Repository navigation
Releases
Floe 0.0.1 Preview
Updated 0.0.1 preview build (2026-11-07)
The DMG has been refreshed while retaining app version 0.0.1. Download it again to get this build; previously downloaded copies are unchanged.
- Stops permission helper launches once Input Monitoring and Accessibility are granted.
- Avoids redundant settings updates and removes the waveform animation scheduler while idle.
- Uses a distinct Floe bundle identity and migrates existing hotwords and preferences.
- Grant all three permissions again when upgrading from the former identity. macOS may retain older display names in its permission UI.
Source for this updated build: ab62ceed4502aaefee649d718f4b9858fd647f8f. The v0.0.1 tag now points to this commit, so the automatically generated source ZIP and tar.gz include this updated build.
Floe 0.0.1 preview: local Chinese and English streaming dictation for Apple Silicon Macs.
Install
- Download
Floe-0.0.1-arm64.dmgbelow. - Open it and drag Floe into Applications.
- Eject the disk image, then launch Floe from Applications.
- Grant Microphone, Input Monitoring and Accessibility access.
- Set Keyboard → Press 🌐 key to → Do Nothing, then hold 🌐 / Fn to dictate.
Requires Apple Silicon and macOS 14 or later. The model is included; speech recognition stays on your Mac.
First launch
This is an ad-hoc signed, not Apple-notarized preview. If macOS blocks it, attempt to open Floe once, then go to System Settings → Privacy & Security → Open Anyway. Authenticate if asked and confirm Open. Only approve the download from this repository. Managed Macs may prevent this exception. Updates may require granting permissions again.
Included
- Local streaming X-ASR bilingual Zipformer model, CPU inference.
- Live text insertion, 5 ms default update interval.
- Editable hotwords and optional correction-based suggestions.
- Glass icon and a small recording waveform.
- DMG with an Applications shortcut and installation instructions.
A SHA-256 checksum is attached. This preview has passed build, streaming text, hotword, audio-tail, clipboard-marker, model-inference and disk-image validation checks. Recognition accuracy and compatibility with every target app are not guaranteed by those checks.