Google's Antigravity Lets Developers Speak Directly to Coding Agents

Google is gradually rolling out Voice Mode for Antigravity, its agentic coding platform. The feature lets developers speak requests, redirect active tasks, answer permission prompts, and hear responses during a continuous session. Google explains the workflow in its Voice Mode documentation .

Gemini Live powers the exchange. Antigravity streams microphone audio to Google’s servers, where the model recognizes speech and generates spoken replies while the coding agent works. The conversation also appears as text and remains available as a transcript after the voice session ends.

Release 2.29 at a glance

Voice Mode arrived with Antigravity v2.29.0 on macOS, Windows, and Linux, according to release coverage . The same release adds Remote Agents, reorganizes Automations into Scheduled, Always on, and Paused lists, and introduces a native Jupyter notebook viewer.

Because Google is staging the rollout, an updated installation may still lack the voice button until the feature reaches that account.

Speak, steer, and approve

A live voice session supports the main interactions developers use to supervise an agent:

• Assign a new coding or research task.
• Refine a plan while the agent is working.
• Redirect a code change during execution.
• Approve or deny permission requests.
• Interrupt a response to provide new instructions.
• Switch languages during the conversation.
• End the session by saying “bye” or selecting the end control.

Antigravity continues showing the agent’s work in the conversation, which allows developers to follow commands, file changes, and results in text. Spoken approval carries the same consequences as clicking an approval control, so the requested action and its scope still require review.

Dictation ends at the prompt box

Antigravity’s existing dictation feature converts speech into editable text and waits for the developer to send it. Voice Mode maintains a two-way audio session, accepts interruptions, speaks the agent’s responses, and continues operating while the agent executes a task.

Controls in the app and CLI

Google provides several speaker voices and playback speeds from 0.75x to 2x. CLI users can configure the session with these commands:

• /voice speaker selects the speaking voice.
• /voice duplex configures turn-taking and interruption behavior.
• /voice volume changes output volume.
• /voice speed adjusts playback speed.

Your audio goes to Google

Antigravity sends microphone audio to Google’s servers for Gemini Live processing. The model’s context is limited to the conversation in which Voice Mode was started, rather than the developer’s entire Antigravity workspace.

Teams handling proprietary code or regulated data should evaluate microphone access, server-side audio processing, transcript handling, and organizational policies before enabling the feature.

Remote agents follow a separate path

Antigravity 2.29 also introduces Remote Agents, which run sessions and tools inside an isolated, Google-managed environment. Those sessions can continue after a developer closes a laptop or browser and can be accessed across desktop and mobile devices.

The desktop app’s local-only restriction keeps current Voice Mode sessions separate from general remote-agent control. Developers should therefore avoid assuming that every remote session can be managed by voice through the desktop interface.

Hands-free supervision

Voice Mode fits workflows centered on supervising long-running tasks, including reviewing an agent’s plan, answering tool permissions during a refactor, redirecting a test run, or starting work while away from the keyboard. It offers less advantage during precise, single-file editing, where typed commands provide faster control over symbols, paths, and code fragments.