Cursor's Nov 6 update adds realtime voice calls with coding agents in the IDE, Agents Window, web and iOS, powered by Grok Voice. How it works, the limits, and who it's for.
On November 6, 2026, Cursor published Voice Chat with Agents ↗: a two-way, realtime voice call with the agent in your chat. You describe a change out loud, ask what the agent is doing, and hear short progress updates while it works. Voice chat works in the Agents Window, the IDE, on the web and in Cursor for iOS, and the docs ↗ went live at the same time. Data Studios ↗ covered it in detail on November 6.
It landed one day after Build Live Apps. Together, the two updates push Cursor further toward supervising agents and away from typing at them.
Voice chat isn't dictation. Cursor's old "Voice Mode" was renamed dictation and still turns your speech into text for you to review and send. Voice chat is a conversation: a voice agent listens and talks back, and it hands coding work to the coding agent in that chat.
During a call, according to the docs, the voice agent can:
Hand off work. It sends a task or follow-up to the coding agent.
Answer questions about your work. It reads diffs, runs and agents to tell you what changed. These lookups are read-only.
Search this agent's history, across earlier messages and earlier calls.
Switch Project workers in a Project that has them.
Hang up when you say you're done.
The coding agent decides what's worth saying aloud. It shares short progress notes, gives a brief answer when it's done, and reports updates from workers it started. The full call transcript lands in the chat.
To start a call, make sure the agent is idle and the chat input is empty, then click the waveform button labeled Start voice chat where Send normally sits. On iPhone you can say "Hey Siri, call Cursor" to start a call with a new agent. The phone has to be unlocked, and you can mute or hang up from a Lock Screen Live Activity or the expanded Dynamic Island.
The voice runs on Grok Voice. It handles listening and speaking, while the coding agent keeps whatever model you picked for it. You can choose from named voices including Ara, Eve, Leo, Rex and Sal, and change voice mid-call. The new voice sticks to that agent.
With Privacy Mode on, voice chat runs with zero data retention, per Cursor. With it off, cloud agent transcripts sync to your other devices while the call is live.
The docs are clear about the edges:
| Limit | What Cursor says |
|---|---|
| Account | Signed-in account required; API-key sessions can't call |
| Silence | Call hangs up after about 10 minutes of silence with no work running |
| Daily cap | A daily limit exists; the number isn't published |
| Not available in | Subagents, Project workers, archived chats, Claude Code chats, teammates' chats |
Cursor doesn't list plan restrictions or a separate price for voice chat. The instruction is simply "Update Cursor to try voice chat."
Early users mostly liked how it feels:
"Btw the real winner 🏆 is the voice chat. Super impressed! Feels really like talking to a smart person."
The more useful reactions were about workflow. "The useful part isn't dictating a longer prompt: it's asking questions while the coding agent keeps working," wrote @X_The_Burner ↗, who added a fair warning: "A confident voice isn't evidence that the code is right." Developer @webtkdev ↗ flagged the first trap new users hit: "Most people hit the mic and wonder why the agent stays silent." That's because the microphone button is still dictation, and voice chat is the waveform next to it. Several developer accounts, including @aiwithshriom ↗, posted how-to threads by the next day.
This is the right feature for the way agents are actually used now. Long cloud-agent runs mean you spend more time checking on work than writing prompts. Asking "what changed in that diff?" out loud from your phone, while the agent keeps running, is faster than opening a laptop and scrolling a transcript. The read-only lookups are a good safety default: the voice agent can tell you about your code, but code changes still go through the coding agent you already trust.
Try it if:
You run long agent tasks or Projects with several workers and want to check progress without context-switching.
You're on iOS. The Siri call and the Lock Screen controls are the most distinctive part.
Hold off if:
Your prompts need precise wording (file paths, constraints, exact refactors). Dictation, where you review before sending, is still better.
You work under strict data rules. Turn on Privacy Mode first, because spoken asides end up in the transcript like anything you type.
You live in Claude Code chats inside Cursor. The button is hidden there.
Compared with alternatives, OpenAI added ChatGPT Voice to Work and Codex in the desktop app ↗ in July, so voice-driven agents aren't unique to Cursor. Cursor's edge is the hand-off to the coding agent and the mobile call flow. For how Cursor stacks up overall, see our GitHub Copilot vs Cursor comparison, our Claude Code review and the best AI for coding 2026 ranking.
Cursor changelog: Voice Chat with Agents ↗, November 6, 2026
Cursor docs: Voice chat ↗, checked November 8, 2026
Data Studios: Cursor adds real-time voice chat with coding agents ↗, November 6, 2026
OpenAI: ChatGPT release notes ↗ (July 23, 2026 entry), checked November 8, 2026