LYNS Voice 1.0.0

The first public release. LYNS Voice is the voice of Claude Code, part of the LYNS family: a holographic orb that listens, thinks, and speaks. The one you speak with is Oracle.

Install (Windows)

1. Download LYNS.Voice_1.0.0_x64-setup.exe below and run it (per-user install, no admin needed).
2. The installer is unsigned (a free, zero-cost personal tool), so Windows SmartScreen will show a warning once: click More info, then Run anyway.
3. On first use, allow microphone access when prompted, and pick your microphone in Settings if you have more than one. The speech models download once (~57 MB for recognition, ~330 MB for the neural voice) and everything runs on-device from then on.
4. The claude CLI must be installed and logged in (claude.com/claude-code); LYNS Voice drives your existing login and stores no API key.

SHA-256 (LYNS.Voice_1.0.0_x64-setup.exe):

10da7bdb9b6db4e9d6dce4e45f0b7594c698855eb8d7cca902367ea3aba904cc

Highlights

• Hands-free voice loop: local Whisper STT with auto-send-on-pause; replies stream sentence-by-sentence through a gapless synthesis pipeline.
• "Oracle" wake word: say "Oracle, " (or "hey Oracle"), fully on-device.
• Neural voice: in-process Kokoro TTS by default; native Windows SAPI as a toggle.
• The conductor: the voice session spawns and steers parallel Claude Code worker sessions, with propose-before-fan-out approval and fleet cost telemetry.
• The Library: your existing Claude Code MCP servers and hooks, inherited automatically and one-click toggleable.
• Command Center HUD: live transcript, session panels, real ConPTY terminals, a diff viewer, token/cost telemetry, themes, PiP mode.
• Security model: dontAsk allowlist (never bypassPermissions), strict CSP, checksummed model downloads. See SECURITY.md.

Windows only for now; macOS is on the way. Full details in the CHANGELOG.