

Running out used to have exactly one cure, which was waiting for the 1st. Credit is the other one. It is a one-off payment, not a subscription — someone who needs a few dollars more this week must not end up signed up to anything.

It comes in three sizes: $5, $13 or $28 of credit, bought here or with the Add credit button in the app. The price is not the credit — $5 of credit costs $10, $13 costs $25, $28 costs $50 — and that gap is deliberate rather than accidental. At parity a refill loses money once Stripe has taken its cut, and it would undercut the plan it exists to protect. Stripe shows the exact price before you pay.

Already using Fetchi? Use Add credit in the app so the refill lands on that account. A website purchase needs the setup code from the return page; the public 0.2.0 download cannot accept that code yet. Read the setup steps before buying.

If it is the count of agent tasks that ran out rather than the spending ceiling, credit is not the fix and the plan is. That distinction is easy to get wrong at the moment you're stuck, which is the moment it costs you $10.

The plan and credit buttons on this page open Stripe Checkout, which collects your email address and payment. After paying, keep the setup code on the return page and follow the setup steps. A code proves which checkout is yours; an email address alone does not.

If you already use Fetchi, pay inside the Mac app to connect the purchase directly to your account. There are three controls under Settings:

• See Pro and Max, then pick the plan that fits — the plan screen. It is not labelled Upgrade, because a one-word verb that sounds like a purchase tells you nothing about what the click does.
• Add credit — a one-off refill, the packs above.
• Manage billing — cancelling, changing the card, downloading invoices. It opens Stripe's own page for your account. Exactly where that button is.

None of the three is a team. There is no seat to pick on that screen and no team to manage from it, which is why a team starts with an email instead.

Every one of them opens Stripe. We never see a card number, and there is nowhere in Fetchi that one could be typed. Before the app offers you a plan at all it asks the service whether payments are switched on, and says so plainly if they are not, rather than opening a shop with nothing in it.

When you do pay, the plan lands when Stripe confirms the payment to our server rather than when the browser comes back — usually seconds, occasionally a minute or two — and Fetchi says so out loud when it happens. If it hasn't after ten minutes, that is worth an email rather than a second attempt. What happens after checkout sets out the whole order, including why a top-up lands quietly instead.

Then you changed your mind, and there is nothing to undo: a checkout you backed out of charges nothing and moves nothing. You go back to Free, which is a trial and not a plan — no card, 15 agent tasks a month while it lasts, and an end date: $2.50 of usage or 14 days from signup, whichever arrives first. It is long enough to find out whether this is for you, and it is not long enough to be anybody's permanent arrangement.

If you are already paying and want to stop, that is one button in the app, and this page says where it is. Cancelling stops the renewal, not the month you have already paid for — the plan keeps working to the end of that billing period and drops to Free when it ends, which is Stripe's own doing rather than something we schedule. Your invoices stay downloadable afterwards.

Anything a page can't answer — including whether a plan is worth it for what you actually do — goes to riyaan.aldimassi+fetchi@gmail.com, where a person reads it.