Tarsen Voice: commands and how it works
Press ⌘J, say what happened, and the board changes. This page lists everything Tarsen Voice can actually do, with things you might say for each, and what it checks before it touches your work. For why a board you talk to is worth having, see the voice Kanban feature page.
Start and stop
| Keys | What happens |
|---|---|
| ⌘J / Ctrl+J | Start talking. During a conversation, it hides or reopens the panel without hanging up. |
| ⌘⇧J / Ctrl+Shift+J | Mute or unmute the microphone. Works while the panel is hidden. |
| Esc | End the conversation. |
All three keys work while you are typing in a field, so you can stop mid-sentence and say the rest. Clicking the orb does the same as ⌘J: it starts a conversation, or hides the panel. It never hangs up. Ending is the End button or Esc.
While the panel is hidden, a running clock sits under the orb, so a live microphone is never invisible. Talk normally. Tarsen works out that you have finished from what you said, not from how long you paused, and you can interrupt its reply.
Voice is off by default. A workspace admin switches it on in Workspace → Settings, and AI has to be switched on there too. You need a browser, a microphone and a connection; there is nothing to install.
Cards
| What it does | Say something like |
|---|---|
| Create a card, with a description, priority, due date and column if you give them | “New card: chase the design feedback, due Friday, high priority.” |
| Move a card to another column | “Move the billing migration to In Review.” |
| Mark a card done, or reopen it | “TR-18 is done.” / “Move TR-18 back to In Progress.” |
| Change the title, description, priority or due date | “Make TR-42 urgent.” / “Push the launch email to next Tuesday.” / “Clear the due date on TR-42.” |
| Assign or unassign people, by name or email | “Give TR-42 to Ada.” / “Take me off the onboarding card.” / “Reassign it from Sam to me.” |
| Archive a card | “Get rid of the old pricing card.” |
| Delete a card permanently, after a spoken yes | “Delete TR-31 permanently.” |
A card is created straight away, without reading the title back first, because you hear it in the reply anyway. With no column named, it goes in the first one. Moving a card into the done column completes it; moving it back out reopens it. When you say a day rather than a date, Tarsen works out the real date from today's.
Columns and projects
| What it does | Say something like |
|---|---|
| Add a column, or rename or reorder one | “Add a QA column after In Review.” / “Rename Doing to In Progress.” |
| Create a project, from a board layout, after a spoken yes | “Start a new project called Website relaunch.” |
| Rename a project | “Rename Tarsen-v1 to Tarsen.” |
| Archive a project, or bring one back | “Archive the Q3 launch project.” |
Before creating a project, Tarsen reads the name back and asks which board layout you want, naming no more than three. Say you don't mind and you get a plain board. It will not delete a project: that is something you do yourself, from the project's settings.
Questions about your work
| What it answers | Say something like |
|---|---|
| What is on a board, column by column | “What's in progress on the Tarsen board?” |
| One card in full: description, people, labels, checklist, recent comments | “What's the latest on TR-89?” |
| Find a card by what it is about | “Find the card about drag errors.” |
| Your open cards, soonest due first | “What am I working on?” / “What's due?” |
| How much you finished, by day | “How many did I get done today?” / “How was yesterday?” |
| Your inbox: mentions, assignments, due reminders, replies | “What did I miss?” |
| What changed recently, and who did it | “Who moved the billing card?” |
| Your plan, usage, members and which features are on | “What plan are we on?” / “Is voice on?” |
| The projects you can see | “What projects do we have?” |
Getting around and typing
| What it does | Say something like |
|---|---|
| Open a screen: home, My Work, inbox, Mission Mode, search, a project, a card, project settings, billing, members, workspace settings, integrations, the archive, your profile or notifications | “Open the Tarsen board.” / “Show me TR-42.” / “Take me to billing.” |
| Type into the field you have clicked into | “Type this into the description: …” |
Typing is the one action that runs in your browser rather than on our server: it puts text into whatever field has focus, which is nothing you could not do with your own hands. For a comment on a card, voice uses a proper comment instead.
How it finds the card you mean
Nobody reads an id out loud, so you can name a card three ways: by its reference (“TR-18”, or “TR 18”), by its title, or by describing it. A title that matches exactly wins. If more than one card could be the one you mean, Tarsen does not guess. It names up to five and asks which.
What the server checks first
The voice model never touches your data. It can only ask for an action, and every request goes through our server, which checks it before anything changes:
- That you are signed in, and that the conversation is yours, in this workspace, and still running.
- That you can see the project or card. Voice sees only the projects you can see; a private project stays private.
- That you can change it. The same permission checks the board and our MCP server use run again against your account, so view-only access stays view-only.
- That the workspace has room on its plan. If it does not, Tarsen says so in one sentence and offers to archive a project or open billing. It does not try to sell you an upgrade.
Every change is recorded in the card's activity with you as the actor, alongside the sentence you said that caused it, so a teammate can see why a card moved without asking.
What waits for a yes
Speech can be misheard in ways typing cannot. So two actions wait for you to agree out loud: deleting a card, and creating a project. Tarsen reads back which card or what name, and nothing is written until you say yes. This is enforced by the server, not just asked of the model.
There is no undo command. A spoken change is an ordinary change, so you reverse it the ordinary way: move the card back, set the old due date, or say it again correctly. That is why “get rid of” archives a card rather than deleting it. To bring back an archived card, say “open the archive” and restore it from there.
Voice can open your billing page, member list and workspace settings. It cannot change any of them, and it cannot delete a project. There is no tool for any of that.
What is sent where
- Talking to Tarsen always goes to OpenAI's realtime API, whichever text provider is configured. Audio runs straight from your browser to OpenAI, because that is the only way to keep replies fast enough to interrupt.
- Our OpenAI key stays on our server. Your browser is handed a token that expires in ten minutes and does nothing else.
- No audio is ever recorded. A voice transcript is text, readable only by the person who spoke it, and it is deleted automatically after ninety days. There is no admin view of it.
- The one sentence behind each change is kept on the card's activity, where anyone who can see the card can read it.
- When your included minutes are used you add your own OpenAI key, and the same audio goes to the same place, billed to your account instead of ours.
The privacy policy has the full wording, including what applies once your content reaches a model provider.
Minutes on each plan
| Plan | Voice minutes |
|---|---|
| Free | 15 minutes for the whole workspace, once, so you can hear what it does. |
| Pro | 30 minutes a month for each person, pooled across the workspace, then your own OpenAI key. |
| Lifetime | 15 minutes a month for the whole workspace, then your own OpenAI key. |
Monthly minutes reset on the 1st. Free minutes are a one-off and do not come back. Voice is billed by the minute rather than by the question, because realtime audio costs money from the moment you start talking.
Something missing here? Email support@tarsen.co and we’ll write it.
Comments, labels and checklists
Everyone watching a card is notified when voice comments on it, as with any comment. You can only rewrite your own comments; ask it to fix someone else's and it will offer to add a new one instead. Labels are set as a complete set, and only from labels that already exist, so a new one has to be made first.