Ameka Voice · macOS

The first AI that runs your agents while you walk away.

Ameka is a voice-controlled layer over every AI agent on your machine — Claude Code, Codex in the ChatGPT app, any AI conversation in any browser. Its own model runs on your Mac, with no API key and nothing leaving the machine. Every session gets a smart loop: on its own timer Ameka reads where that session is, decides what it needs, and engages it — a check-in, its next step, an answer — then tells you in one sentence what moved.

$ curl -fsSL https://ameka.ai/install.sh | bash

macOS 13+ · Apple silicon or Intel

> irm https://ameka.ai/install.ps1 | iex

Or, if you would rather double-click: download install.cmd and run it. Windows will warn that it is from the internet — choose More info, then Run anyway.

Windows 10+ in PowerShell First build · v0.16.16 · 1066 KB · download the archive · sha256 c7b979f398de5ecd2a65b27bc874e240031bb66aca4dc55c80ae143d4d86a268

It installs as a real Mac app. The Ameka mark sits in your menu bar and shows what it is hearing, which session it is talking to, and what it last said — mute it, switch session, or read the log from there. Everything runs on your Mac.

Smart loops

A smart loop is Ameka picking a session up on a timer and actually doing something about it. Every active session starts on a ten-minute loop; each one has its own switch and its own timer in the menu bar, with a live countdown.

the session is workinga check-in, asking for one plain sentence at its next stopping point
it finishedthe next step it named for itself, typed in
it asked a questiona safe answer, or the question reaches you and the session is told to carry on
it is stucktold so, and asked what it needs
anything destructive or outward-facingnever done — one sentence to you instead, and you decide

Ameka reads your brain — a file of who you are and what you are working toward — before every loop, so it knows which sessions matter. It adds what it learns about your work to a section of that file marked as its own.

What it actually does

01

A session finishes a turn — or its loop comes round

Claude Code, Codex in the ChatGPT app, an AI conversation in Chrome or Safari, a script, a cron job — anything that can run one shell command can hand Ameka a briefing, and everything else gets picked up by its smart loop.

02

You hear two sentences

"The webhook is wired and all the tests pass. Want me to promote staging?"

03

You answer out loud

"Proceed."

04

Ameka types it into that exact session

Straight into the right window in the Claude Code app — or silently into a tmux pane, on a machine that runs its sessions there. Then it waits for the next turn.

Say anything on this list

proceed · continue · do it · run it · yes · ship it · keep goingSends your go-ahead to the session.
no · stop · hold on · wait · cancelSends nothing. Holds.
repeat · say again · whatReads the two sentences again.
details · more · explain · whyReads the longer version.
skip · nextDrops it and moves to the next waiting session.
mute · quiet — then unmute · resumeGoes silent. Comes back.
statusHow many sessions are waiting.
tell it … · say … · send …Dictate anything. Read back to you before it is sent.
talking to … · switch to …Points her at another project and brings that session up on screen.
open a session for … · start …Opens that folder in the Claude Code app, in the window you already have.

Anything Ameka does not recognise is read back for confirmation. Nothing is ever typed into a session without you saying yes.

What happens when you install it

The installer asks macOS for two permissions — the microphone, and Accessibility so Ameka can type your answer into a session — then tells you exactly what it can already see: every Claude Code session open on the machine, any recent Codex session, and every ChatGPT or Claude tab in your browser. Nothing to configure, no accounts to connect. If something is missing it names it and the one command that fixes it.

What you need

  • macOS 13 or later, or Windows 10 or later. Python 3.11+ (brew install python on a Mac; the Windows installer fetches it if it is missing).
  • No API key. The voice, the listening and the summary all run on this Mac — a neural voice model and whisper.cpp, installed for you. Bring an OpenAI key only if you prefer their voice.
  • On a Mac, two permissions asked once: Microphone, and Accessibility so Ameka can type your answer back. On Windows, just the microphone.
  • The Windows build is new and does the core loop — hear the briefing, answer, it types into the Claude window. Switching sessions by voice and the menu-bar app are Mac-only for now.
  • AirPods, or any headset. Ameka picks them up by name.

After installing she is already running, and starts at login from then on. Run ameka doctor — it checks every engine, device and permission and tells you exactly what is missing. She keeps herself current too: every half hour she asks ameka.ai whether there is a newer version, verifies it, and restarts on it in about a second. You never install her twice.

Cost and privacy

It costs nothing to run. The voice, the listening and the two-sentence squeeze all happen on your Mac — a neural voice model and whisper.cpp, installed for you — so there is no key to buy and no per-briefing charge. The daemon binds to localhost, nothing is exposed off the machine, and no audio is stored by us or by anyone else. Point the summariser at OpenAI or Anthropic if you would rather, and only then does a briefing cost anything: roughly a tenth of a cent.


iPhone Coming next

The same loop, from anywhere: your Mac keeps working and your phone is the headset. Mac first, iPhone right behind it. Tell us you want it and you go on the list.

Ameka Voice is free while Ameka is being built. It runs on any Mac — you do not need Ameka hardware. Reserve a machine →