Control the hub by voice
Hold Ctrl+Shift+Space, speak, and let go. The hub turns what you said into an action: it starts a new session, or types your words into the session on screen as a prompt for you to review. Your words show up as you speak. To send the prompt at once, press Enter before you let go.
Voice is off by default. You turn it on with the Voice control extension.
Before you start
Section titled “Before you start”- Hex (recommended). Hex does the speech to text, on your machine, on Windows, macOS and Linux. Get it from its website or its releases. Install it and start it once, so that it downloads its speech model. Hex must be running while you use voice. Without Hex, see Without Hex.
- The embedded terminal. Kanban sends prompts to the session in its embedded terminal. A session that runs in another terminal does not get them.
Turn voice on
Section titled “Turn voice on”claude-code-hub ext install voiceThe command turns voice on in the hub’s config.json, and tells you when Hex is not installed. Then restart the hub.
To turn voice off again:
claude-code-hub ext remove voiceTalk to the hub
Section titled “Talk to the hub”Hold Ctrl+Shift+Space while you speak. A pill at the bottom of the window shows Listening. Let go of the keys when you are done.
Your words show up while you speak, about once a second: in the session’s input box, or in the Prompt field of the New session dialog. When you let go, the final text replaces them and stays in the input box. Read it, then press Enter to send it.
To send without the review, press Enter while you still hold the keys. The pill shows sends on release, and the prompt is sent when you let go.
To change your mind, press Backspace while you still hold the keys. The hub stops listening, removes the words it typed and sends nothing. A New session dialog that your words opened closes. Then let go.
The first time, the hub window asks to use your microphone. Allow it: the hub records while you hold the keys and sends the sound to Hex on your machine. If you block it, or your Hex has no live text, Hex records and your words show up only when you let go.
Hex keeps Ctrl+Space for plain dictation into any app, so both work side by side.
Without Hex
Section titled “Without Hex”When Hex is not installed, the hub uses the browser’s speech recognition (Chrome or Edge), on any OS. It works the same way: hold the keys, speak, let go. The difference is privacy: the browser sends the sound to an online speech service while you hold the keys. The first time, a note in the hub window tells you so.
To pick one yourself, set provider in Settings.
A command starts with the word “Hub”. Everything else is a prompt.
| Say | The hub |
|---|---|
| “Hub, new session” (also “Hub, start a session”, “Hub, session”) | Opens Kanban’s New session dialog |
| “Hub, new session, fix the flaky login test” | Opens the dialog with “fix the flaky login test” as the first prompt |
| “Hub, switch to docs panel” (also “Hub, switch docs panel”, “Hub, open session …”, “Hub, go to session …”) | Opens the Kanban session whose name, project or branch has your words. With no clear winner, the session picker opens with your words and the best matches first: press Enter to open the top one, or pick another |
| “Hub, add a task, fix the login page” (also “Hub, new task …”, “Hub, task …”) | Opens Kanban’s Add task dialog for the session on screen. Your words go into Description as you speak. When you let go, a task of 8 words or fewer moves up into Title. A longer one stays in Description as said and gets a title: from taskTitleModel when it is set (see Settings), else the words before its first comma or period. Press Enter to add the task, or press Enter during the hold to add it when you let go |
| “Hub, approve” (also “Hub, allow”, “Hub, accept”, “Hub, yes”) | Allows the permission prompt or plan that the open session waits on |
| “Hub, deny” or “Hub, deny, use the staging database” (also “Hub, reject …”, “Hub, no …”) | Denies it. Words after the command go to Claude as the reason. A question is not answered by voice: its card opens so you pick the answer |
| “Hub, next” (also “Hub, next waiting”, “Hub, next session”) | Opens the next session that needs you: first the ones that wait on you, then the unread ones. Say it again to move on, and it starts over at the end |
| Anything else | Types it into the session on screen as a prompt. Enter during the hold also sends it |
“Hub” must be the first word. Once the hub hears it, nothing more goes into the session: the pill shows the command as you speak. “Hub” followed by words that are not a command shows a toast.
While you hold the keys, a panel at the right side of the window lists the commands you can use on this tab, and lights the one your words match. Not sure what to say? Say only “Hub” and let go: the panel stays a few seconds. Kanban’s help (?) lists the same commands in the Voice group of its Hub tab.
Every way to say it
Section titled “Every way to say it”The panel and the help show one phrase per command. The hub also takes the forms below. Words in brackets are optional, and a/b means either word. Case and punctuation do not matter.
| Command | Forms |
|---|---|
| Prefix | hub, also hob and hab, which speech services often hear in place of “hub” |
| New session | [start/create/new] [a/the] [new] session, then the prompt |
| Open session | open/show [the/my] session or go to [the/my] session, then the name. Also switch [to] [the/my] [session], then the name |
| Add task | [add/create/new] [a/the] [new] task, then the task |
| Approve | approve/allow/accept/yes, with nothing after it |
| Deny | deny/reject/decline/no, then an optional reason |
| Next | next [waiting/session/one], with nothing after it |
Approve and Next take no words after them, so “Hub, approve, but not that file” does not approve.
On another tab
Section titled “On another tab”Voice works on the tab on screen and never switches tabs. Its commands all belong to Kanban, so on another tab, such as Cost, holding the keys shows No voice commands on Cost and the hub does not listen.
Where a new session starts
Section titled “Where a new session starts”“Hub, new session” opens Kanban’s New session dialog as soon as Hex hears it, while you keep speaking. The words after it go into the Prompt field, so you see what starts before it starts. While you speak, the dialog glows and shows Listening, and the Prompt field is read-only. The folder is that of the session on screen, else the project that Kanban’s project filter shows, else the most recent project. Press Enter to start it, or change a field first.
When a prompt is not sent
Section titled “When a prompt is not sent”A prompt goes to the session whose terminal is on screen. When you look at the open session’s board or a pane and its terminal runs in Kanban, the prompt goes to that terminal in the background, and a toast offers Show to bring the terminal up. While Claude works, it queues the prompt and runs it next. When no session terminal can take it, nothing is sent: a toast shows what you said, with a Copy button.
Kanban does not send a prompt while Claude shows a dialog, because the Enter press would answer it. A toast tells you so.
Dictate in the New session dialog
Section titled “Dictate in the New session dialog”The microphone button in Kanban’s New session dialog uses Hex too, when voice uses Hex. Click it, speak, and click it again: the text goes into the Prompt field where the cursor is. Nothing is sent until you press Start.
Otherwise (voice off, voice through the browser, or Kanban alone), the button uses the browser’s speech recognition.
Settings
Section titled “Settings”The extension writes voice in the hub’s config.json. You can change it by hand, then restart the hub.
{ "voice": { "enabled": true, "provider": "auto" }}| Key | Default | Meaning |
|---|---|---|
enabled |
false |
Turns voice on |
provider |
auto |
What turns speech into text: hex, browser, or auto (Hex when it is installed, else the browser) |
taskTitleModel |
none | A Claude model, such as haiku, that writes a short title for a task longer than 8 words. It runs claude -p on your machine with the active config dir’s login, so it uses your Claude plan: about $0.0002 and 3 to 4 seconds per such task |
To use another key, set hub.voice in keys. |
When it does not work
Section titled “When it does not work”- Nothing happens when you hold the keys. Voice is off, or the hub did not restart after you turned it on.
- “Voice: hex …” toast. Hex is not installed or not running. Start Hex and try again.
- “Speech recognition: not-allowed” or “network”. The browser blocked the microphone, or cannot reach its speech service. Allow the microphone, check the network, or install Hex.
- “No speech”. Hex heard nothing. Check the microphone that the hub window uses (the camera icon in the address bar of a browser tab, or Chrome’s site settings for the app), and the one that Hex uses.
- The words are wrong. Hex’s history shows what it heard.