She watches the house.
He runs it.
In every Dalmatian village there was a baba at the window who saw everything, and a dida who kept the household running like clockwork. These are their digital heirs — two self-hosted systems that share a bus, a philosophy, and a stubborn belief that your home's intelligence belongs on your own hardware.
Why I started over
For years, Home Assistant and Frigate ran my house, and I was grateful for every day of it — for the communities who pour themselves into them, for how far they've come, for everything they taught me about what a home could be. And it was that very learning that slowly carried me past them. Today I run neither one. That isn't an easy thing to admit about tools I loved — but it's the whole reason these two applications exist.
None of what follows is a complaint against them; they're remarkable, and they were built honestly for the time they came from. But the longer I lived with them, the more the small daily frictions added up. The one that wore me down most was the simplest: change a single setting, and wait for a restart. A threshold, a zone, one rule — and the whole system reboots before it takes. Add surveillance that waits for motion and quietly loses the person standing still, and automation stitched from hundreds of rules you have to carry in your head, and one day you notice you're working for the tool instead of the other way around.
So I began again from a different foundation. In BABA and DIDA you change a setting and it takes effect at once — no restart, no held breath, the whole system simply follows. BABA looks at every frame whether or not anything moves, so it never loses a person standing still, and it knows who arrived, not just that something did. It even learns your scene on its own — teaching itself which corners are only furniture, and quietly healing the moment things change — and when you do want to shape it, an AI helps you set it up instead of a manual you have to master. DIDA keeps every device in its own isolated process, so one misbehaving thing can never take the house down with it. And the goal I set above all others — to bring every device I already owned across with me, and replace not a single one — is completely met: every sensor, camera, switch and light I had still works, now on a foundation that no longer fights me. Same conviction, same backbone: start from what we know now, and let simplicity be the feature instead of the compromise.
If you've felt it too — the affection for these tools, and the quiet sense that their foundation stayed in the last decade — then these two were built for you as much as for me. Come and see.
And one honest thing about how they came to be: I have not written a single line of this code by hand. Every line is the AI's work — but not one move it made went unsupervised. I directed it, argued with it, and stood behind every decision, the way an engineer answers for a team he's accountable for. Not generated — built, deliberately, by someone who owns every choice in it.
— Ivo
BABA She sees everything.
A self-hosted NVR built on modern ML: transformer detection on every frame, real person re-identification, face recognition and visual scene understanding — on your own GPU.
- —Never loses a person standing still
- —Knows who, not just what
- —Understands your gate, garage and lights
DIDA He runs the house.
Home automation rebuilt on industrial foundations: every protocol isolated in its own process, a validated capability model at the core, and automations that structurally cannot crash it.
- —A failing device can't take the house down
- —Control from the floor plan, voice via Matter
- —Postgres + ClickHouse, not a fragile monolith
Local-first, and stubborn about it
No cloud accounts, no subscriptions, not a single frame of video leaving your network. Everything runs in Docker containers on hardware you own — from a spare mini-PC to a single RTX GPU. Free for personal use under the PolyForm Noncommercial license.