Open-source household AI

A shared AI and memory for families and small groups.

One Maurice everyone talks to — and a memory you all build.

The server runs on a machine you own — a private wiki for each person, and it knows their books and files too.

Free & open-source · One container, on a machine you own · iPhone & iPad native apps
Your data stays home — no cloud account

Chez Maurice on iPhone — a Storyteller conversation

9:41

Bedtime story about whales

Today · 8:14 PM · The Storyteller

Tell me a story about a whale who collects sounds.

Once, in the deep grey water off the coast, there lived a young whale named Brume who kept the sounds she loved in a coral chest…

the creak of a harbour at dawn

a child laughing two bays over

She never kept a sound she hadn't been given. That was the whole of her magic.

Ask the Storyteller something…

Qwen 3 · local

What makes Chez Maurice different

An assistant that becomes part of your team, not one that watches you.

The memory is your own Markdown — visible, scoped, and shaped by deliberate action. Here's what that buys a household.

Memory you choose — and grow

He knows exactly what you hand him.

It runs both ways, and both are deliberate. Load notes, books, files and past conversations into a chat through the composer, and watch the token budget fill — nothing else is in the room. Then write back: anything worth keeping becomes a note in the wiki, ready for next time. You don't just feed a session — you grow a memory you can see.

Load context in — a live token budget you can feel

Write notes back — they join the wiki he reads from next

All plain Markdown, in a git repo you own

léo/allergies.md

# Léo — allergies - peanuts (severe) - shellfish EpiPen in the blue bag.

a note someone writes

Context for this chat

pulled from the wiki

Léo — allergiesnote · just now 0.3k

Trip planningpast conversation 6.4k

what Maurice draws on

Together, with Maurice

Group chats — and Maurice takes the notes.

Have a real conversation with your team — a household, a project, a few friends, all in one thread. Invite Maurice in when you want him: he takes part, answers what's asked, and writes the notes — so what you decide together becomes memory instead of lost scrollback.

Several people in one thread — summon Maurice with a tap

He joins only when called, using the notes in the room

Ask him to capture a decision — it becomes a note you keep

A group chat — several household members in one thread, with Maurice summoned to answer and take notes

Kyoto trip

Marc · Paola · Léo · Maurice

MarcWhat's still open on the list?

Léocan we add the monkey park??

Léo summoned Maurice

MauriceStill open: ryokan booking and JR passes. Monkey park — added. I'll keep the running notes.

Saved to Kyoto trip — notes

One device, every mind

One iPad. Everyone's own Maurice.

Cloud chatbots assume one person per device, per login — but a household isn't built that way, and not every family wants to buy each child their own tablet. Maurice puts everyone on a single shared iPad or Mac: each member has their own conversations, notes and memories, and switching is just a tap and a PIN. The kids share a screen without sharing a brain.

Per-member history, notes and personas

Host guests — and be a guest in another household

Roles for adults, children and guests

mockup

Who's here?

Tap your face to continue

PPaolaadmin

MMarcstandard

LLéochild · 11

NNoéchild · 9

And more

Built like a home, not a dashboard.

Bring your history

Import your existing threads from Claude and ChatGPT — the official export .zips — so your past conversations come with you, searchable.

Many Maurices, a hat for each

Create focused personas — a homework tutor, a trip planner, a storyteller — each with its own voice, model, creativity and bound context. The hat tells a child instantly which Maurice they're talking to.

Households & guests

Run several households on one device, with guests who keep full capability but a limited reach — an explicit allow-list of people and personas.

Native apps, one container

Real SwiftUI apps for iPhone, iPad and Mac — and a single container holding the whole server. One image to run, one volume to back up, nothing else to assemble.

Your models, your choice

Local models through Ollama, or Mistral, Anthropic and OpenAI in the cloud — switch per conversation, grouped by provider.

A browsable web garden

The same Markdown becomes a themeable website — private notes encrypted at rest, with four hand-made garden themes to choose from.

Get Chez Maurice

Three pieces. Start with the server.

The server is the engine — it runs in a container on a machine you own and holds everything. The apps are how you reach it.

Start here

Chez Maurice Server

Required · Docker · a Mac or a Linux machine you own

The chat engine, the garden and your data, in one container. Clone the repository, build the image, bring it up — the guide is five commands long.

Installation guide

One image · arm64 or x86_64 · ~260 MB running

Chez Maurice for Mac

macOS client · via TestFlight

The full desktop experience — chat, hats, the composer, gardens and files — in a native Mac window. Pairs to your server.

Join the TestFlight

Apple silicon · TestFlight beta

iPhone & iPad

via TestFlight

The same app on iPhone and iPad — built as one universal app. Join the TestFlight beta to install it on your devices.

Join the TestFlight

iOS 17+ · iPhone & iPad

How it all connects

Your apps talk to your own server over a private HTTPS link you choose.

Your server, your machine

data · garden · LLM calls

Tailscale / Cloudflare

Your devices

iPhone · iPad · Mac

The whole point

Maurice only knows what you've shown him. That limitation isn't a bug — it's the thing you can tame. He becomes yours through the relationship you build.

A father building an AI for his children · open-source · small on purpose

Why one container, and why your own machine

Maurice ships as a single container image: one thing to run, one volume holding everything that matters, the same image whether it sits on a Mac mini on a shelf or on a small Linux server you rent. That is a deliberate trade, and it replaced an earlier one. The notarized Mac installer was friendlier to open, but it meant two builds, two sets of assumptions about the host, and two ways for an installation to differ quietly from the one I develop against. One image means one place a bug can be.

The cost is honest: you need Docker, and the first run is a terminal rather than a double-click. What it buys is that the thing running in your house is the thing I run in mine — and that the clients can still be real SwiftUI apps rather than a browser tab pretending to be one. What has not changed is the part that matters: a household shouldn't have to hand its assistant to somebody else's cloud to own it.

Who's behind this

Maurice is made by one person — me. I started it as a household AI for my own family, on hardware I already owned, because I didn't want the people I live with to be the product. It's open-source and it's early: a solo project, shipped honestly, not a startup pretending to be bigger than it is. You'll find rough edges. You'll also find that nothing here mines you, upsells you, or phones home. — Candide

Solo-built · in beta · maintained in the open · candide.me