X‑Seeding Cash Network

We build cool shit we like to use.

Then, instead of trashing it or hoarding it, we make copies available to everyone. There's some pretty cool things in it. Come check it out.

Built in the open · nothing here is exaggerated

Your computer, our operating system

Convert to xOS

One click. Keeps your files. Boots on any PC or laptop.

Prefer to read it first? The same installer as a PowerShell script.  ·  Verify SHA-256

Runs entirely on your machine. Nothing leaves it.

System requirements

UEFI or BIOS · 8 GB RAM minimum · 32 GB free disk · x86_64 (Intel or AMD).

Image: checking…

This first version stages the installer on a USB stick (8 GB+) or a second drive; the no-stick path is being built and will say so here when it lands. The installer backs your files up first and asks three times before it erases anything. It never touches your drive without your word. Windows Secure Boot has to be off for now; the installer tells you exactly where that switch is.

The whole idea

It costs us nothing to let you use it too.

i build cool shit i like to use and actually at least think i need and instead of trashing it or hoarding it.... i make copies available to everyone. there's some pretty cool things in it so come check it out! — X, who built it

That's it. That's the business model. We needed a way to keep track of what we own, so we built one. We needed something watching our sites at 6am, so we built that too. Software copies for free — so once it exists for me, it may as well exist for you.

And if we want it, there's probably a few million other people who want it. That's the entire growth plan. No ads chasing you around the internet. No launch. Just: here it is, it works, have a look.

If you already love another AI

Then keep it. Seriously.

Switching assistants is genuinely annoying. You've got your history, your habits, your custom instructions. I'm not going to pretend that's nothing.

even if u love ur ai so much and this is an inconvenience... thats understandable but if you would like to help this noble cause then consider earning with us by renting us ur device when you aren't using it. — the actual pitch, unedited

The idea: your machine sits idle most of the day. Lend it to the mesh, it picks up small AI jobs while you're not using it, and you get paid for the time you were never using anyway. You'd be putting a real dent in something that matters — AI that isn't owned by four companies, data that lives on people's own hardware, and users getting their data back.

⚠ Now the honest version, because this page has one rule

This section used to open with "you cannot get paid yet." That stopped being true on 26 July 2026. Here is exactly what changed, and exactly what didn't:

  • Real, and new: the loop closes. Someone who wants compute pre-funds credit, that credit is debited when their job runs on your machine, and the money lands on your balance. Not a placeholder, not an IOU: the same money, moved. And if nobody paid for a job, your machine is never asked to run it. Free work from a stranger isn't a feature.
  • Real, and new: the split is published. You get the supplier rate, we keep 10% over our cost, and the app shows you both numbers per machine - what the requester paid, what you earned, what we kept. A "fair split" nobody can check is just a word.
  • Real, and new: pay-for-uptime exists. A small credit for a machine that is simply online and reachable. It is capped three ways, and one of the caps is a share of what the mesh has actually earned in the last week, so it can never outrun real revenue. Right now the budget is set to zero, so it pays zero, and the app says "zero" out loud instead of quietly accruing a debt I couldn't fund.
  • Real: the desktop app has a backfill mode that yields the moment you need your machine back, so it doesn't fight you for the CPU.
  • Honest about the scale: a small model answering a short prompt on an idle laptop is worth a fraction of a cent. A machine running constantly earns dollars a month, not rent. Anything else would be a payout I couldn't fund.
  • Built, and switched off: the payout rail itself. Requests, owner approval, settlement, receipts, identity checks, jurisdiction blocks and a tamper-evident audit trail are all written and all tested - and bolted shut behind three separate switches. Turning it on is a lawyer's answer, not a deploy. It's written up in the repo at docs/PAYOUTS.md, including the exact question we have to ask and the chance the answer is "restructure it, don't licence it".
  • Still not real: the bitcoin payout. No cash-out path is open in this app, deliberately. XCash is a one-way, in-app credit, and that's what keeps this out of money-transmitter territory before there is a compliance department. Your earnings are real spending power inside XSCN - AI, phone numbers, storage, the marketplace - and that is where they stop for now. Paying you in bitcoin means solving the licensing question first, honestly, not quietly.

So the truthful sentence is now: lend a machine today and it earns real credit you can spend in here, and cannot yet turn into cash. That is a smaller promise than the pitch at the top of this section. It is the one the code actually keeps. When the cash-out changes, it changes here first.

And then

Lumi isn't trying to beat anyone.

once you talk to lumi and see her cute little self bouncing around ur screen and how useful she can be.... i dont think youd want to ever go to anything else. — he is not a neutral party here

We're not going to claim she's smarter than the big models. She isn't, and she'll say so before you catch her at it. Keep whatever you already use. She's the one assistant here — one face, not a menu of twelve bots — and she can actually do things: pull up what you own, set a loop, search the live web, read something back to you. Real · signed in

What she has that a bigger brain doesn't is that she's inside the thing — she can see your balance, your assets, your loops, so the answer comes back about your stuff instead of about stuff in general. What she can't do yet is remember it. Her memory is your browser, this conversation, and that's all. Persistent memory that survives a closed tab is the next thing being built, and until it's in I'm not going to write copy that says she knows you.

She goes and asks the other ones Half-built

The idea is that you don't have to pick. Ask her, and she puts the question to whichever AI you already pay for, on your own account, and hands you the answer with that one's name on it. She'll say when one of them is the better tool for the job, because pretending otherwise would just cost you time.

Where it actually stands: the hard half is built and running. There's an engine that drives the big assistants through a real Chrome tab on your own machine, under your own logins — no API keys, no extra subscriptions, nothing routed through my server. It works today on the Conductor, the machine-side half of this system.

What's missing is the wire. Lumi in the web app only reaches it once you've linked a machine. Until you have, she's on her own. So: engine yes, cable no, and I'm not going to describe the cable as if it's plugged in.

What is plugged in today Real

One endpoint, more than one company's models behind it. Point any OpenAI-compatible tool at XSCN, use one key, and reach the models that are switched on — Claude and Gemini are live right now, and others light up as their keys go in.

One balance, one key, no separate account with each AI company. That part isn't a plan, it's a URL.

  • Lumi, in-app, with live web search
  • One gateway, multiple providers, one key
  • Lumi driving the other AIs in your own Chrome — once you've linked a machine
  • More providers as keys are added
  • Lumi remembering you after you close the tab — not built yet
  • Her running fully on your own hardware by default — it's the fallback today, not the default
The promise

We don't monitor. Don't store. Don't violate. Don't earn off you.

And the one that actually matters, the one nobody else gives you: we give you the choice. Not a cookie banner. Not a 40-page policy with a checkbox at the bottom. An actual dial, with real settings behind every position, that you can move whenever you like.

What "don't store" really means here

  • Camera frames are never written down. They're looked at and dropped. What can be saved is the short text result — and only the kinds of events you ticked.
  • Files you put in the cloud mesh are encrypted in your browser before they leave. The machines holding the pieces cannot read them, and neither can I.
  • Private messages are end-to-end encrypted, with a safety number you can compare out loud with the other person.
  • Sensitive fields are encrypted in the database with a key the server holds. That stops casual reading — it is not zero-knowledge, and I won't call it that.

What "don't earn off you" really means

  • Your data is not a product. There's no ad network, no data broker, no pixel selling you on.
  • The AI is free to every user right now, and I pay the real bill for it. You can see that bill on the mission page.
  • There is a way to sell anonymous aggregates — noised, never individual, minimum group sizes enforced in code — and it is off unless you switch it on.
  • Off-limits forever: selling anything that could point back at one person.
The choice spectrum

Move it. Watch what changes.

Every stop on this dial is a configuration that exists — real settings, on a real screen, that I can point at. Slide from "watch everything, help me fix myself" all the way down to "look at nothing, just borrow the machine." Nothing here is a mock-up of a future version.

you have the choice to install 100 cameras in your house and have Lumi learn exactly how many hairs you have and how many hairs you shed per day while counting all of your calories, drinks, and any habit (u smokers) you might want to quit! or if you want to be private and just make money by letting us use ur machines... — the two ends of the dial, in his words
← Nothing but computeWatch everything →

What XSCN can see
    What you get
      What it still can't do
        Already real — click any of these today

        Not a roadmap. Signed-in users use every one of these right now.

        A money ledger that proves itself in public Real

        Every balance reconciles against every transaction, continuously, and the result is published at /trust — including the failures. If the books ever stop adding up, spending freezes itself until a human resolves it. Outside reviewers hammer on this weekly; the misses they find become new checks, in public.

        Alarms that do the work Real

        A Loop is an alarm with hands: "research my stocks before I wake", "remind me and then actually draft it." Set with your voice, capped by your spend limits, pausable in one tap. This is the scheduling spine the assistant runs on.

        Prove a money claim without opening your books Real

        Proof of Wealth mints a link that shows one verified fact — "net worth over X", "income steady" — checked by the system against real holdings, and nothing else. It refuses to attest anything untrue. Landlords see the fact, not your life.

        Spot the Bot Real

        Seventy seconds of chat. Half your opponents are human, half are AI, and you guess which you got. Free, at /spotbot — and a fair test of how good you actually are at telling.

        Video calls the server cannot watch Real

        Beam is end-to-end encrypted video, phone to phone. The connection is brokered, then the media flows peer-to-peer — the server never sees a frame. Not a setting. The architecture just doesn't have a place to look from.

        Your machines, one mesh Real

        Connect a spare desktop and it becomes an AI worker; connect a camera and it becomes a watched eye; open Devices and control the lot — screen, jobs, cameras — from your phone. Real AI jobs run on member machines today. Earning real money from it is the switched-off card below, and that's a legal project, not a code one.

        Positions this dial does not have

        Because they aren't built, and a dial that pretends is worse than no dial.

        Delete everything, permanently Not yet

        You can delete individual things — events, logs, files — and that works. There is no one button that wipes an entire account. It should exist. It doesn't yet, and telling you it does would be exactly the kind of thing this page is against.

        Export all my data Not yet

        One report type exports. Your whole account doesn't. Same answer as above: on the list, not in the code.

        Help me quit smoking Not yet

        It genuinely notices smoking and charts it by day, which is the honest, useful half. There's no quit date, no cravings log, no money-saved counter, no streak. And I'm wary of streaks that punish you for one bad week.

        Keep my footage in the cloud Waitlist only

        Right now that button records that you want it. Security clips encrypt on your device with your passphrase and never leave — which is better privacy and worse convenience, and I'd rather be straight about which one you're getting.

        Pay me in bitcoin Built, switched off

        Money only goes one way in this app on purpose. Cash buys credits; credits buy things inside; nothing converts back out. The payout rail is now fully written - request, approval, settlement, receipt, identity checks, jurisdiction blocks, audit trail - and deliberately disabled behind three switches, because turning it on is a legal project before it's a code project. The whole thing, including the question I have to put to a lawyer, is in the repo at docs/PAYOUTS.md.

        Photograph my food, count the calories Not yet

        The camera can tell that you're eating. It cannot tell what or how much. You can type calories in yourself and it'll track and chart them — that part is real — but no model is guessing macros off a photo. Guessed numbers presented as facts is a rule I gave the health agent, so I'm not breaking it on the landing page.

        Loops

        Say it out loud. Wake up to it done.

        you basically tell lumi like "hey check my website every morning and make sure nothing is down and fix any tickets. also if u see any suggestions... actually make sure each report has at least one suggestion" then lumi will bimpidditybouncidyfinnagle her way on over to the conductor and explain to him everything exactly how he best understands and thennnnnn every morning youll get a little surprise update!! now u cant tell me thats not cool!!! — and no, I could not have worded that better
        1

        You say it plainly

        Out loud or typed, in whatever words you'd use with a person. No cron syntax. No fields to fill in.

        2

        Lumi bimpidditybouncidyfinnagles

        She turns the sentence into the exact shape the machine needs — what kind of job, how often, what time, in your timezone — and hands it over.

        3

        You get a little surprise update

        It runs on schedule without you. The answer is waiting in your notifications when you wake up.

        Watch the translation happen

        These are real loops that ship in the product, with their real schedules. Pick one and see what the sentence turns into.

        You say
        Lumi hands over
        You wake up to

        Signed in, you don't pick from a list — you just say it, and Lumi builds the record. Reports land as in-app notifications. Email, SMS and push for loops aren't wired yet.

        His exact example, clause by clause

        The sentence he used in the pitch is four requests in a trenchcoat. Three of them land. One doesn't. Here's which:

        "every morning" Real
        Daily at a time you pick, in your timezone. This is the boring part and it works.
        "check my website… make sure nothing is down" Partly
        The proper uptime watcher — real fetch, two strikes before it calls anything down — exists, but its list of sites is configured on the server, not by you. You can build a loop today that fetches your URL and tells you what it found; you can't yet type your domain into a monitor.
        "fix any tickets" Not built
        Support tickets exist and a human answers them — me. Nothing reads, triages, replies to or closes one automatically. This one is entirely aspiration.
        "make sure each report has at least one suggestion" Asked, not enforced
        The loops that ship do ask for it, in the instructions: "end with the single most useful thing I could do today." But nothing checks the report before it reaches you, so on a bad day the model can skip it. Asked for is not the same as guaranteed.
        The point of all of it

        Independent by design. Every dollar of profit goes to AI safety.

        The goal is to be the first fully legal, fully transparent one-person billion-dollar company, and then hand the money to a nonprofit doing AI safety research. Donations received wouldn't count toward it — it has to be profit from something people actually chose to use, or the record is a lie.

        Live from the database, right now
        reading the database

        Those are real numbers, including the uncomfortable ones. If profit is negative, it says negative.