Changelog

What has shipped, and what has not.

NovaHUD is at version 1.0. This page lists released builds only — nothing is listed here before it is downloadable, and the open items at the bottom are stated as open rather than as coming soon.

  • 1.0 September 2026
    Current build

    The first public release. Everything below is in it.

    • A floating, always-on-top card for every running Claude Code session, above every other window including the Claude Code app itself, permanently bound to its session.
    • Address a session by name, by nickname, by its session title or by its orb colour — or say "everyone" to reach all of them at once.
    • A distinct human voice and a distinct orb colour per session, assigned once and kept.
    • Three listening modes per session: Read aloud, Summary and Quiet.
    • Barge-in: talk over a reply and it stops instantly, mid-word. No stop command, no wake word.
    • Four widget sizes — tiny, small, medium and large — with a default you choose. Smaller sizes drop information rather than shrinking text.
    • Dictation into any app with ⌘⇧D, mute with ⌘⇧M, and ⌘⇧1⌘⇧9 to make the Nth card live.
    • The whole voice pipeline on-device: whisper.cpp for speech recognition, Kokoro-82M on the Apple Neural Engine for the voices, Llama 3.2 3B via Ollama for routing and summaries.
    • A three-day free trial with no credit card, then $5 a month ($3 with a referral code).

    Download 1.0 — DMG, macOS 14 or later, Apple silicon.

Known and open

Stated plainly, because finding it out at install time is worse.

Not notarised yet

The build is signed with an Apple Developer ID certificate but has not been through Apple's notarisation service, so Gatekeeper blocks it on a double-click. Right-click (or Control-click) NovaHUD in Applications and choose Open for the first launch only.

Notarisation is planned. The download page will stop saying this when it is done.

Not on the Mac App Store

NovaHUD is a direct DMG download from this site. A Mac App Store release is planned but has no date.

Apple silicon only

There is no Intel build, and no Windows or Linux version. This is not caution: the local models depend on Apple silicon, and the voices run on the Neural Engine specifically.

Ollama is a separate install

The routing and summarising model is hosted by Ollama, which you install yourself. NovaHUD bundles its own speech recognition and voices, but not that.

No roadmap dates are published here. When a version ships it gets an entry above, and not before.

Running an older build?

There is only one so far. Download the current DMG and drag it over the copy in Applications.