Small, quick, and out of your way.

Why it is light

• One Rust engine does the work. Dictation, text rules, the model router, your vocabulary, settings, history, keys, shortcuts and typing all run in one native program instead of a web page.
• The interface is only the interface. The windows just show what the engine tells them, and the settings window can be closed while Murmint keeps running in the tray.
• Nothing extra starts. Background features of the browser engine that Murmint never uses (translation, media casting, update checks, sync) are switched off.

Why it feels quick

• The microphone starts first. Finding which app you are typing into happens after recording has already begun.
• The connection is warm. While you talk, the connection to your speech provider is already opening, so the text comes back sooner when you let go.
• Live models stream. With a live speech model, your words are transcribed while you speak instead of after.
• One request to format. The whole dictation is cleaned up in a single request, sent to the cheapest model that can handle it, with a fallback if it fails.

Check it yourself

Open Task Manager, go to Details, and add up the memory of the Murmint processes. How fast text appears depends mostly on the speech and formatting providers you choose.