Repository navigation

Tags

v1.2.0

Build for Apple silicon only

Phonon-2's Core ML runner uses Float16, which isn't available on Intel.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Oct 16, 2026
05298e4
zip
tar.gz

v1.1.2

Bump version to 1.1.2 (build 17)

Oct 8, 2026
4d76953
zip
tar.gz

v1.1.1

Bump version to 1.1.1 (build 16)

Oct 8, 2026
469be24
zip
tar.gz

v1.1

Kaze 1.1: ChatGPT Clean Up, leaner memory, calmer notch

- Clean Up can use a model from the user's ChatGPT plan via Sign in with
  ChatGPT (OAuth + PKCE, dynamic client registration, loopback redirect,
  Keychain storage, refresh and revocation). Opt-in; S1-mini stays the
  on-device default. Defaults to GPT-5.6-Luna at the lowest reasoning
  effort, with a model and Thinking picker in Settings.
- Speech models are primed at launch, load on the shortcut press and are
  freed after 3 minutes idle or on memory pressure; S1-mini is freed right
  after each cleanup (1.4 GB -> ~150 MB).
- "Nothing heard" is a wordless shake instead of an error message; real
  errors show inline in one row.
- A word-novelty check pastes the raw transcript if a model answers or
  rewrites instead of cleaning.
- Redesigned Clean Up settings and onboarding step.
- Enables the sandbox's incoming-network capability for the sign-in
  redirect listener (127.0.0.1 only, during sign-in).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Oct 7, 2026
a8e7fa8
zip
tar.gz

v1.0

Merge origin/main (fayazara#10) into v1.0

The 1.0 rebuild replaces every file fayazara#10 touched with a new SpeechAnalyzer
engine; this merge keeps 1.0's tree and preserves fayazara#10's history.