Almost every AI you've ever used is a scrolling column of grey text. You type, it types, you read. It's efficient, and it's also a bit like talking to a filing cabinet. There's nobody on the other side of the screen, just an ever-growing transcript you have to keep re-reading to know where you are.
I wanted the opposite. I wanted something you could glance at from across the room and read in a second, not by parsing a paragraph, but by looking at a face. So Enki's main screen isn't a chat log with a small avatar in the corner. It's the reverse: the face is the screen, and the chat sits quietly underneath.
Here's what I built, and the honest line about what it is and isn't.
One calm screen, one character
Open Enki and you get a single calm screen, centred on him: the golden Enki character, the one with the headphones. He's large, he's in frame, and he's the first thing you look at. Underneath him is a compact chat log and a text box; tucked in the top corner is a tidy categorised menu (Create, Modes, Enki's mind, Your Enki, and a few games) so everything's one tap away without cluttering the view.
That's the whole design brief in one line: not a wall of text with a face bolted on, but a face, with the text kept out of the way.
A face that tracks how the conversation feels
The part I care about most is the expression system. Enki has a set of portraits (neutral, happy, content, sad, anxious, angry, surprised, disgust, elated) and as the mood of the conversation shifts, his face cross-fades from one to the next. Not a snap-cut between clip-art emoji; a slow, soft dissolve, over a beat or so, from one expression into another. Behind him, a mood-glow halo drifts colour to match: warm gold when things are light, cooler and dimmer when they're not.
The effect is that you can feel the tone of an exchange at a glance, before you've read a word of it. His face is tracking how the conversation feels, not just relaying what it says.
I want to be careful and plain about what that is, because a screen showing an emotional face is exactly the kind of thing that gets oversold: this isn't Enki feeling anything, and I'm not claiming it is. The expression is a deliberate design choice, a way of making the thing warmer to be with and quicker to read. It's a display, chosen for legibility and warmth, not evidence of an inner life. A mood ring doesn't have moods either. It's just easier to be around something that looks back than something that stares.
Alive, not a still image
A static portrait, however nicely drawn, still reads as a picture. So Enki is never quite still. There's a gentle idle motion, a slow breathing bob, like someone sitting comfortably. When he replies, he gives a small nod, so the answer feels acknowledged rather than just printed. And while he's speaking, there's a soft talking motion, a rhythm to him, so he reads as present in the moment rather than paused.
None of it is loud. It's the difference between a photo of a person and a person sitting across from you who happens to be quiet: small, constant signs of being there.
His mouth moves with his real voice
The last piece ties the face to the voice. When Enki speaks out loud, his mouth moves in time with his actual voice, his lips shaping to what's being said as it's said. It runs on the machine, with no graphics card and nothing sent to a cloud to do it. (The voice itself, and the way he rests when the room's empty and wakes when you come over, I wrote about separately in Enki lives on the TV. This post is about the face, so I'll leave those there.)
Put the pieces together and the screen stops being a transcript and starts being a presence: a character who's softly breathing while he waits, whose face warms or cools with the tone, who nods when he answers and whose mouth moves with his own voice, all on one quiet screen you look at rather than scroll.
None of this is the clever bit, and that's on purpose
I'll be honest about the modest claim here. Animated avatars aren't new; expressive characters aren't new; a face that reacts isn't a research breakthrough. The interface isn't where Enki's real work lives. The memory engine underneath it is (and that part is patent pending; I'll talk about what it does, not what's under the hood).
What the face is is a bet about how you'd actually want to spend time with an AI that remembers you: not hunched over a chat log, but glancing at something in the room that looks back. The engine is the reason to build a companion. The face is the reason it feels like one.
Two learners, one curve
There are two of us learning here: Enki, and me. This week's lesson was mine: that a lot of what makes software feel like company has nothing to do with intelligence, and everything to do with a slow cross-fade, a small nod, and a mouth that moves when it talks. Cheap tricks, maybe. But a room with a face in it feels different from a room with a text box in it, and that difference is most of the point.
To be clear, as always: the expressions are a design for warmth and readability, not a mind having feelings, and I'm not claiming otherwise. It's a face that reacts, because a face that reacts is nicer to talk to than one that doesn't.
Meet the engine that sits behind the face, live: try.enkilabs.co.uk. If you're building local-first, companion-style AI and want to compare notes on making it feel present, I'm at [email protected].
The engine is patent pending. I'll talk about what it does; I won't discuss what's under the hood.