Bring your own key: what an AI dev teammate actually costs

Per-seat pricing for a tool that works per repository never made sense to us, and a markup on model tokens hides the one number you most need to see. So quellbot has three bills, and only one of them is ours.

Bill one: the model, on your own account

quellbot runs on a credential you bring. On the Claude lane that is either a Claude subscription token from claude setup-token or an Anthropic API key. On the Codex lane it is a ChatGPT subscription, connected once through a device-code approval in the browser, or an OpenAI API key. Model spend goes straight to that provider at their prices. quellbot never marks it up, and never sees the invoice.

The two kinds of credential cost differently, and the console treats them differently:

• Subscription token. Flat-rate, covered by the plan you already pay for. quellbot still meters every run and shows the figure as subscription-covered, so you can see what a run would have cost on a key.
• API key. Metered per token. Every run reports its exact cost on its card and its run page the moment it finishes.

How much is a run? It depends on the model and the size of the change, which is exactly why the number is on every card rather than in a table here. Small, well-scoped fixes on a mid-tier model usually land at a fraction of a dollar; a review of a large diff on a top-tier model can cost more than a fix. The run page tells you, per run, and the Usage page adds it up per month.

Bill two: about a minute of compute, on your Fly account

Every run, whether a fix, a review, a CI triage, or a reply, happens in a throwaway Fly Machine created in your own Fly.io account. Fly bills per second, and a typical run lives about a minute, so this line is small. It also scales to zero: no runs, no machines, no cost. Fly does require a card on the account before it will create machines at all, even on the free allowance.

Bill three: quellbot

Public repositories: $0. Every public repository you own or contribute to, unlimited, with every feature, forever. No trial, no seat count.

Private repositories: Pro, $29 a month or $290 a year (the yearly plan bills ten months). One plan covers unlimited private repositories, personal and organisation, and unlimited teammates. Pro also comes with feedback answered within 24 hours.

Cancel any time from the console and keep access to the end of the paid period. If it is not for you, there is a full refund within 14 days of your first payment. Payments are handled by Dodo Payments as merchant of record.

Budgets that match how you pay

A tool that spends your money on its own schedule needs a ceiling you set, and the ceiling has to match the unit you pay in.

• On an API key, the model gateway debits every call atomically against a per-run safety limit, so a single runaway run cannot spend without end. You set one number, the monthly cap, and it covers every path that spends tokens: fixes, builds, reviews, CI triage, and replies. Cross it and quellbot holds new work until the cap resets or you raise it.
• On a subscription, dollars are not the unit, so there is no dollar cap. A runaway guard caps model calls per run instead.

When something needs you, quellbot raises an alert in the console and, for the two that are actionable, sends one throttled email: your credential stopped working, or your monthly cap was reached. For a hard dollar ceiling across everything, set a spend limit with your provider as well. The monthly cap is a soft stop, and we say so.

Why there is no per-seat price

A review is about the pull request, not about how many people can read it. A CI triage comment is about the failed run. Charging per teammate would mean a team pays more for the same review because more people benefit from it, which is backwards. So the plan is flat, and adding a teammate costs nothing.

All three, on one table