A habit is a question you save once and ask by name. It answers in typed, calibrated shapes and learns from your traffic and corrections. From its 101st decision, every answer costs half as much.
Can I archive this email without reading it?
Agents and pipelines ask the same small questions thousands of times a day. Is this spam? Did that call fail? Which team owns this ticket? A habit turns each one into something that improves with use.
Name the question, like acme.email.archive, and give its type and answers.
Same endpoint, same noul, choice and score answers, mixed freely with inline questions.
Point at any past answer with the right one. It learns your judgment, not your prompt.
After a habit's first 100 decisions, every answer costs half as much. It keeps learning on its own.
A habit's first 100 decisions cost the regular rate. From the 101st on, every decision costs half, however the habit arrives at its answer. There's nothing to tune and no budget to manage.
A typical email here is 310 input tokens: the message plus the question. Counted per habit, per account. Output is always free.
One inbox owner's archive-or-keep decisions, 125 emails the habit never trained on. Each System One model answered on its own, with no human input. The habit learned from the owner's corrections.
Until a habit is confident, a System One model answers for it, so you start at the model's accuracy and go up from there.
Habits live under your name, in collections. @latest keeps learning from your traffic. Publish it to freeze a numbered version that others can ask and fork.
@3 for a published version that never changes
worth-replyingWould a reply make something good happen, or prevent something bad?
ambiguousIs it unclear what the sender wants?
processIs this a recurring exchange that belongs in a process, not an inbox?
Run on the last 100 messages of a real inbox:
300 decisions for less than a cent.
Every habit keeps learning from an intelligent model and from your corrections. Nothing to train, tune or schedule.
Half price from the 101st decision, on every answer. No accuracy budgets, no routing rules, no surprises on the bill.
Habits answer noul, choice and score with probabilities and confidence. Existing System One code keeps working.
@latest improves as you use it. A published @3 is frozen, so anyone depending on it gets the same answers tomorrow.
Up to 256 inputs per batch, each with its own id. Every habit on the same input shares one pass.
Publish a habit for others to ask, or fork a public one and teach it your own judgment.
Pay per input token. Output is free. One rate while a habit starts, half that rate for every decision after.
| A typical email, 310 input tokens | Per million input tokens | Per 1,000 decisions |
|---|---|---|
| Cloudflare clef | $0.240 | $0.0744 |
| TypeSafe jev | $0.042 | $0.0130 |
| Cloudflare clef-flash | $0.038 | $0.0118 |
| Habits, first 100 decisions | $0.040 | $0.0124 |
| Habits, every decision after | $0.020 | $0.0062 |
Other providers' list prices for their System One models, checked October 29, 2026. Output is free on all of them.
A saved, named System One question. You define its type, wording and answers once, then ask it by name. It answers like any System One question and keeps learning from what you send and correct.
No. A new habit answers well from the first request. Corrections make it better at your particular judgment. In the results above, the habit passed every System One model after about 75 corrections.
No. Learning, accuracy and cost are handled for you. You define the question, ask it and correct it when it's wrong.
The same as a hit. After a habit's first 100 decisions, every answer costs $0.02 per million input tokens, however it was answered.
Yes. Edit @latest whenever you like; it keeps your corrections and relearns. Published versions never change, so anyone pinned to @3 gets exactly what they tested.
Ones you ask often, where the answer is in the input itself: triage, routing, tagging, "did this fail", "is this spam".
Only versions you publish. Others ask them by number and can fork them. Your @latest stays yours.
Request an account, create a habit in the console, correct a few answers, and pay half from the 101st decision.