Service eligibility and regional restrictions
PuppyIP serves only compliant overseas businesses and their authorized personnel. Proxy services are not available in mainland China. The service may only be used for lawful business activities outside mainland China. Use of this service within mainland China is prohibited.
Hosting a proxy IP or server overseas does not change these restrictions. The service must not be provided to end users in mainland China through relaying, forwarding, sharing or resale. Before use, read the Terms of Service.
Key Takeaways
- The Agents Window can start a new chat; an IDE chat needs at least one message. When the agent is idle and the composer is empty, click the Start voice chat waveform and allow microphone access.
- Voice Chat is speech-to-speech: Grok Voice handles listening and speaking, while the coding agent keeps using the model selected for that chat. Dictation only puts speech into the composer for you to review and send.
- The call transcript appears in the chat. Live cloud agent transcripts sync to your other devices unless Privacy Mode is on; Cursor says Voice Chat uses zero data retention in that mode.
- By default, a call hangs up after about 10 minutes of silence only when you are silent and the agent has no work in progress. A separate daily call limit exists, but the public docs do not state its exact amount, pricing, or reset time zone.
Start a call from the waveform
Cursor announced Voice Chat with Agents on November 6, 2026, supporting the Agents Window, IDE, web and Cursor for iOS. In the Agents Window, you can open any agent chat, including a new one. In the IDE, the chat must already contain at least one message. Sign in to Cursor before opening the chat you want to use.
In a desktop, IDE, or web chat, wait for the agent to become idle and clear the composer. A waveform button labeled Start voice chat then appears where Send normally sits. Click it and allow microphone access when your operating system or browser asks. If you do not speak right away, the agent checks in after a few seconds. While the agent is working, Stop takes that button's place; wait for the task to finish or stop it before starting a call.
Cursor for iOS is also a supported surface. If the system asks for microphone access, allow Cursor to use the microphone. If access was denied, follow the device prompt to restore it in system settings, then try again.
Voice Chat and dictation are different
Voice Chat is a realtime speech-to-speech conversation: you speak, the voice agent replies out loud, and you can keep talking. When you ask for a code change, the voice agent hands the task to the coding agent in that chat while you continue the conversation.
Dictation turns speech into text in the chat composer. You review and send that text yourself; dictation does not create a spoken agent reply. Use the microphone button for dictation and the neighboring Start voice chat waveform for a call. The IDE docs list Cmd+Shift+Space and Ctrl+Shift+Space as dictation shortcuts; they do not list a Voice Chat shortcut.
Hand off work and check progress during a call
During a call, the voice agent can send a task or follow-up to the coding agent in the current chat and answer questions about the work. It can read diffs, run results, and agent status to explain what changed or what is still running; these lookups are read-only. You can also search earlier messages and calls with that agent.
In a Project with workers, you can move an existing call to another worker. Project worker and subagent chats do not show a Start voice chat button, so start from a chat that has the button and switch workers from within the call.
The coding agent decides what to say: it usually gives short progress notes while work runs and a brief answer when it finishes. Full written replies stay in the chat. It does not promise to read every complete reply aloud; if a turn has no separate spoken update, the call reads that turn's written reply.
Troubleshoot a missing button or failed call
Voice Chat requires a signed-in Cursor account. A session authenticated with an API key cannot start a call. The button may still appear while you are signed out or using an API key, but starting the call can return “Realtime voice requires sign-in.” Sign in with your Cursor account, then start again from an eligible chat.
The button appears only when the agent is idle and the composer is empty; an IDE chat also needs at least one message. The docs list these hidden-button cases: subagents, Project workers, archived chats, Claude Code chats, and chats owned by another teammate. This restriction names Claude Code chats; it does not mean every chat using a Claude model is excluded.
In the IDE, the button is also hidden for cloud agents that are still streaming and when dictation is turned off. If the stated conditions are met and the button is still missing, Cursor recommends updating Cursor. If microphone access was denied, the desktop or browser prompt points to system or browser settings; on iOS, the prompt points to Settings.
Models, voices, and transcript privacy
Voice Chat uses Grok Voice for listening and speech. The coding agent in the chat continues to use the model you selected for it. Cursor lists Ara, Eve, Leo, Rex, and Sal among the available voices. Set a default in Voice or Agent Voice settings, or change it during a call; the in-call choice is saved for that agent. An agent with its own saved voice uses that voice instead of your default.
The call transcript appears in the chat. Live cloud agent transcripts sync to your other devices unless Privacy Mode is on. Cursor describes Voice Chat with Privacy Mode as using zero data retention. The available docs do not explain other data-handling details, so that statement does not establish local-only processing or mean there is no text transcript in the chat.
iOS controls and call disconnects
On iOS, say “Hey Siri, call Cursor” to start a call with a new agent; your iPhone must be unlocked. If iOS cannot show the call on the system call screen, it appears as a Live Activity on the Lock Screen and in the Dynamic Island. You can mute or hang up from the Lock Screen or expanded Dynamic Island; the compact Dynamic Island does not show those controls.
By default, a call hangs up after about 10 minutes of silence from you while the agent has no work in progress. That describes a silence condition, not a maximum call length. Cursor also has a daily call limit and tells you when you can call again if you reach it. The reviewed docs do not give the exact limit, whether calls are charged, or the time zone used to reset the daily limit.
Sources
Frequently Asked Questions
Is Voice Chat included in every plan, country and app version?
The reviewed announcement and documentation list supported surfaces and sign-in requirements, but do not publish a complete plan, country or minimum-version list, or a universal price. Do not apply another Cursor feature’s paid-plan or version requirements to Voice Chat. Check availability in the account and app version being used.
Do I need to configure an xAI API key or choose a Grok Voice model?
These Cursor sources say Grok Voice handles speech and require a signed-in Cursor account. They do not instruct users to configure a separate xAI API key or identify a specific Grok Voice version. Do not apply the standalone xAI developer voice API’s integration, model or billing rules to a Cursor call.