Skip to content

ADR-v2-093 — Local Voice Timer & Break Reminder

Status: Accepted (2026-07-02) · Wave K Context links: [[adr-v2-092-word-count-goal-tracker]] (session utilities), [[project_readback_loop]] (spoken alert), [[adr-011]]

Context

Wave K research (#10) — set a timer or break reminder by voice ("set a timer for 25 minutes", "remind me in an hour") entirely on-device, announced by read-back — no phone, no cloud assistant, no context switch away from the keyboard. A Pomodoro/RSI-break accessibility aid for long dictation sessions. Every commercial voice timer (Alexa/Siri/Google) is cloud-tied; a fully-offline one that lives in your dictation daemon is genuinely unavailable elsewhere. Anchor: on-device assistant / eyes-free reminders line of research; ties to the ergonomic-break literature.

Decision

Add an opt-in Local Voice Timer: [voicetimer] enabled=false. Pure cores: parse_duration(text) → seconds (supports "25 minutes", "2 hours 30 minutes", "an hour", "half an hour"), format_duration(seconds) → a spoken-ready string, and parse_timer_command(text) → ("set", seconds) / ("cancel", None). Dependency-free parsing; the daemon schedules the alert on a background timer and the spoken alert reuses the existing TTS. OFF by default.

Consequences

  • Fully-offline voice timers/break reminders inside the dictation daemon.
  • Pure duration parsing → fully testable; scheduling + TTS reuse existing infra.
  • Privacy (ADR-011): local timer only; nothing scheduled in any cloud.
  • Caveat: fires only while the daemon runs (session-scoped, not a persistent alarm); off by default.