akou

akou

Docker app from geiserx's Repository

Overview

Self-hosted transcription server. A program uploads an audio file and reads back the text with word and segment times and speaker labels, as JSON, SRT or VTT, by long-poll, an event feed or a signed webhook. It also answers the OpenAI transcription endpoint (POST /v1/audio/transcriptions). A web page shows the job queue, the API keys, the models and the settings. This is the server mode of akou, a call recorder for macOS, with the same core.

First use: the speech models (about 3 GB) download on the first job. Set the web page's admin password from the Unraid terminal, at least 12 characters (this reads it without echo): read -rs P && printf '%s' "$P" | docker exec -i -u 1000:1000 akou akou admin set-password; unset P then log in and create one API key per program on the Keys page.

Folders: Unraid creates the appdata folders as root, and akou runs as uid 1000. The container starts as root only to hand the Data and Models folders to uid 1000, then drops to that user before akou starts; that is what the extra parameters and post arguments do.

Network: akou has no TLS of its own and listens on the network only with AKOU_BEHIND_PROXY set to true. Keep it on your LAN, or put a reverse proxy with TLS in front before you reach it from outside.

The image tag is the akou version; there is no latest tag. For a GPU use the -cuda or -vulkan tag (see the docs). More info: https://geiserx.github.io/akou/server/

akou

akou

Release CI License Docker pulls Stars

akou is a desktop app for macOS that records your calls on your own computer, transcribes them on the machine, and answers questions about a call while it is still running. There is no bot in the meeting, no account and no cloud: your own terminal agent (Claude Code, Codex or anything that speaks MCP) can follow and question the call, and the notes land in files you own. The same core runs as a self-hosted transcription server in Docker.

A saved call in akou: calls by workspace on the left, three speakers with their talk time, a transcript with a time and a name on every line, and the notes taken during the call on the right

Features

  • Records the microphone and the call audio on separate channels, with no bot in the meeting and no virtual audio driver.
  • Shows a live transcript as people speak, and runs an accurate pass with speaker labels on your Mac after the call.
  • Keeps a timestamped notepad while the call runs; every note carries the time it was written.
  • Answers a question about the call while it runs, with the Claude Code or Codex you already pay for, a local model, or your own API key. With none, Ask becomes a search of the call.
  • Lets your agent start, follow, question and annotate a call from the terminal, over a local API or MCP.
  • Types what you dictate into any app when you hold a key, and learns a word you fix only if you say yes.
  • Fixes names and product terms across calls from one word list you own.
  • Hands a finished call to your own vault or repository as Markdown, an event log and the audio, through an export folder, hooks, a signed webhook or the API.
  • Runs as a transcription server in Docker, with file jobs, an OpenAI-compatible endpoint, and Wyoming (Home Assistant Assist) and Bazarr's /asr endpoints, each one setting away.

Quick start

brew install --cask geiserx/akou/akou
# or download akou-1.0.0-macos-arm64.dmg from https://github.com/GeiserX/akou/releases and drag akou into Applications
open -a akou

The first window asks what you will use akou for and downloads the speech models (about 3.0 GB). At your first Record, macOS asks for the microphone and system audio; the call then shows up in the sidebar with a live transcript. Needs a Mac with Apple silicon and macOS 14.4 or later. To let Claude Code or Codex follow your calls, install the command line from the akou menu and run akou skill install; see Getting started and Agents and the command line.

Status

1.0.0 is the first stable release. The 0.x releases were prereleases. The app runs on macOS on Apple silicon only, signed with a Developer ID and notarized since 0.6.1. The Windows and Linux archives are the akou command line alone: they manage models and settings and drive a remote akou, but cannot record. Each release's changelog entry lists its known limitations.

Documentation

Docs: https://geiserx.github.io/akou/

License

GPL-3.0-or-later. akou is written from scratch and replaces hark, whose capture design it carries over in full.

Download Statistics

4,023
Total Downloads

Related apps

Details

Repository
drumsergio/akou:1.0.0
Last Updated2026-10-10
First Seen2026-10-11

Runtime arguments

Web UI
http://[IP]:[PORT:8476]/
Network
bridge
Privileged
false
Extra Params
--user 0:0 --entrypoint /bin/sh

Template configuration

Web UI and API PortPorttcp

Web page and API

Target
8476
Default
8476
DataPathrw

Settings, API keys and jobs

Target
/data
Default
/mnt/user/appdata/akou/data
ModelsPathrw

Speech models, downloaded on first use (about 3 GB)

Target
/models
Default
/mnt/user/appdata/akou/models
Behind proxyVariable

Must be true for akou to listen on the network. It has no TLS of its own: keep it on your LAN or put a TLS proxy in front.

Target
AKOU_BEHIND_PROXY
Default
true