Talkloom is a menu-bar app that turns speech into text on your own Mac. No cloud, no account, no subscription. Here is exactly what it does and what it needs.
Yes. The speech model runs on your Mac, so recording, recognition and insertion never touch a network. Unplug the Wi-Fi and Talkloom still works. There is no audio upload to intercept, no transcript sitting on a server, and no account to create. Temporary audio is deleted the moment a transcription finishes.
Double-tap either Ctrl key. A sound and a menu-bar dot confirm Talkloom is listening; double-tap again and it types. There is no wake word, no window to switch to and no button to hunt for. If Ctrl clashes with something you use, pick another key or a full shortcut in Settings.
All of them. Talkloom is a menu-bar app rather than a browser extension, so the text lands wherever your cursor already is — ChatGPT and Claude in the browser, Cursor and VS Code, Terminal, Slack, Notes, a form field on any website. Your clipboard is restored afterwards, so nothing you had copied is lost.
A paragraph-length prompt comes back in about a second. The speech model runs on your Mac’s GPU, so there is no upload, no queue behind other people’s requests and no spinner while a server wakes up. Speed depends on your Mac, not on your connection.
More than 99, including Korean, English, Japanese, Chinese, German, French, Spanish and Portuguese. Talkloom detects one language per recording. If you always dictate in the same language, setting it explicitly in Settings gives noticeably better accuracy than leaving detection on.
Yes. Custom vocabulary takes one term per line — colleagues’ names, product names, API calls — and the speech model reads the list before it transcribes, so it writes them out instead of guessing phonetically. The list is capped at 223 tokens, which is all whisper reads; Talkloom counts that budget for you and names anything that did not fit rather than dropping it silently.
Yes. A snippet maps something you say to text you saved — say “my address” and the address is typed. It is a plain text match with no model involved, and it only fires on a whole word, so a trigger sitting inside a longer word is left alone.
Yes. The Files tab takes an audio file and runs it through the same model on the same machine — nothing is uploaded here either. WAV, MP3, FLAC and Ogg are read directly; voice memos and screen recordings (m4a, mov, aac and the rest) are converted on your Mac first. One file at a time, and dictation pauses while it runs.
The recommended model is a 574 MB download, once, on first launch. Smaller models are available for older machines or when you would rather trade a little accuracy for speed, and you can switch between them at any time in Settings.
macOS 13 or newer on Apple silicon — an M1 or later. Talkloom asks for Microphone access to hear you and Accessibility access to type into other apps; both are granted on first launch and can be revoked in System Settings at any time. Intel Macs are not supported.
$39 once, for one Mac. The first ten dictations are free, so you can check it works on your machine before paying. The purchase includes every future update at no extra cost. There is nothing to renew, no subscription to cancel, and a full refund within 14 days.
macOS 13+, Apple silicon · $39 one-time