Sometimes the recording you want to transcribe does not exist yet. You want to explain a code change, draft an email or give a longer prompt to an AI assistant without typing every word. InstantTranscriber Desktop now has a public beta for those short dictations on Mac, Windows and Linux.
Hold a shortcut, speak and release it. The app sends your recording to the InstantTranscriber service and returns plain text. Choose automatic typing, automatic clipboard output or a result window where you click Copy yourself.
A separate tool for short dictation
The desktop app complements the online transcriber. Keep using the website for longer audio/video files, speaker labels, timestamps, subtitle exports and transcript history. Guest file transcription still works without signup. Desktop dictation needs a connected InstantTranscriber account and internet access, and each recording can be up to two minutes.
This beta returns one final result after you release the shortcut. It does not work offline or display a live stream of provisional words. Capacity can be temporarily busy; the app asks you to retry rather than uploading a recording it cannot admit.
Long prompts for coding agents and terminal workflows
When you work with an agent in a terminal, a useful prompt can need much more context than a one-line command: what failed, what you already tried, which constraints matter and how you want the result checked. Dictate that explanation while the terminal is open, review the text and submit it yourself.
Inspect the failing test and explain its cause. Propose a small fix, keep the public interface unchanged, and run the relevant checks before summarizing the result.
The same workflow helps with code-review comments, issue descriptions, reproduction steps, emails and notes. Use the keyboard for exact syntax and identifiers where needed. The app does not press Enter or automatically run a terminal command.
Why processing runs on a server
You need internet, but you do not need to download model weights, configure a local AI runtime or own a GPU. Recognition is managed centrally, so model updates do not require you to maintain a local installation. Accuracy depends on the microphone, language and recording conditions; review names, code identifiers and commands before using the result.
Choose how your words arrive
- Type into the active app: text goes into the original text field. If you change the destination or an app blocks insertion, the result remains available to copy. InstantTranscriber never presses Return automatically, including in a terminal.
- Copy automatically: each result replaces the clipboard so you can paste it yourself.
- Show result and copy manually: nothing is typed or copied on arrival. A small result window stays visible until you click Copy text successfully or explicitly close it. If copying fails, your text remains visible for retry. Copy or close a waiting result before the next dictation.
The existing Insert and copy combination is also available. Closing the small window keeps the last text in the main app until you clear it or quit.
Setup makes permissions explicit
Open the app, choose Connect & get started, then complete sign-in or signup in your browser. The app guides you through microphone setup before your first dictation. Windows and Linux offer an explicit local microphone test; its audio is discarded without being uploaded.
On Mac, microphone access and the optional Accessibility permission for automatic typing are separate. Clipboard and manual copy modes do not need Accessibility. The app checks permission state quietly and only requests access from setup, rather than repeatedly prompting after a dictation.
A managed computer can still block microphones, typing, clipboard access or installation. The app cannot bypass your employer’s policy. Manual copying gives you another output choice, but is not a promise that every managed device permits it.
Hotkeys, sounds and a background app
The default shortcut is Command+Shift+Space on Mac and Ctrl+Alt+Space on Windows/Linux. Preferences let you change it, switch from hold-to-talk to toggle mode, select a microphone or language, and choose start/stop sounds. Login startup is optional. The microphone does not stay open between dictations.
The recording/result window sits above the dock or taskbar without taking focus when it appears. Its settings button opens the main app. On Wayland, shortcut support depends on the desktop portal, and output uses manual paste rather than bypassing compositor restrictions.
Your existing account allowance
There is no separate desktop purchase. Free accounts include 1 audio hour per UTC calendar month. Premium includes 15 hours per billing month, shared across Desktop, API and MCP. The API Plan includes 100 hours. These audio allowances are separate from web-transcription limits.
A successful dictation uses the actual recording duration rounded up to a second, with no one-minute minimum. Failures, cancellations and no-speech jobs use no audio time. Desktop does not use paid API overage. Check the app for your remaining allowance and reset date.
Public beta downloads and installation status
Download the Mac DMG, Windows EXE, Linux DEB or AppImage. The Mac package supports Intel and Apple Silicon. Direct downloads do not require a store purchase.
The current packages are beta builds without a verified publisher signature; the Mac app is not notarized. macOS or Windows may warn or block installation, particularly on managed devices. Verified signing is still pending. Follow your organization’s policy or keep using the web app if installation is restricted.
Quit the old background app before updating, then install the new version. The beta has no automatic updater. The download page includes SHA-256 checksums so the downloaded files can be checked.
Data and feedback
Audio is processed by the online service and removed after processing or expiry. Temporary result recovery lasts up to 15 minutes. There is no local recording archive; the latest result stays in app memory until cleared or the app quits.
Account-bound technical diagnostics include operating system, architecture, version, permission outcomes, errors and timings. These diagnostic events are retained for up to 90 days and do not contain dictated text, audio, clipboard contents or names of other apps. See the privacy notice for more detail.
Try a short, non-sensitive dictation first. If something fails, include your OS and app version when you send feedback. You do not need to share the recording or clipboard contents. Explore Desktop and download the beta.