Akrib

Download Akrib

Create an account to start your 14-day trial, no card needed. The installer only brings the program; the speech and text models are downloaded on first run, at the size your computer can handle.

Windows

0.5.0
Download · 64 bits 34 MB
How to install

Windows 10 or 11, 64-bit. For the AI models, 8 GB of memory or more helps a lot; an NVIDIA graphics card speeds up transcription.

The installer does not have a paid digital signature yet. Windows may show “Windows protected your PC”: click “More info” and then “Run anyway”. Check the SHA-256 above if you want to be sure.

64 bits · SHA-256
a6a39e356b7bd2326037edf6be64892ab8791cdbf3245cbdd6214476969e22f4

macOS

0.3.4
Download · Apple Silicon 53 MB Download · Intel 55 MB
How to install

macOS 13 or newer. On the first recording, macOS asks for microphone and screen recording permission: allow both in System Settings → Privacy & Security.

The app is not signed by Apple yet. The first time, open Akrib with a right-click → “Open” → “Open”. If macOS says the app is damaged, run in Terminal: xattr -dr com.apple.quarantine /Applications/Akrib.app

Apple Silicon · SHA-256
fc4b1323366935dede0768c604cbaecacc64671d4109af7a53f674c0403de7ff

Intel · SHA-256
2ca7e4cd24fb62264f74e16760b4a26dfd1ef42885e27ff45e13ee30d16d2bad

Linux

0.3.4
Download · x64 60 MB Download · ARM64 63 MB
How to install

A .deb package for Ubuntu, Debian, Linux Mint and derivatives; apt installs the dependencies for you. Recording computer audio needs PulseAudio or PipeWire (the default on these distributions). On Wayland, screenshots may ask the system for permission. Other distributions: AppImage on the releases page.

sudo apt install ./Akrib_*.deb

x64 · SHA-256
6dce811a083bf626aa062d87104b8f8a098429bc7671ed698416cfae49cb8136

ARM64 · SHA-256
80c4689deb9580e16df471fa85aa2f615aac1737f5da14bc16d14c049ab19f0b

What happens on first run

The program picks the models for your computer and shows the size, the license and whether it fits in memory before downloading.

  1. Language and microphone and screen permissions
  2. Speech model: Whisper small, large-v3-turbo or large-v3, with a memory traffic light
  3. Text model: Qwen 3.5 4B or Gemma 4, or your own Ollama, LM Studio or endpoint
  4. 10-second test: record, hear both channels and see the transcript
Open as a page