Skip to content
Remote Control
Voice mode2-0Antigravity AppFeaturesVoice mode Antigravity App FeaturesAntigravity AppAntigravity CLI
Available on:
Antigravity AppAntigravity CLI

Talk with the agent out loud, and hear it answer in real time. With Antigravity voice mode, powered by the Gemini Live model, you speak to Antigravity the way you’d talk to a teammate on a call: give it a task, steer the work while it runs, and answer permission prompts without clicking.

Voice mode in the Antigravity App: ask for a fix out loud, let the agent work, and hear the result. Turn on sound to hear both voices.

Voice mode turns a conversation into a spoken, two-way conversation with the agent. While you talk, the agent does the work in the conversation, so you can read along:

  • Speak naturally: give tasks, ask questions, or think through a problem out loud, the same way you’d type a prompt. Voice mode continues the conversation that you’re in, so the agent has the context of earlier messages, including the ones you typed.
  • Steer while it works: add details, correct course, or stop a task without waiting for it to finish.
  • Answer by voice: allow or deny permission prompts and answer the agent’s questions out loud.
  • Choose what you hear: the agent works quietly by default. Ask for spoken progress updates when you want them.
  • Read along: everything the agent writes, edits, or draws appears in the conversation as usual, and you can follow what you and the agent say while you talk.

The Antigravity App has two ways to use your voice:

Voice modeVoice dictation
What it doesHolds a spoken conversation with the agent.Types what you say into the message box.
How to startClick Start Voice mode in the message box.Click the microphone in the message box, or press Ctrl+M.
How the agent answersOut loud, while its work appears in the conversation.In writing, after you send the message.
When to use itYou want to work hands-free, or talk a task through while the agent works.You want to review and edit a prompt before you send it.

To learn more about dictation, see Voice transcription.

You can start voice mode in a new conversation or in an existing one. In an existing conversation, voice mode continues that conversation by voice.

To start voice mode, do the following:

  1. With the message box empty, click Start Voice mode. It’s the round button with sound bars next to the dictation microphone, and it appears only while the message box is empty. When the agent is idle, it appears in place of Send. While the agent is working, it appears next to the button that stops the agent.

    The message box in the Antigravity App with the bookmarks project and main branch above an empty prompt, the Gemini 4 Argon Medium model and Default permission pickers, the dictation microphone, and the round Start Voice mode button with its tooltip

  2. If the Antigravity App asks for access to your microphone, allow it.

  3. The first time you start voice mode, the Choose Antigravity’s Voice dialog opens and plays a sample of a voice. To hear other voices, click the arrows or press ← and →. To choose the voice that’s playing, click the Use button, which shows the voice’s name, or press Enter.

    The Choose Antigravity's Voice dialog with the subtitle Powered by Gemini Live · You can change this in Settings later, an animated voice orb between previous and next arrows, the voice Fola described as Clear and friendly, nine page dots, and a Use Fola button

  4. Start talking.

When you start voice mode, a panel rises above the message box and shows Connecting… while voice mode connects. When it’s ready, a chime plays. The call bar at the top of the panel shows whose turn it is and how long voice mode has been on, followed by the controls. Until you speak, the panel shows Listening. Ask anything. The conversation also shows a Started voice chat marker with the time.

The voice mode panel above the message box in the Antigravity App. Its call bar shows sound bars next to the label Antigravity, an elapsed time of 0, the Mute mic, Mute Antigravity, and Voice speed buttons, a Minimize chevron, and a red End button, over the text Listening. Ask anything.

Talk to the agent the way you’d talk to a teammate on a call. The agent responds right away to small talk and to questions about something it already said. Anything that needs your code, files, tools, or current information becomes work in the conversation, and you hear the result out loud.

In voice mode, you can do any of the following:

  • Give a task: say what you want, such as “Find the failing test and fix it.” The agent confirms what it’s going to do and starts working.
  • Steer a task: add details or corrections while the agent works, such as “Just check it, don’t change anything yet.” The agent applies the change to the task that’s running.
  • Ask what’s happening: ask “What are you doing?” or “Which step are you on?” to hear the latest progress.
  • Stop a task: say “Stop,” “Cancel that,” or “Never mind.” The agent stops the task that’s running.
  • Pause the agent: say “Wait,” “Hold on,” or “I’m reading” to make the agent stop talking until you speak again. You can also interrupt the agent at any time by talking.
  • Put something on screen: ask the agent to show, print, or draw something, such as a table or a diagram. The agent adds it to the conversation and explains it out loud instead of reading it to you.
  • Hear it word for word: ask the agent to read an answer verbatim, or to walk you through it step by step.
  • Type instead: type a message in the message box and send it. The agent acknowledges it out loud and works on it like a spoken request.
  • Share files: attach a file, such as a screenshot or a log file, to a message. The agent tells you out loud that it received the file.

While you talk, your words appear in the panel above the message box, and they fade a moment after you stop. Then the panel shows the agent’s reply as a lead sentence and short paragraphs, lit up word by word as the agent speaks. What you and the agent say isn’t added to the conversation, but you can open a transcript after voice mode ends. The panel’s transcript can mishear words, but the agent hears your voice directly, so a transcript mistake doesn’t mean that the agent misunderstood you.

Answer permission prompts and questions by voice

Section titled “Answer permission prompts and questions by voice”

When the agent needs your approval for an action, or asks you a question, it reads the request out loud and waits for your answer. You don’t need to click anything.

To answer a permission prompt by voice, do the following:

  1. Listen to the agent’s summary of the action and its recommendation.

  2. Say how you want to answer. For example:

    • Allow once: “Allow it just this once.”
    • Allow for this conversation: “Allow it for this conversation.”
    • Allow for this project: “Allow it for this project.”
    • Allow globally: “Always allow this.”
    • Deny: “No.” To deny and suggest an alternative, say something like “No, run the tests first.”

The agent confirms with a short “Done.” or “Denied.” Some risky actions can be allowed only once. While a permission prompt is waiting, saying “Stop” or “Cancel that” denies it.

To answer a question from the agent, say the option that you want, such as “the first one,” or give your own answer.

By default, the agent works quietly and speaks only when it has an answer or needs your input. The first time you give it a task in voice mode, it tells you that it’s working quietly and that you can ask for updates. When spoken updates are on, the agent gives a short update about every 30 seconds, unless you ask for a different interval.

To change how often you hear progress updates, say something like the following at any time:

ToSay, for example
Turn on updates”Keep me posted” or “Update me.”
Hear more updates”Update me more often” or “Tell me every step.”
Hear fewer updates”Update me less often” or “That’s a bit much.”
Set an interval”Update me once a minute.”
Turn off updates”Work quietly” or “Just tell me when it’s done.”

Spoken updates turn off when voice mode ends, so voice mode always starts quietly.

While voice mode is on, the call bar at the top of the voice mode panel shows the following:

ControlWhat it does
SpeakerShows whose turn it is: You with a microphone while you talk, and Antigravity with sound bars the rest of the time.
Elapsed timeShows how long voice mode has been on.
Mute micMutes your microphone. The button turns red while you’re muted. To unmute, click Unmute mic.
Mute AntigravityMutes the agent’s voice without ending voice mode. The agent’s replies still appear in the panel, so you can read them instead. To hear the agent again, click Unmute Antigravity. The agent’s voice is back on the next time you start voice mode.
Voice speedSets how fast the agent speaks. Click it, and then drag the Voice speed slider to a speed from 0.75x to 2x. The button always shows a gauge; the current speed appears in its tooltip and next to the slider. Your choice applies right away and the next time you use voice mode.
MinimizeCollapses the panel to just the call bar until voice mode ends. To show the whole panel again, click Expand.
EndEnds voice mode.

To mute or unmute your microphone from the keyboard, press Option+M on macOS or Alt+M on Windows and Linux. The shortcut works while you type in the message box, but not while the cursor is in a terminal. While you’re muted, the Antigravity App sends silence instead of your microphone’s audio.

To mute or unmute the agent’s voice from the keyboard, press Shift+Option+M on macOS or Alt+Shift+M on Windows and Linux.

When you switch to a different conversation while voice mode is on, voice mode keeps going, and a floating dock appears in the corner of the window so that you can still control it:

The floating voice mode dock next to the message box: a pencil button, a glowing orb with the tooltip Back to Project Tag Filter Explained, a microphone button, and a red button that ends voice mode

From the dock, you can do the following:

  • Click the pencil to type a message or add files for the conversation that you’re talking in.
  • Click the glowing orb to go back to that conversation. Its tooltip shows the conversation’s title.
  • Mute your microphone, or click × to end voice mode.

To move the dock, drag it. You can also press Tab until the orb is selected, and then press Option and an arrow key on macOS, or Alt and an arrow key on Windows and Linux.

Voice mode can be on in only one conversation at a time. While it’s on, Start Voice mode doesn’t appear in other conversations.

You can end voice mode in either of these ways:

  • Click End in the call bar, or × in the floating dock.
  • Say “Bye” or “End the call.”

Voice mode also ends when you close or reload the Antigravity App window. Any task that the agent is still working on keeps running in the conversation after voice mode ends.

While voice mode is on, the window that you’re talking in doesn’t play notification sounds, so they don’t interrupt the agent’s voice.

When voice mode ends, the conversation shows a Stopped voice chat marker with how long voice mode was on, and a Voice transcript card:

In the conversation, a Stopped voice chat · 2m 02s divider above a Voice transcript card that shows 9 AM · 2m 02s and an Open button

To read what you and the agent said, click Open on the card. The Voice transcript pane opens next to the conversation. To copy the whole transcript, click Copy transcript at the top of the pane.

The Voice transcript pane: the spoken request Hey, so the tag filter isn't clearing when I hit clear. Can you fix it and just let me know when it's done? at 9 AM, the agent's reply I'll check into that code right now., a Worked for 1m row, and the spoken result, with a Copy transcript button at the top

The transcript is saved on your computer, and it’s deleted when you delete the conversation.

Voice settings are in Settings > General > Voice mode:

The Voice mode section in Settings, with a Voice row described as The voice used in Voice mode. Powered by Gemini Live. that shows Arlo and a Change button, a Microphone row set to System default, and a Volume row described as How loud Antigravity's voice is. with a slider at 100%

To change your voice settings, do the following:

  1. Open Settings: press Cmd+, on macOS or Ctrl+, on Windows and Linux, or click Settings at the bottom of the left sidebar. Then go to General.

  2. In the Voice mode section, do any of the following:

    • To change the voice, click Change next to Voice, listen to the voices, and then click the Use button for the voice that you want.
    • To choose an input device, select it in the Microphone list. The default is System default. Voice dictation uses the same microphone.
    • To change how loud the agent’s voice is, drag the Volume slider.

To change the voice speed, use the controls while voice mode is on. You can also ask the agent in voice mode how to change your voice settings.

ActionmacOSWindows and Linux
Mute or unmute your microphone in voice modeOption+MAlt+M
Mute or unmute the agent’s voice in voice modeShift+Option+MAlt+Shift+M
Hear the previous or next voice in the voice dialog← or →← or →
Use the voice that’s playing in the voice dialogEnterEnter
Move the dock, with its orb selectedOption+arrow keyAlt+arrow key
Open SettingsCmd+,Ctrl+,

In Settings > Shortcuts, the mute shortcuts are listed as Mute Voice Mic and Mute Voice Speaker under Conversation.

  • Your voice: Antigravity streams your microphone’s audio to Google servers, where the Gemini Live model recognizes what you say and generates the agent’s spoken replies. When the agent works on something that you asked for, the request includes the audio of what you said.
  • Access: in voice mode, the Gemini Live model can reach only the conversation that you started voice mode in. Work on your files, commands, and tools happens in the conversation, under your usual permissions.
  • Transcripts: in the Antigravity App, what you and the agent say appears in the voice mode panel while voice mode is on, and in the Voice transcript card, saved on your computer, after it ends. In the Antigravity CLI, spoken lines appear in the conversation. The agent’s work and written responses stay in the conversation.

Voice mode has the following limitations:

  • In the Antigravity App, voice mode is available only in local conversations. It isn’t available in Remote Control sessions or in conversations that run in the cloud.
  • Voice mode can be on in only one conversation at a time.
  • You can’t start voice mode in a subagent’s conversation or in a side question.
  • In the Antigravity CLI, voice mode isn’t available in print mode or in sessions with a remote agent.
  • Start Voice mode appears only when you can send a message. For example, it doesn’t appear when you’re signed out or you’ve reached your usage limit.

If voice mode doesn’t work as expected, try the following solutions:

  • Start Voice mode doesn’t appear: clear the message box, including any attached files, and make sure that voice mode isn’t on in another conversation. Check that you’re signed in and that you haven’t reached your usage limit. Voice mode might also not be available for your account yet.
  • “Microphone blocked. Allow mic access.”: the Antigravity App can’t use your microphone. Allow microphone access for the Antigravity App in your operating system’s privacy settings, and then start voice mode again:
    • macOS: go to System Settings > Privacy & Security > Microphone.
    • Windows: go to Settings > Privacy & security > Microphone, and make sure that Let desktop apps access your microphone is on.
  • No audio from your microphone: if No audio from followed by your microphone’s name, or No mic audio detected, appears above the voice mode panel, the Antigravity App isn’t getting any sound from your microphone. Click Switch mic and choose another microphone, or choose one in Settings > General > Voice mode > Microphone.
  • “Couldn’t connect” or “Disconnected”: click Try again. If voice mode still can’t connect, it might not be available for your account yet.
  • “Usage limit reached”: you’ve reached a usage limit for your plan. To learn about limits, see Plans.

Explore related documentation: