ABOUT
We build AI instruments that finish.
A story that ends at bedtime. A practice that fits, so the looking stops. A pattern you can finally see. Each one does a single thing, finishes it, and then stops.
THE THESIS
Most consumer AI is inheriting the incentives of the feed. Time on device. Sessions per week. Retention as the scoreboard. Those incentives work, and for this kind of product they are a mistake: an AI that has to keep you can never tell you that you’re done, and being told you’re done is most of what you came for.
So we took the other side. A story reaches its last page. A practice fits and the searching stops. A session ends because the thing you walked in with is visible now. When somebody stops needing one of these, nothing has gone wrong.
Completion is the product. It is also the business model.
Whether people will pay for something that ends is a real question, and it answers itself fast. There is no engagement number here to hide behind. If the answer is no, we will know early.
HOW WE WORK
Every product carries a few written documents: what it is for, what it may never do, and which arguments have already been settled. The may never do part is machine-readable, and the build checks the code against it. Add a streak, a countdown, a notification that leans on guilt, or a claim we have not earned, and it fails. Not a note in code review — a red build.
This sounds like paperwork and works like the opposite. The boundary gets argued once, in daylight. After that nobody re-argues it at two in the morning against a deadline. When a rule turns out to be wrong, it is challenged in the open and changed on the record.
The models can now do far more than anyone has good uses for. That is the real constraint — not what is possible, but what is worth making — which is why the standard is written down rather than remembered.
We build narrow, ship, and shelve without much ceremony. Some products are dark on purpose. One was a music tool: the models were not good enough yet, so it sits dark, and it will stay dark until they are.
WHAT’S BUILT
Three are public. They have nothing in common except how they end.
- Storie — 有结局的睡前故事。 今夜讲完一个故事,最后道一声晚安。没有悬念,没有下一集,不用惦记着回来。
- Nou — 看清自己的同伴。 一面足够稳定、能让你看清自身模式的镜子。它映照所见,然后指引你找到能陪你走得更远的人。
- Binne — 一种练习,属于你。 找到真正适合你当下状态的静观练习——然后它就属于你,而非我们。它推荐的大多数内容,其实是别人的免费应用。
You buy one, once. Nothing renews, nothing is advertised against, nothing about you is sold. Time you spend here is not worth money to us.
WHERE THIS GOES
Our guess is that consumer AI splits in two. Companions, measured in hours — those will be very large, and for some of the people who use them they will be very bad. And instruments, measured by whether the thing you arrived with actually got sorted out.
The second kind is harder to build and much harder to finance, because when it works there is nothing to look at. The person got what they came for and left.
We would rather be early there than on time for the other one.
Software that counts itself successful when you need it less.
The longer argument for all of this is the manifesto.
IMMANENCE · A SMALL HOUSE OF INSTRUMENTS
隐私 · 条款 · hello@immanence.inc · © 2026 Immanence Inc.