Documentation
Set up VoiceBot
The setup walkthrough opens the first time you launch VoiceBot. You can open it again from the menu bar item Setup guide…, or from Settings. This page follows that walkthrough. Tidy, in step 3, can be skipped.
1. Install VoiceBot and allow it to listen and paste
Install VoiceBot and open it. The walkthrough opens on that first launch. macOS will not let the app hear you, see which app is in front, or paste there, until you allow two things.
- Open System Settings.
- Go to Privacy & Security.
- Open Microphone and turn VoiceBot on.
- Go back to Privacy & Security, open Accessibility, and turn VoiceBot on.
Microphone lets VoiceBot hear you. Accessibility lets VoiceBot see which app is in front and paste there, and it lets the hotkeys work.
After you update VoiceBot, macOS can keep an old Accessibility entry that no longer matches the app. If the hotkey stops working after an update, or VoiceBot no longer pastes into the front app, remove VoiceBot from the Accessibility list and add it again.
2. Speech
VoiceBot turns speech into text on your Mac with a tool called whisper.cpp. You install that tool with Homebrew, then download a speech model from inside VoiceBot.
Homebrew is a small installer for Mac tools. If you have never used it, install it first from brew.sh. The site shows one command to paste into Terminal. After that, Homebrew’s brew command is available.
Open Terminal (it is in Applications → Utilities, or search for Terminal with Spotlight) and paste this:
brew install whisper.cpp
Homebrew may show a notice about downloading GGML files. You can ignore that notice. When the install finishes, open VoiceBot and download a Whisper model from the app. The app does that download. You do not pick the files yourself.
3. Tidy, if you want it
Tidy is optional. It cleans up the words after they have been written down. In the app the choices are named Off, On this Mac, and Cloud. If you do not want tidy, choose Off. VoiceBot still types what you said.
Off
Nothing extra runs. The transcript is left as the speech step wrote it.
On this Mac
Tidy runs on your Mac. Ollama is the recommended way. llama.cpp is for advanced users.
Ollama, recommended. Install it from ollama.com, or in Terminal:
brew install ollama
VoiceBot uses Ollama at 127.0.0.1:11434. The recommended model is llama3.2:3b. The app can pull that model for you.
llama.cpp, for advanced users. Choose this if you already install tools yourself. If you are not sure, stay with Ollama. Once llama.cpp is installed, VoiceBot can use it for tidy on your Mac.
Cloud
You can use OpenAI, Anthropic, or another service that accepts an OpenAI-compatible connection. You bring your own API key. VoiceBot stores that key in the macOS Keychain on your Mac. The key is not sent to Edward.
With Cloud tidy, the transcript text is sent to the provider you pick. The audio never leaves the Mac.
4. The keys that start listening
These are the keys VoiceBot uses unless you change them in the app. On a Mac keyboard, Option is the key labelled Option (sometimes also Alt). Control is the key labelled control.
- Control Option Space turns listening on, and the same keys turn it off.
- Control Option V is push to talk. Hold the keys while you speak, and let go when you stop.
5. Try it
Press the hotkey, speak, and check the text appears in the front app.
If something is wrong
The hotkey does nothing
Check Microphone and Accessibility in System Settings → Privacy & Security, and turn VoiceBot on in both. If you have just updated the app, remove VoiceBot from the Accessibility list and add it again. Then try Control Option Space, or hold Control Option V.
Accessibility shows Not allowed
Open System Settings → Privacy & Security → Accessibility. Turn VoiceBot on. If the switch is on and it still says Not allowed, select VoiceBot, remove it, then add the app again and turn it on.
Tidy was skipped
That can be what you asked for. Off means tidy is skipped, and the words are still typed. If you chose On this Mac with Ollama, tidy is skipped until Ollama is running at 127.0.0.1:11434 and the llama3.2:3b model is there. The app can pull that model for you. If you chose Cloud, tidy is skipped until your own API key is saved in the Keychain. Skipping tidy does not send audio off the Mac.