· 7 min read · Privacy, Ownership
Why your AI assistant should live in your house, not in a data centre
Private by construction, not by policy. The case for owning the machine that reads your life, and what 'own your AI' means in practice.
Every cloud assistant asks you for the same thing: trust us. Trust the privacy policy, the security team, the retention schedule, the promise not to train on your data, the promise that the terms will not change. Some of those promises are kept. All of them are promises.
A machine in your house makes a different offer. It does not ask you to trust anyone, because the data never leaves. We call this private by construction, not by policy, and it is the reason Majordomo exists.
The assistant has to read everything
A useful household assistant is only useful if it reads everything. Not a filtered feed, not the emails you forward to it, everything: bank statements, medical letters, the lease, the children's reports, the argument with the builder. The value comes precisely from the completeness. That is also exactly the data you would never knowingly upload.
Cloud assistants resolve this tension by asking you not to think about it. We resolve it by keeping the data on an encrypted disk in a room you can lock.
Eight things ownership means
When we say own your AI, we mean something concrete.
- Own the machine. It is your property, outright. No seat, no subscription required to keep using it.
- Secure on your hardware. Encrypted disks, no open ports, no remote access unless you grant a support session and end it when you like.
- Works in every scenario. Internet down, provider outage, company gone, power cut: with battery backup it keeps working through all of them. Nobody can switch it off but you.
- Disaster-proof. Encrypted, versioned backups to drives you hold. If the machine dies, your assistant and its memory are restored on new hardware in an afternoon.
- Personal. One assistant, one household. It learns your names, preferences and style and is never shared with another family or used to train another model.
- Confidential documents, shared with no one. Read by a machine in your house. Never uploaded, never used for training, never subpoenaed from a third party.
- Own your data. Standard formats on a disk you hold. Export, back up or walk away whenever you like.
- Full customisation. Choose the models, the rules, the voice, what it may reach. Open-weight and inspectable, so you or we can change anything.
"But cloud models are better"
Sometimes, at the frontier, yes. Our view is that the household job does not need the frontier. It needs a model that reads carefully, remembers reliably and, above all, does not cry wolf. We measure that last point obsessively: the headline number in our evaluation suite is the false-positive rate, because a household forgives a missed catch far more readily than being nagged about things that are fine. Open-weight models on a 96 GB card clear that bar today, and the gap closes every quarter.
The cost argument
A cloud assistant that reads everything is metered on everything. A machine at home costs its hardware once and then a few watts. A Household build draws about 168 W on a typical day, which is roughly a pound or a dollar a day depending on where you live. There is no per-token meter and no reason to ration what it reads.
What you give up
Honesty requires the other side. A machine in the house is a physical object that takes a shelf, makes a little noise under load unless you choose a quiet card, and needs the occasional update. It costs more up front than a monthly subscription. And if you want the very newest frontier model for a specific task, you can still use one; Majordomo simply will not be the thing sending your household data to it.
For most families that trade is obviously right once it is laid out. The assistant that knows everything about you should be the one thing in your digital life that answers to no one else.
Configure a build, or read about what the hardware can actually do.
Own your AI.
Configure a Majordomo with real hardware and estimated pricing, then request a delivery estimate. Nothing is charged until you confirm.
Configure your build