A network powered by the people using it.
Instead of relying on one company's data centres, the network could run partly on the computers of the people using it — their spare storage, spare processing power, and the hours their machines are simply switched on.
The simplest way to picture it: solar panels on a roof. You buy them. You use the power you need. What you don't use goes back to the grid, and you get credit for it. Same idea, with computing instead of electricity.
Storage
The space on your drive you are never going to fill, offered on whatever terms you set.
Processing power
The processor and graphics card you already paid for, working while you are not using them.
Being switched on
A machine that stays reliably online makes the whole network steadier for everyone.
- You choose what to giveYou set the limits. You can switch it off whenever you want.
- It gets checkedWhat counts is capacity that is genuinely useful and genuinely there — not just signing up.
- You share in what it makesWhat comes back is based on what you actually contributed.
WHAT CONTRIBUTION COULD LOOK LIKE
How many devices could you contribute?
2 devices
WHY A NETWORK AT ALL
This isn't charity. It's the whole design.
Lending your spare capacity sounds like doing someone a favour. It isn't. The network is the reason the rest of it is possible — take it away and Home James becomes just another company with a data centre and a monthly bill.
It's how it stays free
Running AI for millions of people in data centres costs an enormous amount. That cost is exactly why every other assistant either charges you monthly or sells what it learns about you. Spread the work across machines people already own and the bill gets much smaller — small enough that people don't have to be the product.
It's how it stays private
A data centre is a single warehouse holding millions of people's lives. That is one place to breach, one company to subpoena, one asset to sell. A network with no warehouse has nothing of the sort to lose, because your context never left your devices to begin with.
It's how it stays yours
Software that runs on a company's machines can be switched off, sold, repriced, or quietly changed. Software running on machines that company doesn't own is a much harder thing to take away from the people using it.
HOME JAMES CREDITS
Your computer could earn its keep.
Most computers sit idle most of the day. A gaming PC with a good graphics card spends its life waiting to be used. The idea behind Home James Credits is simple: when you are not using that power, the network could put it to work — and you get something back for it.
- Your computer helps outOnly when you let it, and only with what you are not using.
- You earn creditsThey add up in a simple Home James wallet, based on what you actually contributed.
- You spend what you earnedThe long-term goal is a Home James card, backed by a bank partner, that works wherever cards work — groceries, gas, the light bill. You would just tap and pay.
Why this matters
Right now a computer is something you spend money on. This would make it something that can earn. Instead of being customers of the digital economy, the people who make the network work would own a small piece of it.
To be clear about where this stands: none of this exists yet. There is no wallet, no card, no credits, and no way to earn anything today. Paying people for computing power and letting them spend it like money means working with banks and payment networks and holding the right licenses — that work has not been done. This is the direction Home James is being built toward, not something being offered. Nothing here is an offer to pay anyone, or a promise of income.
FOR THE TECHNICALLY MINDED
The questions you're actually asking.
If you run your own hardware you have specific concerns, and vague reassurance is worse than none. Here is where each one genuinely stands — including the parts that are still open.
What would actually run on my machine?
Settled: only eligible, sandboxed workloads, and only within the limits you set — CPU, GPU, disk, bandwidth, and hours. Participation is opt-in per device and revocable at any time.
Still open: the exact workload classes. Inference for other users' requests, storage, and routing are the obvious candidates; the isolation model and the scheduler that decides what lands where are still being designed.
Could someone else's personal data end up on my drive?
Design principle: no. Personal context is meant to stay on its owner's devices — that is the entire premise, and a network that quietly redistributed people's private data would contradict it.
Still open: anything stored across the network would need to be encrypted client-side and split so that no single node holds anything meaningful or readable. That scheme has not been finalised, and until it is published and reviewed you should treat it as an intention rather than a guarantee.
How is contribution measured, and what stops people faking it?
Settled in principle: reward measured, verified, useful capacity — not enrolment. A node that claims to be there and isn't should earn nothing.
Still open: proving work was really done, by the machine that claims to have done it, without a trusted middleman is a genuinely hard problem and the honest answer is that it is not solved here yet. Any reward rules will be published in full before anyone is asked to contribute.
What runs locally today versus in the cloud?
Running now: a working local node — a kernel that holds identity, permissions and policy; encrypted local memory; the privacy firewall; evidence and tool handling; and voice. Local models run through Ollama.
Uses external services: tasks that need reach or scale the local machine cannot provide. When that happens the privacy firewall sends the minimum required for the task and withholds the rest.
Local-first, not local-only — and the site says so rather than implying otherwise.
Am I locked into one AI vendor?
No. Models are treated as replaceable workers. The kernel routes each task to whichever local or cloud model fits on cost, privacy and quality, behind one interface and one permission system. Swapping a model out does not change how the system behaves, because the model was never the thing in charge.
Why should I believe the "rules are in charge" claim?
Because it is an architectural choice, not a policy promise. The language model proposes; a separate kernel holds identity, permissions, memory and the right to act. A model that is wrong, jailbroken, or swapped for another still cannot act outside what the kernel permits — the failure mode is a bad answer, not a bad action.
The honest caveat: you are currently taking that on trust, because the code is not public. That is a fair thing to hold us to.
Turning it around
Most platforms make their money from your attention and your personal information. Home James is built to make its money selling useful computing services instead — and to send part of that back to the people whose machines provided them.
No contribution program is open, and no reward rules exist yet. This slider illustrates scale only — it is not an earnings estimate, and nothing here is an offer to pay for capacity. The rules will be published in full before anyone is asked to contribute anything.