Kairos

Kairos plans your day for you. You add tasks; it places them automatically in the free gaps of your calendar, around meetings and with a buffer after each one, in the order given by a WSJF priority score (value plus deadline criticality, divided by effort). Each task shows its score and explains it in one tap. It also protects deep-work blocks, times the real work, and alerts you when a task runs over its estimate.

Its time estimates improve by themselves. An on-device machine-learning model learns from your finished tasks how long the next ones will really take: nearest neighbours on your history, with a ridge regression as a fallback. A new task gets a duration, a range and the list of similar tasks it came from, and the model only replaces the default duration once it beats it on your own history. Training and prediction run entirely on your device.

It was made as an alternative to cloud planners: no account, no ads, no network, and your data stays on your device. Devices can sync through a file you link in a Syncthing or Nextcloud folder, with no server.

The hard parts were keeping one Kotlin Multiplatform codebase running on Android, Windows, Linux, macOS and the browser (Wasm), and making the automatic placement and the learned estimates explainable rather than a black box. The repo follows a spec-first workflow, with a spec for each domain in docs/spec.