Cafeina

Cafeina is a local Windows keep-awake utility for long-running jobs (SQL, loads, downloads) — tray-resident, optional start-with-Windows, counts input events (clicks, keys, mouse travel) as totals only. No network, no telemetry. But nested inside it lives a second project: an opt-in, consent-gated collector of behavioral biometrics — keyboard and mouse — for research on human typing and navigation patterns.

What makes it interesting is the governance effort: LGPD (Brazil's GDPR-equivalent) taken seriously on a personal project. Affirmative consent before any local collection (TaskDialog with durable evidence in consent-log.jsonl), a random research identity (128 bits, not derived from the machine ID), a separate deletion secret that lets a participant erase all their cloud data on their own (DELETE /api/v1/user), a documented data-processing impact assessment (RIPD), and a manual user-triggered sync gated by 4 distinct consent checkboxes. Technical invariant: no ordered key sequence longer than 3, no per-event timestamps, never the typed text — enforced at write time, not as a promise.

Two pieces. A Windows client in .NET 8 WinForms (C#), with low-level hooks for timing + Raw Input to attribute events to the correct physical keyboard (device change splits the session), custom-drawn controls, runtime i18n pt-BR/en-US, and local SQLite for aggregates only (bigrams/trigrams as histograms — never text). Packaged as a self-contained single-file exe (win-x64, ~157 MB) downloadable directly from the site. A server in Fastify + better-sqlite3 in a Docker container on the VPS, behind nginx + Let's Encrypt, with an idempotent upload endpoint, watermark, delete-all-by-key, rate limits, and a static homepage that doubles as the LGPD privacy notice.