Skip to content

Releases: foxl-ai/notes

Foxl Notes v0.1.3

30 Mar 05:55
d5ce6ff

Choose a tag to compare

Foxl Notes v0.1.3

What's New

  • Transcription Language Setting — new dropdown in Settings to select transcription language (8 languages + auto-detect). Changes apply mid-recording.
  • Client-side VAD — silent audio chunks are no longer sent to the ASR server, preventing model hallucination during silence.
  • Streaming duplicate fix — fixed silence commit loop that caused repeated segments in transcript.
  • Windows ASR rewrite — server now uses faster-whisper only (no Qwen/CUDA dependency), with robust EPIPE handling and simplified install.
  • Cross-platform client — handles both server response formats (Windows segments + macOS stable_text).

Local ASR

  • macOS: MLX backend (Qwen3-ASR-0.6B)
  • Windows/Linux: faster-whisper (Whisper base, int8, CPU)

Downloads

macOS (Universal)

  • DMG: Foxl Notes-0.1.3-universal.dmg
  • ZIP: Foxl Notes-0.1.3-universal-mac.zip

Windows (x64)

  • Installer: Foxl.Notes.Setup.0.1.3.exe
  • Portable: Foxl.Notes-0.1.3-win.zip

Platforms

  • macOS (arm64, x64) — MLX backend
  • Windows (x64) — faster-whisper (CPU, int8)
  • Linux (x64)

Foxl Notes v0.1.2

30 Mar 02:43
d5ce6ff

Choose a tag to compare

Foxl Notes v0.1.2

What's New

  • Qwen3-ASR ONNX INT8 for CPU — replaces faster-whisper with the same Qwen3-ASR model exported to pure ONNX Runtime. No PyTorch needed, 30 languages, ~0.38x RTF on desktop CPU.
  • Deep link auth fix (Windows URL hostname parsing)
  • Upgrade to Pro redirects to app.foxl.ai
  • Offline transcription endpoint

Local ASR Install

CPU machines now install:

  • onnxruntime + librosa + tokenizers (lightweight, no PyTorch)
  • Downloads Qwen3-ASR-0.6B ONNX model (~2.5GB) from HuggingFace

Platforms

  • macOS (arm64, x64) — MLX backend
  • Windows (x64) — ONNX INT8 (CPU) or qwen-asr (CUDA)
  • Linux (x64)

Foxl Notes v0.1.1

30 Mar 01:34
d5ce6ff

Choose a tag to compare

Foxl Notes v0.1.1

What's New

  • Windows ASR server with CPU fallback (faster-whisper int8 for CPU, qwen-asr for CUDA)
  • foxlnotes:// deep link authentication (Google, Apple, Magic Link)
  • Fixed deep link URL parsing on Windows (hostname vs pathname)
  • Local ASR reinstall support
  • Fixed relay URL (relay.foxl.ai)
  • Upgrade to Pro redirects to app.foxl.ai
  • i18n: model-agnostic language detection text
  • Health check before recording with local ASR
  • Auto-save after AI summary

Platforms

  • macOS (arm64, x64)
  • Windows (x64)
  • Linux (x64)