Prototype
Камертон
A personal app written in Metalogos that learns from a single person's data
1 of 3 done · In progress: Mini App and measurements
Screenshots and mockups
Screenshots soon
Kamerton is a personal one-person (n-of-1) experiment. Each day the program offers one practice from a library, asks what happened, and learns from the answers which practices suit this particular person. A numerical core (probabilistic choice among options) makes the decisions while a language model only words the messages. Ideas are not taken on faith but tested on the owner's own data: a dedicated "honesty valve" separates a real effect from chance. It is not a product and is not published — it has a single user.
What exists
- A daily offer of one practice from a library
- A learning selection algorithm (Thompson sampling) — learns from replies and occasionally explores
- An "honesty valve" — checks whether a practice beats the baseline instead of mistaking chance for success
- A Telegram bot interface with SQLite storage and a weekly backup
- Mini App prototypes (in progress)
- Automated CI checks against a pinned Metalogos version
Highlights
- An honest single-person test — a dedicated valve that separates effect from chance
- The whole app is written in Metalogos, exercising the language in real use
- Privacy by construction — data stays on the owner's server and camera frames are never stored
Not there yet
- An end-to-end run on a real server is not yet confirmed
- The Mini App is only partly delivered
- The two-week check that the loop works without coercion has not been passed yet
Progress
- Core and Telegram loopDone
- Mini App and measurementsIn progress
- Two weeks of the loop running without coercionPlanned
Next step: Confirm an end-to-end run on the server and start the two-week check.
Stack
- Metalogos
- JavaScript
- HTML
- Metalogos built-in server and standard library
- Telegram Bot API
- SQLite
- Self-hosted
- GitHub Actions
- Numerical selection core (no model in the loop)
- Local language model for wording