Private community AI cluster · ✦ · Kubínska Scéna
Gotham 2.0
Thinking small is no longer an excuse.
The neighbourhood AI that lives at home. A brain of our own for the community and small businesses—boxes, cables, a cellar. No corporation between you and your AI.
Today: models running on two Spark boxes · the Gatekeeper is up · looking for 3–6 partners
Why
Today, we rent our intelligence.
Every question your family or your business asks lives in someone else’s data centre. Small communities and small businesses have become consumers of other people’s thinking.
What if neighbours built a brain of their own? Not metaphorically. Boxes, cables, a cellar—and rules we agree on ourselves.
The idea
A whole made of wholes
The machine computes.
It doesn’t understand.
Consciousness and meaning belong to people. Each of us is a monad—a unique whole that can’t be copied or dissolved. Monads join into larger wholes and don’t stop existing.
That’s exactly how we build Gotham: every member and every box stays its own—and together we’re more. AI is a tool in the service of people.
What it is
Three boxes, three roles.
The Gatekeeper lets people in, the Memory keeps the data, the Brain does the thinking—and each piece can be swapped without touching the others.
Gatekeeper
Lets in members only and hands out tasks. The only way to the Brain. Today it runs the chat, the files and the automation.
A small Linux server
Memory
Files, data, backups—for everyone. It remembers what matters, even while the Brain sleeps.
NAS · 64 TB · planned
Brain
Answers, writes, computes. It remembers nothing and never goes on the internet.
up to 8 × NVIDIA Spark
It all fits in one cellar and belongs to the community. No server room, no air conditioning—in winter, the cluster heats the place.
What it’s for
Four things you get.
Helper
Work, letters, translations, summaries, spreadsheets, help with school. For your business and your family, day and night.
Creativity
Stories, poems, games, ideas. An AI with time for silly stuff too—no counting credits.
Community
A shared project of neighbours and friends. We learn to work with agents together, and we have our own chat and shared files.
Privacy
Only your AI reads what you write. No big corporations, no ads, no profile of you.
The same class of models you know from the internet—it just runs here, and we play by our own rules.
What the Brain can do · as of 19 October 2026
Not one model, but a small team.
Five models run on two boxes. Each does what it’s built for.
- ConversationThe main model for writing, questions and thinking. In one conversation it can read hundreds of pages at once.main box
- ProgrammingA model for code. It reads even more at once, even a larger project.helper box
- Quick answersA small, nimble model for short questions.helper box
- Code completionA helper that completes code as you type.helper box
- Search by meaningFinds what you mean in your documents, even if you don’t use the exact words.helper box
Roughly fifty skills for agents.
The agents working on top of the Brain have learned procedures at their disposal. In four groups:
Software development
Programs in modern languages such as Go and Rust, desktop apps for all the common systems, 2D and 3D games including multiplayer, app design and graphics. Even new tools for the agents themselves.
Quality and security
Tests before code, trials in an isolated environment and in the browser, code review, programming to recognised security principles, protecting finished programs from reverse engineering, and careful handling of passwords and access. Methodical bug-hunting, and “done” only once it’s verified.
Integrations
Connections to other systems and devices, email, chatbots for everyday messaging apps including our [matrix], and automation of repetitive work.
Cluster and agents
Running and managing the cluster, even from a phone; planning before working; several agents at once; long tasks with a running log; and notes so the agents don’t start from scratch every time.
On top of that: fluent Slovak, answers based on official documentation instead of guesswork, brainstorming, and concise answers without the fluff.
Privacy
The Brain has no door out.
The Brain is shared—conversations aren’t. Private rooms in the [matrix] chat are end-to-end encrypted: not even the server can read them; only members hold the keys.
Privacy here isn’t meant to be a promise in the terms of use. It’s meant to be configuration. The Brain is designed with no route to the internet, so even a broken model would have nowhere to send data. A model can only propose an action—the Gatekeeper carries it out, after checking who’s asking.
Status: we’re building the Brain’s disconnection from the internet in phase two. Until it’s finished and proven by testing, we don’t claim it works.
The full build in numbers
Bigger on the inside, modest on the outside.
boxes
each smaller than a shoebox
of memory for AI
like eight top-end laptops in one brain
peak power draw
1,920 W—less than an electric kettle
people it can carry
estimate: 150–300 members, 40–80 of them at once
The figures are for eight boxes; the numbers of people are an estimate, not a measurement. The speed comes from MoE-type models, which use only a small part of themselves for each word—that’s why even a large model answers briskly on a single box.
Bring your own Spark
Bring a box—the Brain grows.
Your ticket in is a box of your own: an NVIDIA Spark or an ASUS Ascent GX10, 128 GB of memory, 240 W. You plug it into the shared Brain, but it doesn’t dissolve into it—it stays yours.
2 of 8 slots taken · 1,024 GB in total when full
- The power of the whole Brain, not just your own box.
- Your box stays yours—take it home whenever you like. Leaving means pulling a cable, not fighting over a contract.
- Two boxes per member at most—nobody takes over the Brain.
- Overheads are split monthly by the tokens used. No flat fee.
- On your own you have 128–256 GB. Together you have 1,024 GB—and the whole community benefits.
Social design
Rules as the architecture of a community.
Ticket in
1–2 Sparks per member.
At least one, at most two. The Brain belongs to everyone—and nobody takes it over. The box (about €5,000 incl. VAT) stays yours.
Running costs
Pay for what you use.
Electricity and overheads are recharged once a month by the tokens used. No margin, no flat fee, no surprises.
Community
Decisions are made together.
Rules, maintenance, shared know-how. A community isn’t an audience—every box and every hour of help makes the whole bigger.
We’ll agree the membership rules together. This is a proposal, not a price list.
Budget
The whole thing costs less than the village tractor.
From the first box to the full build of eight. The budget, formulas and all, is public—rework it for your own street.
The first two boxes, the Gatekeeper, encrypted chat and shared files.
The community Memory (NAS), isolating the Brain, and networking.
Growing to eight boxes—only if the community grows. Members bring most of it with their own Sparks.
A small business can buy one or two boxes on its own. Eight neighbours together have a brain none of them would have alone. It looks like building a cluster—it’s building a community.
Status
Where we are today.
Phase 1 now
The Gatekeeper and the first boxes
The Gatekeeper is up: the community [matrix] chat, shared files, automation and the first agent. Two Spark boxes work as main and helper, joined by a direct ConnectX link (100 Gb/s), and to the Gatekeeper over a 2.5 Gb network. They run open models.
Phase 2 next
Memory and isolation
The community Memory (NAS), cutting the Brain off from the internet and verifying it, and an encrypted path to the models. It starts once the first boxes’ performance has been measured.
Phase 3 vision
Eight boxes
The full 1,024 GB build, open to the whole community. Only if measurements confirm it will be used and there’s real interest—otherwise, nothing is bought.
Every expansion comes after measurement, not enthusiasm. Phase three is a vision, not a commitment.
Join in
This isn’t a manual for one village.
It’s a manual for your street and your business. The whole plan is public under a CC BY-SA 4.0 licence.
Come and see.
We’ll show you the boxes live—the hardware, the isolation and the whole path from question to answer.
Bring 1–2 Sparks.
The shared Brain grows by 128–256 GB. The box is an excuse to work together.
Pay only for what you use.
Monthly, by your own tokens. The Brain grows exactly as fast as trust does.
A whole in which no one is lost.
Your data. Your model. Your rack.