Long architecture descriptions.
Multi-file edits, "build me a feature that does X" prompts. Speak the whole spec — the constraint, the file, the edge case — and send. The composer takes a hundred-word prompt as easily as a twelve-word one.
Hold the hotkey. Speak. Your prompt lands in the composer, the chat, or wherever your cursor sits.
Voca is a system-wide voice-to-text tool. There's no Cursor extension, no per-app config, no setup. Press the hotkey, talk to your editor the way you'd talk to a senior engineer, release. The prompt arrives. Works in Cursor's composer, in Cmd+K, in the chat panel, in the terminal pane, and in every other text field on your computer.
Available on Windows, macOS, and Linux.
No install for the demo, no card required.
A vague twelve-word prompt gets generic code back. A hundred-word prompt that explains the constraint, names the file, and mentions the edge case gets something you can ship. Everybody knows this. The reason it doesn't happen is typing.
Most engineers type at around 40 words a minute. Most engineers speak at around 150. The four-times gap between the two is where prompt quality dies. When typing feels slow, you compress. You drop the constraint, skip the edge case, and send the short prompt that produces the mediocre output. Then you re-prompt twice when one careful prompt would have done it.
Voca closes the gap. You speak the prompt the way you'd describe the problem to a colleague at the whiteboard. Cursor gets the full picture on the first try.
Inside Cursor, that's a few places.
Outside Cursor, the same hotkey works in Slack, Linear, GitHub, your browser, your notes app, and every other text field on the same machine.
Multi-file edits, "build me a feature that does X" prompts. Speak the whole spec — the constraint, the file, the edge case — and send. The composer takes a hundred-word prompt as easily as a twelve-word one.
Hotkey, speak the change you want on the line your cursor is sitting on, release. The prompt drops into Cmd+K and you stay in the file.
The chat panel for the questions that aren't quite a generation request. Voice in, model out, voice in again. The full sentence you'd say out loud, not the half-sentence you'd type.
Commit messages, gh CLI commands, a curl with three flags you can never remember. Voca works in the integrated terminal pane the same way it works in the editor.
The dirty secret of the 2026 dictation market: almost none of it ships for Linux. Wispr Flow is Mac and Windows. Willow is Mac, Windows, and iOS. Voibe, VoiceInk, Superwhisper, Monologue, Weesper — all Mac-only. Cursor itself runs perfectly well on Ubuntu, Pop_OS, Fedora, and Arch. The dictation layer is the missing piece.
Voca ships native Linux builds. Same hotkey, same workflow, same browser fallback if you don't want to install anything at all.
Some dictation tools "improve accuracy" by reading the contents of your active window every few seconds and shipping screenshots to their servers. Useful for a marketing demo. Less useful when the active window is a private repo, a customer's PII, or a half-written security report.
Voca doesn't do that. The hotkey starts the mic. The mic captures your voice. The audio goes to our server, gets transcribed, and is deleted the instant the text comes back. We don't read the screen. We don't keep the audio. We don't train models on it.
Wherever your cursor sits. Composer, chat, Cmd+K, terminal, file editor.
Full sentences, the why, the constraints, the edge case. Speak naturally. Voca handles the cleanup.
0.4 seconds, median. Faster than the model will start replying.
See the full breakdown of Voca vs Wispr Flow.
Three engineers prompting Cursor all day. Seven engineers who barely use voice. Voca is the only major dictation tool with a billing model that handles that shape directly: $6/seat/month for the heavy users on Pro, $0.10/hour pooled credits for the rest. One invoice on the 1st of the month.
Open the browser demo and dictate one prompt. If it lands the way you'd hope, grab the desktop app. Thirty free minutes a month, no card required.
No. Voca operates at the OS level. Anywhere your cursor sits in any application, including all four input surfaces in Cursor, transcribed text appears on release of the hotkey.
Yes. Voca ships native Linux builds for the major distros, and Cursor runs natively on Linux. Most other commercial dictation tools (Wispr Flow, Willow, Voibe) don't ship for Linux at all.
No. Voca does not read, screenshot, or otherwise capture your active window. The microphone captures audio when you hold the hotkey. Nothing else.
No. Audio is deleted as soon as transcription completes. We do not train models on your audio or your text. Transcripts are stored only if you opt in.
Cursor doesn't currently have native voice input across all panels. Voca works system-wide, including in Cursor, but also in everything else you have open.
Yes. The browser demo runs on any modern browser without an account or a card. Useful for testing the latency and accuracy before committing to a desktop install.
Mixed Pro and credit-pool seats, per-member usage caps, and team billing on the 1st of the month. See /for-teams.
Voca is $6/month flat or $0.10/hour pay-as-you-go. Same hotkey on Mac, Windows, and Linux. Also worth a look: Voca for Claude Code.