Skip to content

YazSes v2.0.0-dev.3 — Voice-First Interaction Layer (developer preview 3)

Pre-release · not published to PyPI/Snap/APT (publish workflows are pinned to the v0.x/v1.x lines; v2 dev tags intentionally do not ship). v1.4.1 remains the stable release. Every v2 feature is OFF by default.

dev.3 completes all 13 flagship + experimental features across Waves A, B, and C — the full v2 vision is now implemented, tested, and integrated. Still 100% on-device and privacy-first (ADR-011); 13 v2 ADRs (adr-v2-000…adr-v2-013).

Wave A — daemon-wired dictation upgrades

Confidence Ink (ADR-v2-001), Prosody pause→sentence (ADR-v2-002), Spoken Edit Mode (ADR-v2-003), Context-Primed Dictation (ADR-v2-004).

Wave B — new capabilities

  • Personal Adapter P1 (ADR-v2-009) — corpus prompt-mining biases STT toward your jargon.
  • Spoken Recall & Ambient Scratch (ADR-v2-005) — yazses recall / yazses scratch.
  • True Code-Switch (ADR-v2-008) — routing layer, dormant until a CS adapter is supplied.
  • Voice-to-Tool / Spoken MCP (ADR-v2-006) — voice→tool planner + confirm guard.
  • AT-SPI Voice Pilot (ADR-v2-007) — "click Save"-style desktop control.

Wave C — experimental (all --force to enable)

  • Modality Role Router (ADR-v2-011) — assign each input its fastest role.
  • Accessibility Continuum (ADR-v2-012) — Whisper/Low-Effort Mode lowers the mic gate for quiet speech; semantic-capture seam.
  • Gaze-Routed Dictation (ADR-v2-010) — route dictation to the looked-at window, with a focus fallback + destructive-confirm.
  • Glasses↔Desktop Bridge (ADR-v2-013) — pair a phone/glasses as a mic; the desktop runs STT + injection (reuses the remote transport).

Manage everything with yazses features

27 capabilities registered — all v2 ones off by default; experimental ones refuse enable without --force. New config sections: [confidence] [context] [recall] [agent] [pilot] [modality] [continuum] [bridge], plus commands.spoken_edit, prosody.pause_sentence_ms, and gaze.route_dictation.

Quality

876 tests green across Linux × macOS × Windows on Python 3.11 and 3.12. Each feature is guarded and off by default, so the v1 dictation path is unchanged when everything is off.

What's next (toward v2.0.0)

UX depth on the highest-value features (Confidence overlay markers + voice re-pick; Spoken Edit destructive-confirm; live backends for the experimental Wave C set) and a docs pass. v2.0.0 ships to all channels once the publish workflows are extended to the v2.x line.