Talk to the board

Voice control lets you dictate into any text field (a ticket description, a reply, a chat) and drive the board with spoken commands: open a ticket, move it, search, submit. Speech is transcribed by a model that runs on the board’s own server, so your audio never leaves the machine rev0 runs on.

Before you begin

• A microphone, and the browser’s permission to use it (it asks the first time).
• The first time you use it, the board downloads the speech to text model, which takes a moment.

Turn voice control on

1. Open Settings > Automation > Voice control.
2. Tick Enable voice control.
3. Pick an Activation:
   • Push-to-talk (hold a key) (the default): the board listens only while you hold the Push-to-talk key, and sends what you said when you let go. The key is Backtick by default; Right Ctrl, Right Alt / Option, Right Shift and F8 also work.
   • Wake word (always listening): the board listens all the time, and a command must start with a wake word: “Kanban”, “board” or “computer” (optionally after “hey” or “ok”), as in “Kanban, next card”. Dictation into a focused field never needs the wake word.

This section belongs to this device: it saves as you change it, with no Save.

Result: a Voice pill appears at the bottom right of the board. It reads Listening… while it hears you and Transcribing… while it works. Click it, or press Cmd+Shift+V (Ctrl+Shift+V on Windows and Linux), to start or stop listening.

Dictate into a field

1. Click into any text field: a ticket’s Description, the reply box, a chat message.
2. Speak (holding the push to talk key, if that is your activation).

Result: your words are typed where the cursor is. To leave the field hands free, say “stop dictation”; your next words are read as a command again.

Fix what it mishears

Settings > Automation > Autocorrect fixes words as they are dictated.

1. Keep Autocorrect dictation ticked (it is on by default).
2. Under Correction dictionary, click + Add correction, write what it hears in heard (for example “cabana”) and the right word in fix (for example “kanban”).

When a dictation has corrections to apply, a review box shows each suggested fix before it is typed. Tap a highlight to keep the original word, then click Insert, or dismiss it to insert nothing.

Result: the words you add are fixed every time you dictate them on this device.

Say a command

With no text field focused, what you say is read as a command. Things you can say, at the bottom of the voice settings, lists examples:

To	Say
Move around the board	“next card”, “previous card”, “next column”, “next board”
Open or close a ticket	“open ticket 316”, “open”, “close”
Create a ticket	“new ticket”, “create a ticket titled export as CSV”
Move a ticket	“move to in review”, “move to done”, “move left”
Search	“search login bug”
Send what you wrote	“submit”
Open a view	“settings”, “catch up”, “help”, “toggle theme”, “show archived”
Stop listening	“stop listening”

Result: the board does it, and the Voice pill shows what it did. A phrase it does not recognise shows Didn’t catch that with what it heard.

Add your own commands

1. Open Settings > Automation > Custom voice commands.
2. Click + Add command.
3. Write the phrase. Use a slot for the part that changes: {ticket}, {status}, {title} or {query}, as in “send {ticket} to {status}”.
4. Pick the action: Open ticket {ticket}, Move to {status}, Create ticket {title}, Search {query}, Close ticket, Submit / send, Stop dictation (blur field), or any keyboard shortcut command.
5. Click Save.

Result: your phrase works on every device you open the board from. Your commands are matched before the built in ones, so you can reword a built in command too.

The LLM fallback

Settings > Automation > LLM fallback has Understand free phrasing with an LLM, off by default, and the LLM model it would use (claude-haiku-4-5). It is meant to send a phrase that no command matched to that model to work out what you meant.

Note

The board does not answer these requests yet, so with the fallback on, an unmatched phrase still ends in Didn’t catch that. Add a custom command for phrasing you use often.

Pick the speech to text model and language

1. Open Settings > Automation > Transcription.
2. Pick the Speech-to-text model:
   • Whisper base (the default): fast, English and other languages.
   • Whisper small, Whisper large-v3-turbo and Whisper large-v3: more accurate and slower, in that order of size.
   • Distil-Whisper large-v3: English only, fast.
   • Whisper large-v3-turbo (whisper.cpp, Apple Metal GPU): the fastest on a Mac with Apple Silicon. It needs the voice-metal extra on the server (uv sync --extra voice-metal in a source install).
   • The models marked coming soon are not usable yet.
3. Pick the Speech-to-text language: English (the default), French or Auto-detect.
4. Click Save.

Result: the next thing you say is transcribed with that model. A bigger model is downloaded the first time you use it.

If something goes wrong

Enable voice control in Settings > Automation > Voice control first

You pressed the voice shortcut with voice control off. Turn it on in Settings > Automation > Voice control.

Voice error

The pill turns to Voice error when listening fails, most often because the microphone is blocked or the model could not load. Allow the microphone for the board’s address in the browser, and check that the model you picked is not one marked coming soon.

Didn’t catch that

The phrase matched no command. Check Things you can say, say it with the wake word if your activation is Wake word, or add it as a custom command.

Related

• Keyboard shortcuts, for the push to talk and toggle keys.
• Write a ticket an agent can finish, for what to dictate.