Nov 7, 2026

How to Use Cursor Voice Chat with Agents: Start a Call, Talk While It Codes, Fix Missing Buttons

You can now talk to a Cursor agent out loud — describe a change, ask what it is doing, and hear short progress updates — without typing every prompt. On November 6, 2026 Cursor launched Voice Chat with Agents: a realtime two-way call that works in the Agents Window, the IDE, on the web, and in Cursor for iOS. This guide is for people who already use Cursor (or are about to) and want a spoken workflow today. You will use buttons, microphone permissions, and Voice settings — not scripts or APIs.

Everything below was checked on November 7, 2026 (Asia/Shanghai) against Cursor’s Voice Chat changelog and the Voice chat docs. Menu labels change; those pages are the source of truth.

Key facts (from Cursor):

What it is: A realtime spoken conversation with an agent. You talk; the agent talks back; coding work can be handed off without typing (docs; changelog).

Where it works: Agents Window, IDE, web, and Cursor for iOS (docs).

Not the same as dictation: Dictation (earlier releases called this Voice Mode) puts text into the chat input. Voice chat is a two-way call with spoken replies (docs).

Voice stack: Listening and speaking run on Grok Voice. The coding agent in the chat keeps using the model you already picked for it (docs).

Sign-in required: Sessions authenticated only with an API key cannot start voice chat. You need a signed-in Cursor account (docs).

Privacy Mode: With Privacy Mode on, voice chat runs with zero data retention (docs; changelog).

Limits: By default a call hangs up after about 10 minutes of silence from you while the agent has no work in progress. There is also a daily limit on voice calls; Cursor tells you when you can call again if you hit it (docs).

Short version

Update Cursor so you have the Voice Chat release, then open an agent chat (Agents Window, IDE, web, or iOS).
Clear the chat input and wait until the agent is idle. A waveform button labeled Start voice chat appears where Send usually sits — next to the separate microphone button for dictation.
Click Start voice chat, allow the microphone if asked, and start talking (or wait a few seconds for the agent to check in).
Ask for work out loud. The voice agent hands coding tasks to the coding agent in that chat and keeps talking with you while the work runs.
If the waveform button is missing, clear typed text, wait for the agent to finish, avoid subagents / archived chats / Claude Code chats / someone else’s chat, and in the IDE make sure the chat already has at least one message.

Who this is for

This guide is for people who use Cursor through the app, web, or iPhone — not for building custom voice APIs.

You should read this if:

You want to describe changes out loud while an agent works, instead of typing every follow-up.
You keep confusing the new voice call with the older dictation / Voice Mode microphone.
You use Cursor for iOS and want Hey Siri, call Cursor, Lock Screen mute/hang-up, or Dynamic Island controls.
You care about Privacy Mode and zero data retention on voice calls.
The Start voice chat button never appears, or the call fails with Realtime voice requires sign-in.

If you mainly want Cursor to build and publish a small web app from a description, that is a different Nov 2026 feature — see How to Build Live Apps in Cursor. Voice chat is about talking to the agent, not about hosting.

Before you start: checklist

Update Cursor on desktop (and the iOS app if you use it). The changelog says to update Cursor to try voice chat (changelog).
Sign in with your Cursor account. API-key-only sessions cannot start voice chat (docs).
Pick a surface you will use: Agents Window, IDE chat, cursor.com web agents, or Cursor for iOS.
Have a microphone the OS or browser can use. You will allow access when Cursor asks.
Optional: Turn on Privacy Mode first if you want voice chat with zero data retention (docs).

Voice chat vs dictation (do not mix them up)

Cursor documents two different features that both involve a microphone (docs):

	Voice chat	Dictation
Button	Waveform — Start voice chat	Microphone — Start voice input (Agents Window), Voice Input (IDE), Record voice (iOS)
What you get	A spoken reply from the agent	Text in the chat input
Who sends the prompt	The voice agent can hand off work for you	You review the text and send it
Shortcut	None listed for starting a call	`Cmd+Shift+Space` / `Ctrl+Shift+Space` in the IDE; hold `Ctrl+M` to talk in the Agents Window

Earlier releases called dictation Voice Mode. That name still shows up in older tips and muscle memory. In the IDE, Cmd+Shift+Space / Ctrl+Shift+Space still toggles dictation, not the new call (docs; changelog).

Practical rule: if you only want text in the box, use the microphone / Voice Input. If you want the agent to answer out loud and keep a conversation going while code runs, use the waveform Start voice chat button.

How to start a voice chat (desktop, web, Agents Window)

Follow Cursor’s documented steps (docs):

1. Open an agent chat

Agents Window: open any chat, including a new one.
IDE: open a chat that already has at least one message. A brand-new empty IDE chat will not show the waveform button yet.

2. Clear the chat input and wait for idle

The waveform button appears only when:

the agent is idle, and
the chat input is empty.

While the agent is running, Stop takes that spot instead of Send / Start voice chat (docs).

3. Click Start voice chat

Click the waveform button labeled Start voice chat. It sits where Send usually is, next to the microphone button for dictation (changelog; docs).

4. Allow microphone access

Allow microphone access if your computer, browser, or phone asks. The call connects. Start talking, or wait a few seconds — the agent will check in (docs).

If you previously denied the mic, Cursor shows:

Desktop/browser: “Microphone blocked. Allow it in your system or browser settings.”
iOS: “Microphone blocked. Allow it in Settings.”

Turn microphone access back on there, then start the call again (docs).

What you can do during a call

Voice chat is speech-to-speech. You speak to a voice agent; it replies in a natural voice while you keep the conversation going. When you ask for code changes, that voice agent hands the work to the coding agent in the same chat and keeps talking with you while the work runs (docs; changelog).

During a call, the voice agent can (docs):

Hand off work — send a task or follow-up to the coding agent in the current chat.
Answer questions about your work — read diffs, runs, and agents to tell you what changed and what is still running. These lookups are read-only.
Search this agent’s history — search earlier messages and calls with this agent.
Switch Project workers — in a Project with workers, move the call to a different worker.
Hang up — end the call when you say you are done.

The coding agent decides what is worth saying out loud. Cursor says it shares:

short progress notes while it works,
a brief answer when it finishes,
an update when workers it started report back.

If the coding agent does not write a spoken update for a turn, the call reads its written reply instead. Full written replies still land in the chat; you usually hear a short update rather than the entire text (docs; changelog).

Transcripts: the call transcript appears in the chat. Cloud agent transcripts sync to your other devices while the call is live, unless Privacy Mode is on (docs).

You can keep working while you are on a call — the coding agent keeps running and speaks short updates when there is something to report (docs FAQ).

Example things to say (plain language)

These are example prompts for a spoken workflow. Adjust them to your project; Cursor does not publish a fixed phrase book.

“Summarize what you changed in the last diff.”
“Add a login form to the settings page and tell me when the preview is ready.”
“Is the cloud agent still running, or did it finish?”
“Search earlier in this chat for the migration plan we discussed.”
“Switch me to the other worker in this Project.”
“We’re done — hang up.”

For longer agent workflows that touch browsers you are already signed into, pair voice chat with careful permission settings — see How to Let AI Agents Use Your Logged-In Chrome. For connecting Cursor to IFTTT-style automations over MCP, see How to Connect IFTTT to Claude, ChatGPT and Cursor.

Grok Voice, named voices, and which model codes

Listening and speaking run on Grok Voice. The coding agent in the chat keeps using whatever model you already selected for that agent (docs; changelog).

Pick your default voice in Voice or Agent Voice settings. Cursor lists many named voices, including Ara, Eve, Leo, Rex, and Sal. You can also change the voice during a call; that choice sticks to the agent you are talking to. An agent with its own saved voice uses that instead of your default (docs; changelog).

Gotcha: changing the spoken voice does not change which coding model edits your files. If you want a different coding model, change it the same way you already do in Cursor agent settings (for model-oriented setup notes, see also Cursor Composer 2.5).

Requirements (official list)

Cursor’s requirements for voice chat (docs):

A supported surface — Agents Window, IDE, web, or the Cursor iOS app.
A signed-in account — sign in to Cursor with your account. Sessions authenticated with an API key cannot start voice chat.
Microphone permission — allow Cursor to use your microphone when the OS or browser asks; fix blocks in system/browser/iOS Settings if you denied it earlier.

There is no separate “buy Voice Chat” product line in the docs page above — the hard blockers called out are surface, sign-in (not API key), and microphone.

iOS: Siri, Live Activity, Lock Screen

Cursor for iOS adds a few call extras (docs; changelog):

Siri: say “Hey Siri, call Cursor” to start a call with a new agent. Your iPhone needs to be unlocked.
Live Activity: when iOS cannot show the call on the system call screen, the call appears on the Lock Screen and in the Dynamic Island.
Mute / hang up: use the Lock Screen or the expanded Dynamic Island. The compact Dynamic Island does not show those controls.

Update the iOS app along with desktop so the waveform / Siri entry points match the Nov 6 release notes.

Call limits and Privacy Mode

Silence hang-up

By default, a call hangs up after about 10 minutes of silence from you while the agent has no work in progress (docs). If you step away mid-task, either leave work running or expect to start a new call.

Daily limit

There is a daily limit on voice calls. Cursor does not publish the exact number on the Voice chat docs page; if you reach the limit, Cursor tells you when you can call again (docs). Plan important demos earlier in the day if you already burned through calls.

Privacy Mode

With Privacy Mode on, voice chat runs with zero data retention (docs; changelog). Cloud agent transcript sync to other devices while a call is live is also suppressed when Privacy Mode is on (docs).

If your team requires Privacy Mode for other Cursor features, turn it on before you rely on voice for sensitive repos.

Troubleshooting FAQ (from Cursor’s docs)

Why don’t I see the Start voice chat button?

Cursor’s FAQ (docs):

The waveform button shows only when the agent is idle and the chat input is empty. Clear typed text, or wait for the agent to finish.
In the IDE, the chat also needs at least one message.
The button is hidden in:
- Subagents and Project workers
- Archived chats
- Claude Code chats
- Chats owned by another teammate
In the IDE, it is also hidden for cloud agents that are still streaming and when dictation is turned off.
If none of these apply, update Cursor.

The Start voice chat button can still show when you are signed out or using an API key, but the call itself needs a signed-in Cursor account. Sign in with your account, then start the call again (docs).

Does voice chat read every reply aloud?

Usually not. The coding agent chooses short spoken updates; full replies stay in the chat. When it does not write a spoken update for a turn, the call reads the written reply (docs).

Can I keep working while I’m on a call?

Yes. The coding agent keeps running while you talk and speaks short updates when there is something to report (docs).

I clicked the microphone and only got text in the box

That is dictation, not voice chat. Use the waveform Start voice chat button instead (docs).

Suggested first-session walkthrough

Use this as a 10-minute practice loop after you update Cursor:

Sign in on desktop. Open the Agents Window and start a new chat (or open an IDE chat that already has one message).
Type nothing in the input. Confirm the agent is idle so the waveform appears.
Click Start voice chat, allow the mic, and say: “Don’t change any files yet. Just tell me what this project looks like from the open files.”
Ask a small, safe follow-up: “Explain the last diff in one sentence.” Confirm you hear a short spoken update and see the transcript in chat.
Optionally open Voice / Agent Voice settings and try another named voice (Ara, Eve, Leo, Rex, or Sal).
Hang up by saying you are done (or use the UI hang-up control).
On iPhone (optional): unlock the phone, say “Hey Siri, call Cursor”, mute from the Lock Screen or expanded Dynamic Island, then hang up.

If that loop works, you can move on to real coding prompts — still reviewing diffs the same way you would after typed agent runs.

How this fits with other Cursor workflows

Live Apps (Nov 5, 2026): describe an internal tool and publish a link — covered in How to Build Live Apps in Cursor. Voice chat can sit on top of that workflow (talk while a cloud agent builds), but Apps has its own publish and sharing rules.
Browser automation: if the agent needs your already-logged-in Chrome, set that up separately with clear permissions — logged-in Chrome guide.
IFTTT / MCP automations: connect Cursor to external triggers carefully — IFTTT MCP setup.
Model choice for coding: voice uses Grok Voice for speech; coding still uses your selected agent model — Composer 2.5 notes if you are comparing coding models.

What we are not claiming

Exact daily call quotas (Cursor says a daily limit exists and will tell you when you can call again; the Voice chat docs page does not list a number).
Pricing tiers required for voice beyond the documented sign-in / surface / microphone requirements.
That every spoken sentence will be read aloud (Cursor says short updates are the default).
Behavior of third-party “Voice Mode” tutorials written before Nov 6, 2026 — many of those describe dictation, not this call feature.

Sources

Cursor changelog: Voice Chat with Agents (Nov 6, 2026)
Cursor docs: Voice chat