$ 100% local dictation for Linux

Speak.
It types.

Hold a key and talk. Freeflow cleans up what you said and drops it at your cursor, in any app. Nothing you say leaves your machine.

How it works

No cloud. Three steps.

  1. 1

    Hold the hotkey and talk

    Freeflow records only while you hold the key. Release to stop.

  2. 2

    Your machine transcribes it

    whisper.cpp turns your voice into text right on your computer. Nothing is uploaded.

  3. 3

    Clean text lands at your cursor

    A small local AI tidies the words, then Freeflow types the result where you were.

Features

Everything runs on your machine.

AI cleanup

Fillers gone, punctuation fixed, capitalized. You talk loosely, it writes clean.

um so basically can you uh send the the file tomorrow

Can you send the file tomorrow.

Tones per app

Formal for email, casual for chat. Freeflow picks the tone by the app you type into.

Formal Neutral Casual

Personal dictionary

Force the spellings you care about. Names, tools, jargon land right every time.

hyperland Hyprland Hyprland Ollama

Local history

Your recent dictations stay on disk, yours to keep or wipe. Never synced anywhere.

Dark mode

A settings app that matches your desktop, light or dark.

Tray and hotkey

Lives in your system tray. One chord to dictate, hands-free mode on a double tap.

100% local.

$ freeflow status ● all local

Install

Up and running in three commands.

Fedora, Arch and Debian friendly · MIT · uninstall.sh removes every trace.

Read the install guide

About

Why this exists.

Every mainstream dictation tool streams your voice to someone else's server. Freeflow started as a bet that a modern Linux machine can do the whole job itself: whisper.cpp handles the listening, a small local model does the tidying, and your words never touch a network.

It is free, MIT licensed and built in the open. Read the code, file an issue, or rip it apart and make your own. The only account you need is the one you already have: your own computer.

MIT licensed 280 tests 0 network calls while you dictate Source on GitHub