Whisper without the command line

The same local model — without the terminal

Whisper is a model, not a program: Python, a command, a text file at the end. Here that recognition lives inside an app — one shortcut to dictate, a window to drop files into, subtitles over any video.

No Python Runs on your GPU Free

Side by side

	Whisper (the model)	Dictator
What you install	Python, dependencies, model weights, ffmpeg	One installer; the model downloads itself on first use
How you use it	A command in the terminal per file	A shortcut anywhere in the system, or drag a file into the window
Dictation into other apps	Not out of the box	Yes — the text lands where the cursor is
Live subtitles	No	Yes, with the translation underneath
Translation	English-only translation task	198 languages through AI, 45 offline
Speaker separation	Separate tooling required	Built in, runs locally, free
Price	Free, open source	Free; cloud extras from about $6/mo
Open source	Yes	No

What the wrapper adds

Six global shortcuts

Record, paste the last result, dictate with translation, translate the selection, translate what is on screen, live subtitles.

History that stays local

Every transcript is kept on your disk and searchable; nothing is sent anywhere in the local mode.

Text tools after the transcript

Summarise, remove filler words, rewrite, business tone — each result saved as a separate version.

Phone included

An Android app and a keyboard that dictates into any chat, with translation on the way out.

When plain Whisper is the better choice

If you are building a pipeline — batch-processing an archive, wiring recognition into your own code, or you need the licence and auditability of open source — use Whisper directly, or one of the open wrappers like Buzz or Vibe. This app is for the person who wants the text now, not a script.

Frequent questions

How big is the model?

From a few hundred megabytes to about one and a half gigabytes, depending on the size you pick. It is downloaded once.

Does it work without a graphics card?

Yes, on the processor — slower, and a lighter model helps.

Are there minute limits?

Not on local recognition. Cloud features are metered because a server pays for them.