PRICING · BRING YOUR OWN MODEL

Priced in loops.

You pay for a verdict on your output, not for retries, and never for tokens: you bring your own model and we never proxy it. The price depends on what surface you check, and the calculator is public, so you can reproduce any bill yourself.

The numbers

100 free loops every month.

No card to start, and 1,000 a month once you sign in with GitHub. When you need more, the paid plans below are monthly, the way Anthropic prices Claude, and the card is only asked for when you upgrade.

Get a key → Try it free first

The plans

One loop underneath all four plans. You pay less when you pool the lesson back, more when you keep things sealed.

start here

Free

$0

100 loops/month anonymous, 1,000 with a GitHub sign-in. Held to the world's conventions (the global pattern cache). You contribute the abstracted lesson from each loop back to the public donkeys (a donkey is a deterministic check that runs against your output), never the data. That is how you pay, and it is required on Free.

the default

Pro

$20 / month

10,000 loops/month. Adds auto-consistency: bring your own repo and the donkey holds your code to its own conventions. Pooling is on by default; you can turn it off.

sealed / regulated

Max

$200 / month

1,000,000 loops/month. Adds a hermetic, sealed mode with no egress; pooling is off by default, for regulated use.

teams

Enterprise

contact us

Custom, uncapped loops; an on-prem or sealed deployment; a service-level agreement, dedicated support, and a private ratchet (your own store of accrued conventions, kept separate from every other tenant). Contact us.

Why a loop, and why you can check the bill

A loop is one judged atom. Not a token (those are your model's, and your cost, not ours), and not a retry (you are not billed for the donkey handing work back). Because we never proxy your tokens, there is no margin for us to take on your model spend.

Each surface meters the atom differently. For the prose, document, and conversation donkeys it is one check: one loop per check, so $20 buys 10,000 checks. The code donkey (the deterministic check for source code) meters by what it actually reads, function by feature, which is the unit the calculator computes; its rate is being finalized. The calculator is open source, so anyone can reproduce a bill: the metered units times the published rate, both checkable.

The loop calculator (loops.py) → The public repo →

WYSIWYD, the standalone extraction product, is a separate wedge with its own per-page pricing (Free / Pro / Enterprise) at wysiwyd.doloop.io; it is not metered in loops.

What a missed one costs: regretmath

The other side of the bill. In a cross-release study of real codebases, a convention break that ships sits a median of ~3,047 commits of the repository's own history before anyone goes back and fixes it (an immutable-default slip ~3,697, an unhandled error ~2,426). The counter is public, REGRETMATH.md, the same way the loop calculator is. Caught at the commit, that is regret you never pay.

In your own terms: saved = bugs_caught × hours_per_bug × your_loaded_rate. The loop is what you pay; the commits of delay are what you avoid. Both numbers are public and reproducible.