macOS
Your systemMacs with Apple Silicon (M1 or newer), macOS 12 Monterey or later. Intel Macs aren’t supported yet.
Open the .dmg and drag Bolikh to Applications.
Step-by-step setup for macOS →Bolikh bol “speak” + likh “write”
Your speech becomes clean text in any app: email, team chat, documents, the terminal, the browser. It runs on your own computer, so your voice never leaves it.
Click where you want the text, then use one key. There is nothing to switch to and no window to copy from.
Press and hold your dictation key: Right Option on a Mac, Right Ctrl on Windows and Linux. A small bar at the bottom of the screen shows Bolikh is listening.
Talk the way you normally would. Pauses, “um”s and “uh”s are fine. Say “new paragraph” or “new line” to format as you go.
Let go. In a fraction of a second the cleaned-up text appears where your cursor is, and your clipboard is put back the way it was.
So um I was thinking we could, uh, move the meeting to Friday hmm if that works for everyone.
So I was thinking we could move the meeting to Friday if that works for everyone.
Um, uh, hmm, erm and mm are removed, and the commas around them are tidied up.
Pick your key and how it behaves. On a Mac: Right or Left Option, Right Ctrl, Right Cmd, Right Shift or Fn. On Windows and Linux: Right or Left Alt, Right Ctrl, Right Windows key or Right Shift. Press Esc at any time to cancel.
In our tests on an Apple Silicon Mac, a 9.5-second sentence was transcribed in 0.24 seconds. No graphics card needed.
If you can type in it, you can dictate into it: mail, chat, browsers, office apps, terminals and code editors.
Say “new line”, “next line” or “new paragraph” and Bolikh formats the text instead of typing the words.
Teach it how to spell names, brands and jargon. Every time it hears one, it writes it your way.
Text is pasted through the clipboard, then whatever you had copied before (text or an image) is put back.
Your last 200 dictations stay on your computer so you can copy one again. Turn it off or clear it any time. Audio is never saved.
Lives in the menu bar or system tray, can start when you log in, and plays soft start and stop tones you can switch off.
Talk in any of these and Bolikh writes it in that language, with punctuation and capitals. There is no language menu to change.
Hindi and Urdu are next. They gave Bolikh its name.
Many dictation apps send your audio to a server. Bolikh runs the speech model on your own machine. After a one-time model download it works with Wi-Fi switched off. The code is public, so anyone can check this.
Pick your system. On first launch, Bolikh walks you through microphone and permission setup and downloads the speech model.
Macs with Apple Silicon (M1 or newer), macOS 12 Monterey or later. Intel Macs aren’t supported yet.
Open the .dmg and drag Bolikh to Applications.
Step-by-step setup for macOS →Windows 10 or 11, 64-bit.
Run the installer. No admin rights needed.
Step-by-step setup for Windows →Ubuntu 24.04 or later, Debian 13, Fedora 40 or later. 64-bit Intel or AMD PCs, X11 and Wayland.
Install the package, then join the input group.
Step-by-step setup for Linux →Bolikh’s setup screen takes you through each step and ticks it off when it’s done. Here are the same steps, so you know what to expect.
Open the .dmg you downloaded and drag Bolikh into Applications. Then open Bolikh from Applications.
If macOS says it can’t verify Bolikh, see “macOS won’t open Bolikh” below.
Click Test in the setup screen and say a few words. macOS asks if Bolikh may use the microphone: click Allow. Bolikh then tells you the microphone works.
Click Open Settings in Bolikh. In Privacy & Security → Accessibility, switch Bolikh on, then come back. The step ticks itself off.
This lets the key work while you’re in other apps, and lets Bolikh paste for you.
Bolikh downloads its speech model once (about 670 MB) and shows the progress. After that it works offline.
Click in the practice box, hold Right Option, say a sentence and let go. Bolikh then waits in the menu bar. Open Settings from there to change the key or how it behaves.
Run the Bolikh setup file. It installs for your account only, so you don’t need admin rights.
If Windows shows “Windows protected your PC”, see the question about it below.
Open Bolikh, click Test and say a few words. Bolikh tells you when the microphone works.
If it says it’s very quiet or can’t find a microphone, go to Settings → Privacy & security → Microphone and turn on “Let desktop apps access your microphone”.
Bolikh downloads its speech model once (about 670 MB) and shows the progress. After that it works offline. Windows needs no other permissions.
Click in the practice box, hold Right Ctrl, say a sentence and let go. Bolikh then waits in the system tray. If you can’t see its icon, click the ^ arrow next to the clock.
On Ubuntu or Debian:
sudo apt install ./Bolikh-linux-amd64.debOn Fedora:
sudo dnf install ./Bolikh-linux-x86_64.rpmBolikh reads the dictation key directly from the keyboard, which works on both X11 and Wayland. Add yourself to the input group, then log out and back in:
sudo usermod -aG input $USERCheck which session you’re running:
echo $XDG_SESSION_TYPEIf it says x11, skip this step. On Wayland, install wtype for KDE Plasma, Sway or Hyprland. On GNOME, install ydotool and make sure its ydotoold service is running.
sudo apt install wtypeOpen Bolikh, click Test and say a few words to check the microphone. Then let it download the speech model once (about 670 MB).
Click in the practice box, hold Right Ctrl, say a sentence and let go. Bolikh then waits in the system tray.
Open the Apple menu → About This Mac. If it says Chip: Apple M1 (or M2, M3, M4 and later), Bolikh works. If it says Processor: Intel, it doesn’t yet: the speech engine Bolikh uses isn’t available for Intel Macs at the moment.
Click Done, then open System Settings → Privacy & Security. Scroll down, click Open Anyway next to the Bolikh message, and confirm. You only need to do this the first time.
Click More info, then Run anyway. You only need to do this the first time.
This can happen after an update. In Privacy & Security → Accessibility, select Bolikh, remove it with the − button, then open Bolikh and switch it on again.
Run groups and check that input is in the list. If you only just added it, log out and back in. On Wayland, also check that wtype or ydotool is installed. Bolikh’s setup screen shows which of these is missing.
On Windows, Bolikh can’t type into apps that run as administrator unless Bolikh runs as administrator too. On Linux, most terminals paste with Ctrl+Shift+V, which Bolikh doesn’t send yet, so dictating into a Linux terminal doesn’t work for now. Terminals on macOS and Windows work.
Bolikh uses your system’s default input. Change it in your sound settings (macOS: System Settings → Sound → Input; Windows: Settings → System → Sound → Input), then try again.
Check your internet connection and that about 1 GB of disk space is free, then click Retry in the setup screen. On office networks, the firewall must allow huggingface.co. Every file is checked when it arrives, so an interrupted download can’t leave a broken model behind.
The download is fine. macOS shows this for apps that aren’t notarized by Apple yet. Move Bolikh to Applications, open Terminal and run:
xattr -cr /Applications/Bolikh.appThen open Bolikh again.
macOS: System Settings → Privacy & Security → Microphone, switch Bolikh on, then quit and reopen Bolikh. Windows: Settings → Privacy & security → Microphone, turn on microphone access and “Let desktop apps access your microphone”.
Bolikh isn’t code-signed yet, and some antivirus tools are suspicious of any unsigned app. Bolikh contains no ads, trackers or network code beyond the one model download, and its full source code is public. Restore it from quarantine or add it to your antivirus’ allowed apps.
Bolikh’s window uses Microsoft Edge WebView2, which Windows 11 and most updated Windows 10 PCs already have. If yours doesn’t, the installer downloads it from Microsoft automatically. This needs an internet connection and only happens once.
Install with apt rather than dpkg -i: sudo apt install ./Bolikh-linux-amd64.deb downloads everything Bolikh needs from your distribution. If you already used dpkg, run sudo apt -f install to finish. On Fedora, dnf install handles this automatically.
Your distribution is older than Bolikh supports (for example Ubuntu 22.04 or Debian 12). Bolikh needs Ubuntu 24.04, Debian 13, Fedora 40 or newer.
Bolikh has to be running: look for its icon in the menu bar (Mac) or system tray (Windows, Linux) and open it if it isn’t there. Turn on Launch at login in Settings so it’s always ready. Also check that setup is finished: every step in the setup screen should have a tick.
Bolikh types wherever your cursor is when you let go of the key. Click into the text box first, then hold the key and speak.
Speak at a normal pace, keep a steady distance from the microphone and reduce background noise. A headset helps in noisy rooms. For names and jargon it keeps getting wrong, add them in the Dictionary tab.
Download the new version from this page and install it over the old one. Your settings, dictionary and history are kept. On macOS you may need to switch Bolikh on in Accessibility again after updating (see above).
Open Bolikh’s Settings. Dictation key picks the key and Behaviour picks Smart, Hold or Toggle. Fn is available on Macs only.
Ubuntu shows tray icons out of the box. On GNOME in Fedora or Arch, install the AppIndicator extension. Without a tray, Bolikh keeps its window open so you can still reach Settings.
macOS: drag Bolikh from Applications to the Bin. Windows: Settings → Apps → Installed apps → Bolikh → Uninstall. Linux: sudo apt remove bolikh or sudo dnf remove bolikh.
It joins two Hindi and Urdu words: bol, “speak”, and likh, “write”. That is the whole app: you speak, it writes.
Only once, to download the speech model on first launch. After that, dictation works fully offline.
Any app where you can type: Gmail and other web apps, Outlook, Slack, Teams, WhatsApp Desktop, Word, Google Docs, Notion, VS Code, terminals and more. It types at your cursor by pasting, the same way you would.
English and 24 other European languages, shown above. Bolikh detects the language on its own. Hindi and Urdu are planned for a later version.
Filler sounds are almost never wanted in written text, so they are taken out and the sentence is tidied up. Real words are left alone. You can switch this off in Settings if you want a word-for-word transcript.
No. Audio stays in memory only while you speak and is discarded as soon as it has been turned into text. Nothing is sent to us or anyone else.
Yes. The full source code is on GitHub under the GPL-3.0 licence, so anyone can read it, check what it does with their voice, and build it themselves.
Downloading and trying Bolikh is free. Pricing will be announced before version 1.0.
Aazad Software Solutions, a software studio that builds apps, websites and AI automation. The speech model is NVIDIA's Parakeet TDT 0.6B v3, used under the CC BY 4.0 licence.
For macOS, Windows and Linux. No account to create.
Download Bolikh