Voice input in Cursor: built-in dictation and Potok

Potok transcribes Russian speech; this site has been translated into English for convenience.

Cursor has had built-in dictation since version 2.0: the Ctrl+M key starts voice control of the agent, and in 3.1 recognition became more accurate because the whole recording is processed at once. It works only for commands to the agent. You can dictate text into any other Cursor field, such as code, a comment or an ordinary prompt, with the Potok app (potok.now), which recognizes about 900 characters per minute.

Did you open Cursor to explain a task to the agent by voice instead of typing a long prompt? Since version 2.0, Cursor has its own dictation button, but it is designed for one specific scenario, not for any text field of the program. Voice input in Claude Code is a similar story.

Voice input in Cursor: what the agent's built-in dictation can do

Cursor is a code editor with an AI agent for Windows and macOS. Built-in dictation appeared in version 2.0: the Ctrl+M shortcut starts voice control of the agent through built-in speech recognition. During recording, the screen shows a sound wave, a timer and cancel and confirm buttons, and the dictated phrase does not go to the agent until you confirm it. In version 3.1 recognition became more accurate: instead of parsing speech in pieces, Cursor processes the whole voice recording at once, so the final text is more reliable. But this dictation has a narrow task: it turns voice into a command for the agent, not into text for an arbitrary field of the program. You cannot type code, a comment on a line or an ordinary text prompt by voice through Ctrl+M, because the built-in feature has no separate button for that. The detailed list of changes in version 3.1 is in the official Cursor changelog.

How Potok differs from Cursor's dictation

Potok works not as a separate voice command to the agent but as ordinary text input: hold the hotkey (Ctrl+Space by default, different from Cursor's Ctrl+M), say a phrase, release, and the text lands where the cursor is. It can be a line of code, a comment, an ordinary prompt field or the same agent window: for Potok there is no difference, it simply inserts the finished text into the active field. Punctuation and capital letters are placed automatically, and you do not need to dictate "zapyataya" (comma) or "tochka" (period) separately. Recognition is in the cloud, with a model trained for Russian speech, including terms and mixed Russian and English text, which is often needed when describing code. Potok's speed is about 900 characters per minute by voice against 200 when typing. An overview of how to dictate text in other programs is in the section about apps.

When to do without Potok

If you only need to quickly explain a task to the agent by voice (what to rewrite, which function to add, which error to fix), Cursor's built-in dictation via Ctrl+M is enough: it is already set up and works right after updating to version 2.0 or newer. Potok is needed when you want to type the code itself, comments or a long text prompt by voice into any other field of the program. Keep in mind here: Potok is a cloud service, recognition runs on servers in Russia, and it does not work without the internet. It has no voice commands other than dictating the text itself.

Frequently asked questions

Built-in dictation and Potok

Primary sources: cursor.com · cursor.com · cursor.com