Byblos User Manual

Version 0.1.0

Getting Started

First Launch

When you first open Byblos, an onboarding wizard guides you through:

  1. Microphone permission — Byblos needs mic access to hear you.
  2. Download a model — Pick a speech model and optionally an AI model.
  3. Ready — Byblos appears as a waveform icon in your menu bar.

Permissions

Without Accessibility permission, Byblos copies transcriptions to your clipboard instead of typing directly.

Recording & Transcription

Starting a Recording

Three ways to record:

  1. Left-click the menu bar icon — click once to start, click again to stop.
  2. Hold-to-record hotkey — hold Option (configurable), speak, release.
  3. Transcript workspace button — click the red record button at the bottom.

While Recording

Stopping

Where Does the Text Go?

  1. Your active app — text is pasted into whatever app you were using.
  2. Transcript history — every transcription is saved automatically.

Tip: If text isn't appearing, check Accessibility permission. Without it, text is copied to your clipboard — press ⌘V to paste.

Undo

Say "scratch that" or "delete that" to undo the last transcription.

Dictation Modes

Right-click the menu bar icon → Mode to switch.

Clean (default)

Removes filler words (um, uh, like, you know), fixes punctuation and capitalization.

Email

Professional tone with paragraph breaks. Great for Mail, Gmail, Outlook.

Notes

Converts speech into bullet points. Pairs well with Notes.app, Obsidian, Notion.

Translate

Speak in any language, get English text. 99+ languages supported via Whisper.

Raw

Exact transcription with no processing.

Code Comment

Concise output prefixed with //. For developers dictating comments.

Agent EXPERIMENTAL

AI assistant mode. See Agent Mode.

Automatic Mode Switching

When enabled, Byblos picks the right mode based on which app you're using: Mail → Email, VS Code → Code, Notes → Notes, etc.

Transcript Workspace

Right-click → Show Transcripts.

File Transcription

Drag audio/video files onto the Transcript workspace, or click Import Audio.

Supported: WAV, MP3, M4A, FLAC, OGG, MP4, MOV, MKV.

Managing Models

Settings → Models.

Speech Models (Whisper)

ModelSizeBest For
Tiny74 MBQuick notes, low memory
Base141 MBGeneral use
Small465 MBBetter accuracy
Large v3 Turbo1.5 GBFast + excellent quality
Distil-Large v31.4 GBNear-best quality, great speed
Medium1.5 GBHigh accuracy
Large v32.9 GBMaximum accuracy

CoreML encoders are downloaded automatically for 3x faster transcription on Apple Silicon.

Recommendation: Start with Distil-Large v3 — near-best quality at good speed.

Custom Vocabulary

Settings → Vocabulary. Add replacements for names and jargon that Whisper misspells.

Examples: "byblos" → "Byblos", "kubernetes" → "Kubernetes", "react" → "React"

Settings Reference

General

Models

Download, activate, and remove speech and AI models.

Audio

Input device, noise suppression, VAD.

Agent Mode EXPERIMENTAL

Requires a local LLM.

Local LLM Setup (Optional)

Download an AI model in Settings → Models. Recommended: Qwen 3 8B (4.7 GB, needs 16GB+ RAM) or Qwen 3.5 4B (2.7 GB, 8GB+).

The LLM runs in a separate helper process alongside Whisper, both using Metal GPU.

License & Support

Free for personal use. No limits, no nag screens, all features available.

Commercial use (work, business, revenue) requires a license: $49/user/year. Honor system — we trust you.

To activate a commercial license: Settings → About → paste key → Activate.

support@byblos.im · GitHub Issues

Troubleshooting

No transcription

Text not appearing in app

Hotkey not working

Logs

~/Library/Logs/Byblos.log

Quick Reference

ActionHow
Start/stop recordingLeft-click menu bar icon
Hold-to-recordHold Option
Open menuRight-click menu bar icon
Stop from overlayClick the overlay
Undo lastSay "scratch that"
Auto-stopPause ~3 seconds
Show TranscriptsRight-click → Show Transcripts