Hold one key, talk, and clean text appears wherever your cursor is. It runs on your machine, so nothing is uploaded.
Meetings, interviews, voice memos. Drop the file in, or capture the call as it happens — no bot joins, the audio comes straight off your machine.
Windows and macOS. No account.
images/shots/hero.png · 1600 × 1000
Open Dictum on the Home screen with Dictation selected, after two or three real dictations so Recent is not empty. Resize the window to 1600 × 1000, hide any personal text, and capture the window only (Win+Shift+S, or Cmd+Shift+4 then Space on Mac). Do not include the desktop behind it: the frame already draws a border.
It types where you already type
How dictation works
How transcription works
One shortcut, set once, working in every window. Whatever your hands already do is usually the right answer.
Ctrl + Spaceimages/shots/step-key.png · 1400 × 700
Settings → the hotkey row, mid-capture, with the key combination showing. Crop tight to that row and the two lines around it rather than the whole Settings window: the card is wide and short, so a full window shrinks to nothing. 1400 × 700.
Ramble. Change your mind halfway. The ums and the false starts come out on their own, and the punctuation arrives without you saying “comma”.
images/shots/step-talk.png · 1400 × 700
The orb mid-dictation, floating over a half-written message in a real app. The point of this one is that Dictum is on top of something you already had open, so keep the other app clearly visible. Blur anything personal. 1400 × 700.
The text lands in the field your cursor was in. Nothing to copy across, no second window to go back to.
images/shots/step-typed.png · 1400 × 700
The same app as the previous shot, one moment later, with the finished text in the box and the orb gone. Shooting these two back to back from one dictation is what makes the pair read as before and after. 1400 × 700.
Or hit record and let it take the call audio straight off your machine. No bot joins the meeting, and there is nothing to invite.
images/shots/step-drop.png · 1400 × 700
The Home screen in Transcription mode, mid-drag, with the drop state lit and a real audio file over the window. 1400 × 700, window only.
You on one side, everyone else on the other, with timestamps you can click back into.
images/shots/step-voices.png · 1400 × 700
The Transcriptions popup on the Formatted tab, scrolled so several turns of both speakers are visible in their two colours with the player bar in shot. An mp4 works even better here: fifteen seconds of clicking a timestamp and hearing it jump. 1400 × 700.
Copy it, or export to PDF, Word or Markdown.
images/shots/step-export.png · 1400 × 700
The Share popover open over a finished transcript, with the three export formats readable. 1400 × 700, window only.
Your style
Three settings decide how it sounds. All three are yours.
Tell it how you write, app by app. Slack gets two lowercase lines. A client email gets full sentences and a sign-off.
Slack short, no greeting
Gmail full sentences, sign off with my name
images/shots/prompts.png · 1200 × 800
Settings → Personalised Formats, with the Slack tile open in the edit modal so the prompt box and the live preview are both visible. Fill the preview with a sentence a person would really dictate. 1200 × 800, window only.
Names, acronyms, whatever your team invented. Add the word and what it sounds like, and it stops coming back wrong.
Delacroix heard as “della kwa”
keydictum heard as “key dictum”
images/shots/vocabulary.png · 1200 × 800
The Vocabulary page with six or seven real entries — a surname, an acronym, a product name — each with its sounds-like filled in. An empty-state screenshot sells nothing. 1200 × 800, window only.
Say something short, get something long. Signatures, addresses, the reply you send twelve times a week.
“my signature” the whole four-line block
“office address” street, city, VAT number
images/shots/shortcuts.png · 1200 × 800
The Snippets page with one snippet expanded so both the trigger and the block it expands to are readable. Use a real signature with the personal bits swapped out. 1200 × 800, window only.
Privacy
The models sit on your disk. Your audio never becomes an upload.
A plane, a train, hotel WiFi that loads nothing. It does not notice.
No recordings on a server. Your voice trains nothing.
GDPR
Local by default. Cloud audio is processed and dropped in the same request.
There is a cloud mode for when you want it. Prepaid credit, never a subscription.
The code is on GitHub. Open a pull request and it can ship in the next build.
Open the repositoryMost of what is in Dictum started as somebody describing an annoyance.
Ask supportNot out yet. Leave an email and you get one message, on the day it is.
Tell me when it is outWindows 10 and up. macOS 13 and up on Apple Silicon.
