Fono

Hands-free wake-word activation: idle, listen for a spoken phrase, and
start dictation or the assistant with no key — locally, on the ONNX
runtime already in the binary, and auto-served over Wyoming for Home
Assistant.

Added

• Optional always-on wake-word activation ([wakeword], default
  off). Fono can idle and listen for a spoken wake phrase, then start
  dictation or the assistant on the same path the hotkey uses — no key,
  no hands. Detection runs locally on the ONNX runtime already in the
  binary via openWakeWord,
  so it adds no new dependency and no measurable size, and your audio
  never leaves the machine while idle on the default path. The listener
  suspends during any active recording or assistant turn and resumes
  when Fono goes idle. Ships with a clean Apache-2.0 default phrase as
  the only enabled model, plus an opt-in community phrase catalog that
  is downloaded on demand, never bundled, and shows its NonCommercial
  license as a notice when you pick one. When the LAN Wyoming server is
  enabled, Fono automatically serves wake detection over it — exactly
  like it serves STT and TTS, with no extra switch — so Home Assistant
  discovers Fono as a drop-in wake-word provider and detection runs on the
  Fono box with audio staying on the machine. Opt-in and off by default,
  behind an explicit "idle mic audio leaves the machine over the LAN"
  warning, Fono can instead forward audio to an external
  wyoming-openwakeword service. fono doctor reports the wake-word
  configuration and that privacy warning. The clean-license hey_fono
  default model is not yet hosted, so the local always-on listener stays
  off until you enable it; the auto-served Wyoming path uses the community
  hey_jarvis model as a temporary fetchable default in the meantime.
  Engine and licensing rationale is in ADR 0012.

Full Changelog: https://github.com/bogdanr/fono/compare/v0.11.1...v0.12.0