Google Antigravity Voice Mode Launches

• On 6 November 2026 Google introduced voice mode in the Antigravity App and CLI, enabled by Gemini Live: you speak to the agent like a teammate on a call, steer it while it works and hear spoken replies in real time, while its edits and results still appear in the conversation^[1]
• Voice mode runs on the Gemini Live model, which hears your voice directly rather than a transcript and can be interrupted by simply talking; it can reach only the conversation it was started in, so file, command and tool work happens there under your normal permissions^[1]
• 9to5Google places it in the version 2.29 desktop update, which also brought a redesigned conversation page and cloud Remote Agents for paid Google AI Pro and Ultra subscribers, and it says Voice Mode sits next to the existing microphone button^[1]
• TestingCatalog had spotted the feature in a build on 1 November, describing a live-conversation voice mode with voice choices, speed controls and audio visualisation, apparently built for an upcoming 'Chief of Staff' agent, which shows Google's voice push predates the launch^[3]

Notable features

• Guardrails: the agent reads permission requests aloud with a recommendation and accepts replies like 'allow it just this once', but destructive actions (deleting files, elevated commands, force-pushing) cannot be approved by voice and need an on-screen Allow, though they can be denied by voice^[1]
• It stays quiet by default and speaks only when it has an answer or needs input; say 'keep me posted' for updates about every 30 seconds; choose from nine voices, speed from 0.75x to 2x, and mute with Option+M (macOS) or Alt+M^[1]
• CLI: run /voice live or press F6, with /voice speaker, /voice speed and /voice duplex on for interruption; over SSH, agy mic-serve on your own machine plus the ANTIGRAVITY_MIC variable routes your local microphone and speaker; CLI version 1.5.0 or later is required^[1][4]
• Google's docs separate voice mode from voice dictation, which only types your words into the message box (Ctrl+M in the app, /voice dictate or F5 in the CLI)^[4]