Dictate in any app · Code
Cursor's whole workflow is writing prompts, and prompts are far faster to say than to type. Typally works at the keyboard level, so anywhere Cursor accepts typing, it accepts your voice instead.
From the people who build Typally. If Cursor's own option covers your use, use it — this page will tell you when it does.No dictation of its own. Cursor is built around a chat panel and inline prompts, both of which are ordinary text inputs.
The quality of what an AI editor produces tracks the detail in the prompt. Typing punishes detail — every extra clause costs keystrokes — so people write terse prompts, get generic code, and iterate three more times.
In practice people use it for: Composer prompts describing a feature in full, inline edit instructions, explaining an existing file before asking for a change, and the long context paragraph that saves two rounds of clarification.
Setup happens once, outside Cursor, and then applies to every program on the machine:
Install, sign in, and it waits in the tray.
Whatever has the cursor receives the text, so click the right place in Cursor before starting.
Push to talk, say your piece, release to transcribe.
Cursor sends on Enter in some inputs. If dictation ends with a return character, a half-formed prompt can fire before you finish thinking.
Typally records for as long as you hold the key, so the full requirement — goal, constraints, what to avoid, how you will know it worked — arrives in the Composer box as clean text. Custom vocabulary keeps your file, function and library names intact, which is what usually makes dictated technical prompts unusable.
Instead of typing 'add auth', a developer holds the hotkey and says: what the auth flow should do, which provider, how errors should surface, which existing file to follow the patterns from, and what should not change. The prompt takes forty seconds to speak and lands closer to working code than three rounds of typed one-liners.
Say the whole requirement in one breath: what you want, the constraints, what to avoid, and how you will know it worked. A spoken paragraph costs nothing extra and routinely replaces several typed rounds.
Cursor sends on Enter in some inputs. If dictation ends with a return character, a half-formed prompt can fire before you finish thinking.
Very. Longer prompts with more context produce better output, and speaking makes long prompts cheap. The bottleneck in AI-assisted coding is usually prompt detail.
That depends on the tool. Typally exists because a built-in voice input cut off after about a minute; it records until you release the key.
Yes — it is a normal text field, so anything typing at the system level reaches it.
Start on the free plan (100 minutes a month, no card). Pro lifts the cap and adds meeting notes for $7.99/month.
Also dictate in: VS Code · ChatGPT · Linear · CLion · Notepad++ · see all apps