DAWN Changelog

The native menu bar has two entries: Open Dashboard and Exit. Light mode and a cursor-first launch keep DAWN out of the way until you ask.

The Whisper engine is bundled alongside Silero voice detection; the approximately 148 MB speech model downloads on first use. Recognition runs on your Mac, and voice is user-started.

The capture-only Halo flashes what it heard and reports the result when the job is done. Questions come back by voice. Choose Follow or a draggable Pet that keeps its style and position across restarts. Point & Ask opens the microphone at the key press and keeps a short release tail.

Gain staging, polarity audit, bus routing, clipping detection, reference-track loudness matching and mix technical checks, with Production, Mix and Export session modes and sourced production guidance.

Tempo, track count and track names are answered straight from your Live set. Phrases like quieter or louder by N dB, colour a track, turn a device on or off, and sequences of up to 16 clauses run locally without a model round trip.

When DAWN needs an answer it waits for you instead of failing, and a turn that ends badly gets one recovery attempt before it reports back.

Every chat request passes through a single deterministic pipeline before the selected model is called, so exact requests are handled predictably and the model plans only what remains.

A stable prompt prefix improves caching, earlier tool results replay in compact form, and session queries return a screenful by default. Turn limits now scale with the request.

Replies read like a colleague instead of a machine receipt, with a minimal tool card in place of raw tool details. Device parameters are read and written in Live's own display units.

Correlated receipts, project-bound ownership, guarded writes and verified readback make action status clearer. Writes Live rejects are reported as not applied.

Hosted, custom-API and Apple speech recognition are retired. Voice is still user-started and never always listening.

The app uses DAWN Engine. DAWN Core is a separate open-source library and is not the shipping chat engine.

A plan can address every track it creates, stopped turns no longer complete after cancellation, and Retry no longer resumes a turn that has no journal.

Whisper non-speech output stays out of commands, live partials no longer delay the final transcript, and mishearings of the word track are repaired. Phrases such as the Lead track name the track Lead, and everyday device words resolve to Live stock devices.

Tempo compensation now settles correctly after a change.

Analytics privacy choices are enforced, usage metrics are anonymous aggregates with no identifiers, local provider addresses are validated before probing, and key dependencies are upgraded to fixed releases.