Developer tool
Lyrebird
- Product design
- Development
- Open source
- Role
- Industry
- Year
- Product design, Development
- Developer tool
- 2026
Project overview01
Just speak. It writes.
macOS has no fast, private way to dictate into any app. Cloud tools send your audio to someone else's servers. Dictation software wants a subscription, training, or a permanent connection.
Lyrebird is the small thing that was missing. We designed and built it, and it is open source and free.
The challenge02
The product only justifies itself if it is faster than typing. Every second of interface between pressing the shortcut and having usable text is a second that pushes someone back to the keyboard.
At a glance03
The problem, and what changed
What was missing
macOS has no fast, private way to dictate into any app. Cloud tools send the audio away, and dictation software wants a subscription, training or a connection.
Interaction model
The whole interface is a keyboard shortcut: press, speak, release, and the text is on the clipboard. No window to manage, no account, no setup past picking a model.
Menu bar presence
A menu bar presence rather than an app: state is legible at a glance, and the settings are the few that actually change the outcome.
Scope of work04
What the work covered
The interface is a keyboard shortcut
A global shortcut starts recording. Stop, and the text is already on the clipboard. There is no window to find and no button to hunt for.
Whisper on the machine, not in a datacentre
The model runs on the Mac's own chip. Nothing is uploaded, which removes the privacy question rather than answering it.
Live in the menu bar
It has to be present without being an app you open. The menu bar is the honest place for a tool used dozens of times a day.
No account, no subscription, no telemetry
Free and open on GitHub. For a tool this small, anything else would cost more trust than it earns in revenue.
How it works
Lyrebird lives in the menu bar. Press the shortcut and it records. Release, and Whisper processes the audio locally in seconds, leaving the text on the clipboard ready to paste wherever you need it. No intermediate step, no loading screen, no internet.
Privacy by design
Audio never leaves the machine. Whisper runs locally through Core ML on Apple Silicon, or the standard inference engine on Intel. It works on a corporate network, on a plane, or with no connection at all. Nothing is logged, nothing is sent, nothing is retained.
- Local only
- No API keys, no account, no telemetry. The audio and the transcript stay on the Mac.
- Any app
- A global shortcut and the clipboard, so it works wherever the cursor already is.
- 90+ languages
- Automatic language detection, Spanish, English and Italian included.
Outcome
- A free, open app available for direct download, running on Apple Silicon and Intel, macOS 12 and up.
- Automatic language detection across Spanish, English, Italian and over ninety more.
- Zero network dependency: it works with the wifi off.
