DailyBrief

DailyBrief watches which folders you actually worked in - directory paths, file names and timestamps only, never file contents - and every morning writes you a one-page briefing in Georgian: what got done, what is left, how to split the day. It carries a Trello-style kanban board that both you and the AI edit, re-checks tasks against real git activity every few hours, and shows the token cost of every run with a kill switch.

Task boards go stale because nobody updates them, and AI assistants go expensive because they read everything. So the pipeline was inverted: PowerShell scripts do the gathering and compress it into a small digest, the model reads only that and returns a tiny patch, scripts apply it. When nothing happened, the model is never invoked at all.

01

It reads activity, not content - directory paths, file names and timestamps only, and nothing leaves your machine except the compressed digest.

02

Drop a folder onto the app and it becomes a task; a git-aware scanner baselines it and tracks commits, changed files and dirty state.

03

Automatic checks every 5 hours append what actually changed to each card, and close a task when the evidence is clear.

04

Token-diet by design: idle checks short-circuit to zero tokens, Sonnet only for the brief and day replans, Haiku for mechanical runs.

05

Every run is journaled with duration, tokens and cost, and future runs show a history-based ETA before you start them.

06

You can open the exact context that was sent to the model as a TXT - nothing about the pipeline is hidden.

07

Deleted tasks are archived and never resurrected by the AI.