HOMI is the private AI hub for the whole family. It helps with the kids' homework, runs the smart home, keeps track of life. Local 7B–13B models pre-loaded — bring your own OpenAI, Anthropic, or Google key when you want frontier power.
HOMI is built around the four things a family actually needs an AI for. The same hub that turns the lights off also reads bedtime stories, reminds you about the school recital, and notices the kid hasn't slept enough this week.
The shared brain for your household. Voice-controlled, always on, knows everyone in the family.
Homework help, spelling drills, multiplication tables, bedtime stories. With parental boundaries baked in.
Lights, thermostat, curtains, TV, AC, locks, cameras. One voice, one hub, no five apps.
Pulls in your wearables and surfaces patterns. Not medical advice — just better daily decisions.
Because it's right there in the kitchen, by voice, with no app to open and no screen to fight. Reads spelling words. Walks through long division. Tells the bedtime story when you've already told three. With limits the parents — not big tech — set.
HOMI runs serious models on the device — enough for daily voice, automation, and most kid-tutoring tasks. When you want frontier power (long-form reasoning, complex homework, creative writing), it routes through your own provider key. We never store, proxy, or train on your data.
Pre-loaded open-source models for voice, automation, summarization, and most tutoring tasks. Runs at < 30W. Offline-capable.
Plug in your own OpenAI, Anthropic, Google, or Mistral API key — pay them directly. HOMI is never in the billing path or the data path.
Built on the same open-source ecosystem powering 1M+ smart homes. Zigbee, Matter, Thread, BLE all on-board. Works without an internet connection.
Most "AI hubs" are dressed-up routers running someone else's cloud. HOMI is a real inference-grade machine — enough silicon and memory to run 7B–13B models on-device, in a passively cooled box that sits silently on a shelf.
When HOMI arrives, it's already configured. Plug it in, scan the QR, and it walks you through pairing your devices, your wearables, and (optionally) your AI provider key.
HOMI boots into OpenClaw OS on first power-on. Ethernet recommended; Wi-Fi 6 supported. Local models load in under 90 seconds.
Opens homi.setup on your phone. Pair smart home devices, wearables, and add household members and kids with their own profiles.
Paste your OpenAI, Anthropic, or Google key for frontier-model access. Skip it and HOMI runs entirely on local models. Both work.
2,500+ smart home devices, every major AI provider, the calendars and wearables you already use.
The next decade of home AI will hear every conversation in your house — your kids' homework questions, your sleep, your routines, what your family is worried about. Whoever owns that data owns something the prior internet era never got close to.
Alexa, Google Home, Apple — they're racing to be that. Their model is the same as it has been: store the data, train on it, sell ads against it, lock you in.
HOMI flips it. The hardware sits on your shelf. The data lives there. The AI runs there too — local 7B–13B models for daily use, with the option to plug in your own API key when you want frontier power. We are never in the billing path and never in the data path.
It's the same playbook we ran with IMMI: open-source upstream, real legal docs on day one, and a hard guarantee that if Prizmox vanishes tomorrow, your hardware and your data both keep working.
Early-bird preorder locks your unit in the first 500-unit production run. Ships Q3 2026. Full refund any time before shipment.
Both — and you choose. HOMI ships with 7B–13B parameter open-source models pre-loaded on the device. They handle voice, automations, summarization, and most kid-tutoring tasks fully offline.
For frontier-grade reasoning (long-form homework, complex writing, deep research), you can paste in your own OpenAI, Anthropic, Google, or Mistral key. Your key, your account, your bill — HOMI is never in the middle.
You can also point it at a self-hosted Ollama instance on your own machine if you want everything to stay on-prem.
Each kid has their own profile. Parents set time caps, allowed subjects, content boundaries, and review session logs from a parent dashboard. Younger kids (4–8) get story mode, jokes, and reading practice. Older kids (8–14) get subject tutoring, spelling, math, and homework Q&A.
Kid sessions default to local models so nothing leaves the device. Parents can opt-in to cloud routing per profile if they want stronger tutoring quality — and that uses the parent's own API key.
HOMI is COPPA-aware by design: no ads, no data sales, no third-party trackers, no voice recordings shipped off the device.
For automations, lights, locks, climate, sensors, calendar, reminders, family logistics, and voice queries — yes. For Amazon-shopping-by-voice or first-party Spotify integration, you'll keep an Echo for that one job. Most families do.
The thing HOMI does that Alexa and Google Home don't: it works for kids, it doesn't ship your house's audio to a server farm, and the AI behind it is yours to choose.
HOMI integrates with Apple Health, Garmin, Oura, Whoop, and Fitbit. It pulls sleep, activity, recovery, and stress signals into the family dashboard and surfaces patterns over time.
This is wellness, not medical: HOMI doesn't diagnose conditions, prescribe anything, or replace a doctor. It nudges routine — better sleep windows, screen-time limits, hydration reminders — based on signals you've already chosen to share.
The data stays on the hub. We don't sell it, we don't share it, we don't build a model on it.
HOMI is a standard inference-grade machine running OpenClaw OS, which is built on the open-source smart home and AI stack. Your data, automations, models, and hardware all keep working.
You can update directly from upstream, back up your configuration to a USB drive, and export everything as standard files at any time. We don't gatekeep anything.
No. HOMI is a one-time purchase. We do not charge monthly fees for any feature, ever.
If you choose to plug in a cloud AI key, you pay that provider directly at their rates — typically a few dollars a month for a family. With local models only, you pay nothing beyond electricity.
For most families, an Anthropic or OpenAI API key works great — you'll spend $2–10/month for typical household use. Google Gemini and Mistral keys also work. If you don't want to use any cloud at all, the on-device models are enough for daily voice, automation, and most tutoring.
HOMI guides you through key setup at first run. If you want to skip it, every cloud feature falls back to local automatically.
An ARM/x86 inference-optimized SoC, 32 GB+ unified RAM, 512 GB NVMe SSD, Wi-Fi 6, Gigabit Ethernet, on-board Zigbee 3.0 / Matter / Thread / BLE radios, USB-C, and full microphone array. Passive cooling, <30W idle, silent.
The full spec is published before you preorder. No mystery silicon, no NDA chips.
1-year hardware warranty on manufacturing defects. 30-day returns for any reason. Full refund on preorders any time before your unit ships.
A small team that thinks the family AI shouldn't be owned by Amazon, Google, or Apple. We build hardware that puts the brain in your house, not in a server farm. HOMI is our second product after IMMI Photo Box.
We stay off the founder-Twitter treadmill on purpose. The product is the argument.