Para is a macOS menu bar assistant that provides hands-free dictation powered by NVIDIA’s Parakeet v2 0.6b speech model. The app runs entirely on-device, offering privacy-friendly transcription with contextual UI controls.
- ✅ Repository scaffolded with placeholder SwiftUI scenes and service stubs.
- ✅ High-level implementation plan in
Docs/implementation-plan.md. - ✅ Parakeet v2 0.6b NeMo checkpoint downloaded locally (
Docs/model-metadata.md). - ✅ Conversion plan and ONNX export script scaffolded (
Docs/model-conversion.md,Scripts/export_parakeet_onnx.py). - ⏳ Pending: Xcode project setup, audio capture wiring, Parakeet runtime integration, and UI build-out.
- Global hotkey to start/stop dictation (toggle or hold-to-talk).
- Floating “listening pill” overlay that visualizes live waveform input.
- Dark ShadCN-inspired control window for microphone selection, output preferences, and transcript history.
- Local Parakeet v2 0.6b transcription via MLX or ONNX Runtime.
- Optional clipboard copy and auto-paste into the focused app.
- Launch-at-login toggle, permission onboarding, and diagnostics view.
Docs/ Implementation plan and future design notes
Resources/ Placeholder for asset catalogs, sounds, etc.
Sources/
App/ SwiftUI app entry point and placeholder views
MenuBar/ Menu bar coordinator logic
Overlay/ Listening pill overlay controller and view
Audio/ Audio capture service (mic permissions, taps)
Transcription/ Parakeet model integration stubs
Preferences/ User preferences models and storage
Utilities/ Global application state management
- Create the Xcode project and hook it up to the existing
Sourceslayout. - Implement global hotkey handling and tie it into
AppState. - Wire up audio capture, device selection, and waveform visualization.
- Integrate Parakeet v2 0.6b (select MLX or ONNX runtime).
- Finish the control window UI and clipboard/paste automation.
We’ll keep this README updated as milestones land. PRs and issues will track individual work items once development begins in earnest.