Cursor Voice Chat for Agents: What It Does, Where It Stops

Cursor shipped Voice Chat with Agents on 6 November 2026: a real-time, two-way voice call with your coding agent, inside the editor.

You describe a change out loud and hear the agent work. It returns short spoken progress notes while it runs, then a brief spoken answer when it finishes — the full written reply stays in the chat. That is a deliberate design choice, and it is the one that decides whether this fits your workflow.

One caveat up front, because it shapes everything below: the reporting this brief draws on is based on Cursor's published changelog and voice-chat docs, checked 7 November 2026, with no hands-on testing. Voice quality, latency and how well the system handles accents or technical vocabulary are not knowable from a changelog.

What actually shipped

Voice chat runs in the Agents Window, the IDE, on web, and in Cursor for iOS. While the coding agent works, you can keep talking — ask what changed in a diff, check on a run, search earlier messages, or switch between workers in a Project.

The speech layer runs on Grok Voice, with Ara, Eve, Leo, Rex and Sal as the named voice options. Two details matter more than the voice list: with Privacy Mode on, voice chat runs with zero data retention, and the coding agent keeps whatever model you already selected. Grok Voice handles the speech, not the code.

Starting it is one button. When the agent is idle and the chat input is empty, a waveform button labelled "Start voice chat" sits where Send usually sits. On iOS there is a hands-free route — "Hey Siri, call Cursor" — and Lock Screen and Dynamic Island controls arrive in iOS 1.16.0.

This is not dictation

The naming invites confusion, so it is worth separating. Dictation — previously called Voice Mode, toggled with Cmd+Shift+Space in the IDE — turns your speech into text in the chat input and stops there. Voice chat is a two-way conversation where the agent answers out loud. Dictation is talking at the keyboard; voice chat is talking to the agent.

The limits, as published

Cursor documented this launch's constraints unusually plainly, and they are the part to read before you plan around the feature:

• API-key sessions cannot start voice chat at all. You need a signed-in Cursor account. If you run Cursor on an API key, the feature does not exist for you.
• There is a daily call limit, and Cursor has not published the number. That is a real gap — you cannot plan around a limit you only discover by hitting it.
• About 10 minutes of silence hangs up the call by default when the agent is idle.
• The start button is hidden in more places than you would guess: subagents and Project workers, archived chats, Claude Code chats, and chats owned by another teammate.
• No new plan price was announced with the launch. Voice chat is a feature inside Cursor as it stands, not a paid add-on.

Is it worth turning on?

The honest split is about how you consume agent output, not about the technology.

If you reason by talking — rubber-duck debugging where the duck edits the code — this targets you directly. So does working away from the keyboard: the Siri shortcut and the Lock Screen controls signal that Cursor sees the best moment for this as the one when you are not at your desk.

If you read every word of every agent reply before deciding, expect to turn it off. Spoken summaries are lossy by design, and a voice telling you the short version is a slower path to information you were going to read anyway. Open-plan offices add a social question on top of the technical one.

Two things to watch. First, the unpublished daily call limit is the single number that decides whether this survives a full workday of real use. Second, this is Cursor's second launch in two days, after Build Live Apps on 5 November — a pattern worth tracking if you are deciding where your team's agent workflow lives.