Billie Flow v0.2.1

Billie Flow v0.2.1 is the install-tested public proof-of-concept release.

What changed since v0.2.0:

• First launch now opens Settings through SwiftUI's supported openSettings action.
• Local cleanup now works with the pinned mlx-lm runtime instead of falling back to raw ASR.
• Setup no longer waits indefinitely when its final verification command exits immediately.
• The release gate now requires a real on-screen first-launch window and warning-free production cleanup.
• Clean install, setup, recording, and copied-state screenshots are included in the source repository.

This remains intentionally narrow: Apple Silicon, macOS 26, English speech, fixed local models, copy-only output, and no updater or support promise. The 22 MB app archive does not duplicate model weights. On first setup it asks before downloading about 3.5 GB, including the fixed models from Hugging Face.

The app is ad-hoc signed and not notarised because this project does not use an Apple Developer Program membership. Follow the Gatekeeper steps in the README if macOS blocks the first launch.

Download the Apple Silicon zip and its matching .sha256 file below.

Billie Flow v0.2.0

Billie Flow is an unsupported proof of concept for local English speech-to-text on Apple Silicon.

This release contains the arm64 macOS app and its SHA-256 checksum. It bundles the pinned setup helper and worker source, but no model weights. On first launch, the app asks before downloading roughly 3.5 GB of Python dependencies and fixed Whisper/Qwen models from Hugging Face. Inference then runs locally.

Requirements: Apple Silicon and macOS 26 or later.

The app is ad-hoc signed and not notarised. Read the Gatekeeper warning and installation instructions in the README before opening it.