Skip to content

All projects

Vurb

Product spec Shipped product: figures from vurb.me

A shipped desktop dictation app: hold a key, talk, let go, and the words are typed wherever the cursor is. Speech is transcribed on the user’s own computer, so the audio never leaves it.

Status
Live at vurb.me
Relevant to
AI and agents
Live
vurb.me

Results · Shipped product: figures from vurb.me

  • to transcribe a 12-second clip on Apple Silicon with the best modelProduct spec
  • detected automatically, including switching between languagesProduct spec
  • installers shipped, free to startProduct spec
Vurb home page: hold a key, talk, let go
Vurb home page: hold a key, talk, let go

The problem

Most people talk much faster than they type, but cloud dictation sends your voice to someone else’s servers. I wanted dictation that works in every app, works offline and keeps the audio on the machine.

What I built

  • Hold-to-talk in any text field, a hands-free mode on double-tap, and Esc to throw a dictation away.
  • Local transcription with OpenAI Whisper via whisper.cpp, with automatic language detection, including switching languages mid-dictation.
  • Optional AI cleanup that is off until the user turns it on. When it’s on, only the text goes to Anthropic, using the user’s own API key.
  • Custom vocabulary, searchable history and stats, plus a marketing site, pricing and installers at vurb.me.

Limitations

  • A small product I build and sell on my own. The numbers here come from vurb.me, not from a published benchmark.

Stack

OpenAI Whisper, whisper.cpp, Anthropic API