Skip to content

A small, open-source Mac dictation app that uses your ChatGPT sign-in. Speak into T3 Code, your editor, or any text field.

3.2 KB
Raw Blame

Releasing GPT Voice

Build and package releases locally. An App Store Connect app record, TestFlight build and App Store listing are not required.

Build

  1. Update VERSION and increase BUILD_NUMBER.
  2. Run swift test.
  3. Set GPT_VOICE_SIGNING_IDENTITY to the intended signing identity.
  4. Run ./scripts/build.sh --universal.

The result is dist/GPT Voice.app, with Apple Silicon and Intel executable slices. The build enables the hardened runtime for certificate-signed builds and includes the microphone entitlement. It keeps the original bundle identifier for existing settings.

Apple validation

For the Developer ID distribution workflow, submit the signed app to Apple with your configured credentials. After an Accepted result, staple the ticket and check it:

xcrun stapler staple "dist/GPT Voice.app"
xcrun stapler validate "dist/GPT Voice.app"
spctl --assess --type execute --verbose=2 "dist/GPT Voice.app"

Package the installer

Use a separate Python environment for packaging:

python3 -m venv .local/release-tools
.local/release-tools/bin/pip install dmgbuild==1.6.7
.local/release-tools/bin/python scripts/package-dmg.py

The script requires a valid Developer ID signature, a stapled ticket and a passing Gatekeeper assessment. It creates the themed Finder disk image with an Applications link. It mounts the result and checks every architecture signature and executable before marking the package ready. Do not add Finder metadata to the app after signing. In particular, hiding its filename extension can invalidate strict signature checks. --preview permits a local installer preview.

Sign the DMG with Developer ID, submit it for notarization, and staple the accepted ticket. Run xcrun stapler validate dist/GPT-Voice.dmg before release. Check it with hdiutil verify and spctl --assess --type open --context context:primary-signature --verbose=2.

Check the download experience

  • Mount the final DMG and inspect its icon layout and instruction text.
  • Copy the app to Applications. Eject the DMG before launching the installed copy.
  • Test onboarding, sign-in, permissions, a recording, paste and cancel.
  • Check light and dark mode and Reduce Motion.
  • Download the uploaded asset through its public release URL and compare its SHA-256 checksum.
  • Check Gatekeeper with quarantine on the downloaded copy. Do not disable security checks.
  • Prefer a separate Mac with no earlier GPT Voice grants for a final first-install check.

An Intel slice can be built on Apple Silicon, but Intel execution requires a separate check.

Publish

Attach GPT-Voice.dmg, an app ZIP, checksums, release notes and the promo MP4 to the matching GitHub release. The website's download link must point to the uploaded release asset. Update the README, website and launch drafts to the same version and verified release status. Keep signing keys, credentials and local capture logs out of Git.

Packaging modes

For the v0.1.4 packaging workflow, run scripts/package-dmg.py --unnotarized. This mode requires a certificate signature and preserves the actual validation results in the JSON report. The default mode requires Developer ID signing, a stapled ticket, and a passing system assessment. Keep the package report accurate for the selected mode.