Dibur · Free and open source
Stop typing.Just talk.
Press a shortcut, speak, and the text lands wherever your cursor is. Even the English mid-sentence comes out spelled right.
- Everything stays on your machine
- Open source
- No account, no subscription
v1.0.2
How it works
Press your shortcut
One keystroke, from any app in the system.
Talk normally
Hebrew, English, and every mix of the two. The way you actually speak.
The text is in
It lands right where your cursor is. No copy, no paste.
Every dictation tool trips on the same sentence
Transcribing Hebrew? They all manage. But say a tool name mid-sentence and you get word salad. Dibur was built for exactly these sentences.
Without Dibur
open a pool ricvest after the komit, push it to Versel, and check Super Base
With Dibur
open a pull request after the commit, push it to Vercel, and check Supabase
Speak with an Israeli accent, get correct English
Over 350 terms are built in: the tools, products and concepts we all work with. Say them in Hebrew and they come out spelled right.
- cloud codeClaude Code
- cloud em deeCLAUDE.md
- super baseSupabase
- verselVercel
- pool ricvestpull request
- gitahubGitHub
- komitcommit
- vordpressWordPress
And that is just the start. Add your own words: client names, project names, anything that keeps coming up.
Works where you work
Anywhere with a text field. The text goes straight into the app in front of you:
- Claude Code
- Cursor
- VS Code
- Terminal
- Gmail
- Slack
- Notion
- Obsidian
- Telegram
- Chrome
What you get
A Hebrew-specific engine
Not generic Whisper. A model fine-tuned for Hebrew, with automatic punctuation and readable paragraph breaks in longer dictations.
Over 350 terms built in
Claude Code, Supabase, Vercel, CardCom, Rav Messer and more. Even spoken in Hebrew, they come out written correctly.
Your own words
Add client names, project names and internal terms, and they come out right the first time.
A shortcut you choose
Set it once. From then on it is one keystroke away, anywhere in the system.
Local history
Everything you dictated stays on your machine, to revisit or delete whenever you want.
Automatic updates
The app updates itself. Every update is signature-checked, or refused.
Your voice never leaves the machine
Transcription runs on your own CPU or GPU. There is no server in the middle, no cloud, and no account to create. On first launch, one transcription model of about 900 MB downloads to your computer. It stays there and keeps working even when you disconnect from the internet.
Requirements
- macOS 12 or later, both Apple Silicon and Intel
- Windows 10 or later, 64-bit
- About 1 GB free for the transcription model, downloaded once on first launch
- A microphone, and accessibility permission so the text can be pasted into the app you are working in
For Windows users: what you will see
The macOS build is signed and notarized by Apple and opens without any warning. The Windows build is not signed yet, so SmartScreen will show a blue screen saying the publisher is unknown. Click "More info", then "Run anyway". It appears only the first time, and it goes away once a Windows signing certificate is in place.
One minute to install. Done.
Free, open source, and everything stays on your machine. Your next message is one you say, not type.
Download free