Cursor Voice Chat: How It Works and What Cursor Hasn't Said

Cursor Voice Chat: How It Works and What Cursor Hasn't Said

Cursor voice chat shipped on November 6, and it isn't the dictation feature you already know. According to Cursor's changelog, you can now hold a live, two-way voice call with your agents. It works in the Agents Window, the IDE, on the web and in Cursor for iOS.

You talk. The agent talks back. Meanwhile, a coding agent does the actual work.

That's a real change for anyone who's asked on Cursor's forum for a way to talk to a Cursor agent and hear it answer. Here's how it works under the hood, where it fits in a real day, and the gaps Cursor hasn't filled in yet.

We read the changelog and the voice chat docs. We haven't run long calls ourselves, so nothing below is a latency or quality verdict.

How Cursor Voice Chat Works: Technical Breakdown

There are two agents in the loop, not one. You talk to a voice agent. Per the docs, it can send a task or follow-up to the coding agent in the current chat, then keep talking with you while that agent works.

The coding agent doesn't read everything aloud. It shares short progress notes while it works and a brief answer when it's done.

Cursor voice chat runs on Grok Voice. Cursor's docs say so in one line and nothing more. You get many named voices, including Ara, Eve, Leo, Rex and Sal, and you can switch between them mid-call.

Starting a call is picky. The waveform Start voice chat button sits next to Send. It only shows up when the agent is idle and the chat input is empty, so you can't call an agent that's mid-run.

So the setup looks like this:

1. Sign in to Cursor. Being on the latest version helps.
2. Open an agent chat. In the IDE, it needs at least one message first.
3. Clear the input box and wait for the agent to go idle.
4. Click the waveform button and allow microphone access.

Calls end on their own. A call hangs up after about 10 minutes of silence from you. That timer only applies while the agent has no work in progress.

Cursor Voice Chat vs Dictation

This is where people will get confused. Cursor's old Voice Mode is now called Dictation, and it hasn't gone anywhere. Cmd+Shift+Space still toggles it.

Voice mode never talked back. Agents now do.

Here's our rule of thumb. If you know exactly what you want, dictate it. If you're thinking out loud, call.

Hey Siri, Call Cursor: The iOS Flow

On iPhone, saying Hey Siri, call Cursor starts a call with a new agent. A Live Activity on the Lock Screen lets you mute or hang up.

That's the part that could actually change habits. Your phone becomes a way to start agent work when you're nowhere near a keyboard.

Real-World Use Cases

Steering a long refactor. Start the agent, then send spoken follow-ups instead of typing them between runs.

Rubber-duck debugging that answers back. Explain the bug out loud. The voice agent can turn what you said into a task for the coding agent.

Away-from-desk triage. Say Hey Siri, call Cursor to start a new agent on a failing test. Then mute or hang up from the Lock Screen.

Easier on sore wrists. For developers with RSI, a full voice loop beats dictating a prompt and clicking Send.

Hard Limitations and Flaws

There's a daily cap, and no number. The docs say there's a daily limit on voice calls. They don't say what it is, or whether it differs by plan.

Pricing isn't stated. Neither the changelog nor the docs say which plans get voice chat, or whether calls use your agent quota. Sign in to Cursor is the only requirement listed.

The data path is fuzzy. With Privacy Mode on, voice chat runs with zero data retention. But the docs don't say where audio is processed or whether xAI sees it. Check with Cursor before you turn it on for company code.

Transcripts travel. During a call, cloud agent transcripts sync to your other devices unless Privacy Mode is on. That's handy on a personal laptop, less so on a shared screen.

Summaries hide detail. Spoken progress notes are short by design. You'll still need to read the diff before you merge anything.

Verdict: Your Developer Action Plan

Cursor voice chat turns talking to a coding agent into an actual conversation. The two-agent split is the smart bit: one agent talks while the other codes.

But it's a v1 with blank spots. Here's what we'd do this week:

1. Run one test call on a low-stakes task. Use it to answer three things the docs don't:

• Handoff fidelity. Ask for a specific change, like renaming a function across two files. Then check the chat. Did the task the coding agent got match what you said?
• Mid-run follow-ups. Change your mind out loud while it works. See whether the follow-up reaches the coding agent right away or waits.
• Your cap. Count your calls until you hit the limit. Right now that's the only way to learn the number.

1. Turn on Privacy Mode first if you work on client or company code.
2. Ask your Cursor admin or account rep three things: the daily call limit, which plans include it, and where audio is processed.
3. Keep Dictation for exact prompts. Use voice chat for steering, not for spelling out file paths.
4. Read every diff. A short spoken summary doesn't count as code review.

If you code with your hands full, or typing hurts, try it today. Everyone else can wait for Cursor to publish the limits.