Why another habit tracker

The app usually breaks before the habit does

Ask someone why they stopped using their habit tracker and you rarely hear that they stopped caring about the habit. You hear that they missed a few days and the streak went to zero, or that logging turned out to be more work than the thing being logged, or that they opened it one day, looked at a grid of squares, and couldn't think of a single reason to keep going.

All three are design problems, and they have the same root: the tracker can only hold one shape of data — did you do the thing, yes or no — and almost nothing people actually want to change fits that shape.

Three things a checkbox can't hold

The reason. You skipped because you were travelling. That fact, recorded eleven times, is the most useful thing in your entire history, and there is nowhere to put it. So it's lost, every time, and six months later nobody can tell you why your third week is always the one that goes.

The near miss. You wanted to and didn't. In a break habit that's the single most important event there is — it's the one that contains the trigger — and in most trackers it doesn't exist, because nothing happened.

The number. "Four alcohol-free days a week" is a real goal with a real target. A daily yes/no can't represent it, so the app makes you choose between pretending your goal is daily abstinence and not tracking it at all.

Why the interface is a conversation

Because all three of those are trivially easy to say and genuinely annoying to enter. "Skipped Tuesday, flight" is four words. As a form it's a date picker, a reason field somebody has to have designed, and a taxonomy of reasons that will be wrong for you. Multiply by every habit and every interesting day, and the reason nobody captures context is obvious: the capture cost more than the insight was worth.

The same asymmetry runs the other way. Months of check-ins contain real answers about how you work — which conditions predict the sessions you complete, whether one habit is quietly propping up another — and reading them properly has always meant exporting a CSV. Asking a question in a sentence is not a gimmick here. It is the only thing that makes the data you've been generating for years worth anything.

What we won't do with a language model

It never produces a number. Streaks, rates, totals and comparisons are computed from your check-ins by ordinary code, and the model's job is to decide which operations to run and what to pass them. A tracker that approximates your progress is worse than useless — it's a tracker that can flatter you — so the arithmetic is kept somewhere it can't be improvised.

It also doesn't coach. No streak guilt, no manufactured urgency, no notification designed around retention rather than around you. The nudges exist to catch a gap at day three, when it's still recoverable, and that's the whole ambition.

And it's not a walled garden

Everything the app can do is exposed over MCP, so the assistant you already have open can check you in and answer questions about your history. Partly that's about friction — the lowest-friction surface for you probably isn't ours, and pretending otherwise loses to the sofa. Mostly it's that an account you can only reach through one vendor's chat window isn't really yours. Every plan exports everything, always.

Where it is now

Early. The pages on this site describe what's being built, the waitlist is how you get in first, and founding accounts lock their price permanently. If you've abandoned three habit apps already, you're exactly the person worth hearing from — tell us where it broke down.