Lisa Smart: the concept of Lisa at the heart of the home
This is the second post about a Lisa that does not exist yet. Like Lisa Touch, Lisa Smart is today a concept: a flavour in the specification and a few decisions. No smart image has been built, and no device runs it.
Lisa Smart is Lisa for the home. It has two jobs. It is a smart speaker with a small screen. And it is the hub for the home’s connected devices: the lights, plugs, sensors and everything else people now call the Internet of Things.
What is decided
It is the same Lisa. Lisa is one platform in four flavours: desktop, server, touch and smart. They share the same frameworks, the same contracts and the same release, with an image of its own for each flavour on each architecture. Lisa Smart is not a separate project. It is the home flavour of the platform that already runs our desktops and servers.
A speaker with a small screen and no keyboard. You speak to it, and the screen shows you what it heard and what it did. What runs is one full-screen surface over the same frameworks: no windows, no Dock, no menu bar. A control a finger presses on that screen uses the same touch-primary size as Lisa Touch, at least 44 points.
The hub for the home. We ruled on 2 October that Lisa Smart is where the home’s connected devices meet. A home today collects devices from many makers, each with its own app and its own account. Lisa Smart is meant to be the one place they all answer to.
Plug and play. This is the part we care about most. Lisa detects every smart device on the home’s network and connects it, with nothing to set up. You should not have to install an app, make an account with the device’s maker, or type a code to make a light switch on. The device joins the network, and Lisa connects it.
It stands on the same contract. A smart device conforms to the same desktop profile of the platform contract as a desktop: graphics and input, audio, networking and Bluetooth, on a Fedora family base.
The voice is Lisa’s own
A speaker in the home only works if people trust it. The rules for Lisa’s voice are already written, for every flavour, and Lisa Smart inherits them:
- “Hey Lisa” is on the device. Voice activity gates the microphone, and silence stores nothing.
- Near is not to. A finished utterance is transcribed locally, then a classifier decides whether it was spoken to Lisa or only near it. An utterance that was not addressed to Lisa is discarded.
- The wake word earns its place. It ships only with a measured false-accept rate, and until then it is off. Push-to-talk always works.
- It is our own. Someone else’s wake word in a product called Lisa is not a substitute.
Speech for applications goes through the same Speech framework as on the desktop, on the local engines and under the person’s choices.
What is deliberately open
The ruling says what Lisa Smart does. It does not yet say how. Nothing is written about:
- which devices the hub speaks to, and over which protocols;
- how Lisa detects every device on the network and connects it without any setup;
- which daemon holds the authority over the home’s devices, as every other authority on Lisa has exactly one owner;
- what the small screen shows when nobody is speaking to it;
- which hardware to bring up first.
The docs say so in one line: the hub is a role without a page. Lisa’s way is to write the page first and then build to it. Plug and play is easy to promise and hard to make honest, so that page will say exactly what “nothing to set up” means, device by device, before any code claims it.
Where to read more
- Flavours: the four flavours and the ruling on Lisa Smart.
- The agent: the voice rules, “Hey Lisa” and what stays on the device.
- lisaos.app/smart: the product page, marked specified, not built.