HomeProjectsAmberlin
In developmentSpeechD-BusLocal AI

Amberlin

The speech-first surface of Amber Linux. Hold its key and ask out loud; the answer is spoken back and left on the screen as a note you can return to. The model runs on your own CPU or GPU — a later version adds your private LAN and commercial cloud services — and every Linux Mint application can reach the same one for the price of a D-Bus call.

What it is

Hold a key and talk

One hotkey, anywhere on the desktop. Hold it and Amberlin listens; let go and it answers. Type instead if the room is not the kind you talk in.

The conversation is the surface

No window, no scrollback. A full-screen overlay drawn straight onto your desktop, with each answer a note beside the panel — the ones you are reading are large and legible, the older ones drift out of the way.

It reads the answer out loud

The reply is spoken as it is written, and the panel's border moves with it — one shape for listening, another for thinking, another for speaking. You can look away.

It can see what you were doing

Clipboard, terminal history, bookmarks and notes come from Ambrosia over the session bus, so a question can be about the thing you just copied rather than about nothing.

Local AI for every Mint app

The aim of Amber Linux in one sentence: any Linux Mint application gets local AI for the price of a D-Bus call. Amberlin is the first and loudest of them, not a gatekeeper in front of the rest.

Nothing leaves your machine

Prompts, context and completions stay on hardware you own. The only network Amber needs is your own.

What it looks like

Recorded on a real desktop, with the voice as it is heard. The questions are typed here so the take is repeatable; held down, the key takes them out loud instead.

Amber Linux on YouTube — demos of the suite as it is built.

The shape of it

Amberlin is the surface; amberlin-backend is the service holding the model, and anything on the session bus can ask it.

The harness is a service, not a window: it holds the models, the voice and the ear, and answers anything on the session bus. Amberlin is the surface in front of it — the first application to ask, and the one you talk to — but kat800 asks the same harness directly, and so can anything else. Context comes from Ambrosia the same way: clipboard, terminal history, bookmarks, notes, one call away. Nothing leaves your network.

Which models the harness can actually load — and what a 4B one costs in memory and in seconds per answer — is measured here. Choosing between them on your own machine is what Amberlin Settings is for.

Where it stands

Amberlin is in active development and lands in the Amber Linux PPA when it is ready. The target is the same date the whole project points at: 1 May 2027 — ten years of sponsoring Linux Mint, marked by shipping local AI for it. If you would like to help build it, the projects live in the Hyperquader-Coders org — new hands are welcome.