Open source · Bring your own models

A conversational AI you talk to face to face.

A full-body character lives in a small 3D set and joins you in a video call. You speak or type, it listens, reacts, and answers in voice with matching expression and body language.

Early development. Nothing here is finished yet.

Why a face changes everything

Most assistants are a text box. You type, they answer, and the conversation stays abstract. A face that listens and reacts changes what talking to one feels like: it becomes something you talk to, not something you type at.

This one remembers you, helps you plan, runs tasks you schedule, answers from your own documents, and can run your smart home through Home Assistant. The character renders on your machine, the models are yours, and the memory lives where you put it.

What it does

Six things that set it apart from a chat window.

It has a face.

A full-body VRM character with lip sync, expressions and idle life, rendered in the browser with three.js.

It reads the room.

Voice, text and — only if you turn the camera on — your face feed one estimate of how you seem, and the character's mood answers it.

It remembers.

Episodes, facts with the dates they were true, plans, and a self-model it edits. In your MariaDB, your SQLite, or your browser.

It does things later.

"Every weekday at eight, summarise my notes folder." Visible in a panel, cancellable, and confirmed aloud when you create it.

It answers from your documents.

Your PDFs, notes and databases, with citations you can open.

It runs on your models.

Ollama, LM Studio, vLLM, llama.cpp, OpenRouter, NVIDIA, Anthropic, OpenAI, or models you download into your own browser.

How it works

Pick your models

Choose where speech, language and voice run — locally, remotely, or in your browser.

Allow the mic

Talk or type. The character hears you and watches your camera only if you turn it on.

Talk

Ask, plan, read documents together, or just think out loud. Replies come in voice with matching expression.

It remembers

Facts, episodes, plans and schedules are stored in the database you pointed it at.

Privacy promise

No telemetry, ever. Nothing leaves your machine unless you point it somewhere.

We ship no models. Anything that runs in your browser is downloaded by you, after a screen showing its size, licence and source.

Your memory lives in your database, not ours.

The camera is off until you turn it on, and video never leaves the page.