Domain roles · invitation preview

A role in a domain — not a node you can take away.

Structure, responsibility and context — something you take part in, not something you take with you.

Today it proves that open-source code actually works: it runs each package's own test suite and keeps only the reproducible result as evidence. 1,997 proven results so far — across Rust, Go, Python, JavaScript and PHP — and every one you can re-check on your own machine.

Zynko isn't an LLM — no GPU farm, no terabytes. It proves what it knows, and it stays small enough to run on your own hardware.

Today's AI recomputes everything, every time — burning power over and over for answers it already found. Zynko proves an answer once — however long that first proof takes — then never runs it again: the result is kept and reused, forever, at no further cost. The saving isn't a faster chip; it's the thousand re-runs that never happen. Not a promise — something you can check.

Invitation-only for now · a role is bound to your key and to the domain — non-transferable, and it expires

What you get — and what you don't

You don't get a node — you get a role. The node stays, but as the verification step, under the domain's rules, not as a tool you can carry off.

RoleCan doCannot do
Observer reads capsules and status, checks chain integritydoes not validate, write or forward
Validator runs the deterministic check — this is the node roledoes not create capsules, does not grant roles
Relay forwards capsules, builds topologycannot modify a capsule — any change is a different hash
Domain-Maintainer owns the domain's field set, grants and revokes roles cannot overwrite a sealed capsule; cannot grant for another domain

Why it cannot be taken away

Not because we encrypt it — because a copy cannot be continued. Every capsule hash covers the domain's identifier and the previous capsule; without the domain's key you cannot produce a valid successor to a copied chain, and it cannot be passed off as a different domain either.

What we do not claim: copying bytes cannot be prevented, and anyone can start their own domain with this same code. What the system guarantees is that it is not this domain.

The "prove once, never recompute" argument above has a name: dedup. The system asks by the content's fingerprint whether it already exists, and adopts any result it finds after verifying it — it doesn't recompute, but it doesn't take it on trust either.

See it answer — and check it yourself

proven, not guessed

The shape of what you get — an illustrative example of the format, not a specific claim. A real answer always carries its own re-runnable proof.

You ask

"Does acme-lib@2.4.1 still pass its own test suite on my platform?"

Zynko answers — proven

"Yes — 1,240 / 1,240 checks pass, re-run and byte-identical, with a result id you can look up."

You verify

Run the same check on your own machine → the identical result. You never take our word for it.

Capsule

The executable truth.

In the Capsule House, information becomes precise.

We don't interpret — we confirm.
We don't guess — we demonstrate.
We don't approximate — we reproduce.

A capsule is a new kind of unit: reliable, repeatable, verifiable.

Step inside and see how truth is made.

Enter

What Zynko can do — right now

  • Prove that real code works. It takes an open-source package, runs the package's own tests, and confirms the result really holds — so you get code that works, not code that only looks right. Done today across Rust, Go, Python, JavaScript and PHP.
  • Check security the same way — findings are demonstrated, never merely claimed.
  • Never asks you to take its word. Every answer it gives, you can re-check on your own machine — you verify it yourself.
  • Stay deterministic. The same question gives the same proven answer, every time — no guessing, no hallucinated confidence.

Where we're heading

  • Open the network to your machines — invited partners, verified nodes. That's this page.
  • A factory per language, running in parallel — proven first for Rust and Go, now copied and retargeted language by language.
  • A mini-civilization per subject — code, math, astronomy, biology… each with its own knowledge and its own human proposals channel.
  • Zynko writes its own tools. People teach; the system increasingly builds and repairs its own code.

Now — your plan

Zynko grows in the direction people point it. Every subject can become its own mini-civilization, and every one has a proposals channel — a place to say what's missing and where to go next. An idea starts as a hypothesis; Zynko turns it into a proof the moment there's a precise way to test it. So — what would you want it to prove?

Join with your machine

invitation · lifetime access

We're opening the network to invited partners who lend a machine — a spare GPU box, a server, even a modest node. In return, founding contributors keep free, unlimited access for life. Every machine that joins makes the network stronger — its combined compute is what lets Zynko grow. Your machine stays yours: its private keys are generated on your side and never leave it.

Lifetime accessfounders

Take part while the network is being built, and you keep unlimited access to Zynko for life.

Your keys never leaveby design

The onboarding script generates your node's private keys on your own machine. We never receive them.

Any decent machinelow bar

A worker node needs only 2 vCPU, a few GB of RAM and some SSD. No datacenter required.

Apply to join

Already taking part? See your dashboard ↗  ·  Cautious? Read the node package first ↗ — it's plain and inspectable.

Have a GPU? That's where video lives

generation runs on partner GPUs

After all of that this might sound like a contradiction — so let's be exact. Running the network needs no GPU; the deterministic, verifiable work runs on ordinary CPUs, and that's the whole point. Generation is the one exception. Video, image and audio generation is physics-bound to the accelerator — we don't pretend to do it on a CPU, and we say so plainly. It runs on partner GPUs. That's a partition, not a race we claim to win: the provable work is ours, forever; the creative generation stays where it belongs. If you have a GPU sitting idle, that's the compute the network actually needs, and it earns you the same lifetime access.

Video & other generationneeds GPU

Generative capabilities — video above all — run on the GPUs partners lend. This is the real compute the network is asking for.

Same lifetime accessfounders

A lent GPU counts like any contribution — founding partners keep free, unlimited access for life.

Your card, your machineby design

Your GPU stays in your own box, does only the work it's given, and its keys never leave it.

Safe for your machine — and safe for the network

you never take our word for it

Joining costs you nothing but spare compute. Your machine is never asked to hold anyone's secrets, it only ever sees its own task, and nothing it produces can quietly harm the network — because every result is checked before it counts. And you never have to trust us on any of this: you can verify it yourself.

Every result is checked

Nothing a machine produces is taken on faith. A wrong or bad result is caught and dropped before it can affect anything — good work counts, bad work goes nowhere.

Your machine stays walled off

An invited machine only ever sees the task it's working on — never the rest of the network. It can't reach, read or touch anything else.

You hold no secrets

Your machine gets only the work it needs — no keys, no private data. Its own keys are created on your side and never leave it.

How onboarding works

invitation → verified in five steps
Invitation

You get an invite and a one-line setup command.

Run one script

Read it first — it's plain, versioned and checksummed. It sets your machine up; your keys are made locally and never leave.

Connect

Your machine joins with restricted, task-only access.

Produce

It pulls work and produces results, on its own schedule.

Verified

Each result is checked before it counts — then it's on the record.

Apply to join

What a node needs

start with three · grow to a cluster
Worker nodethe low bar
2 vCPU · 4–8 GB RAM · 50–100 GB SSD · Docker / Podman
Leader nodeone per cluster
4–8 vCPU · 16–32 GB RAM · 200–400 GB SSD · fast network
Auditor nodethe checker
2–4 vCPU · 8 GB RAM · 50–100 GB SSD
GPU nodefor generation
a modern GPU (a few GB of VRAM or more) — this is what powers video and other generative work
A regional clusterone of many
7–15 machines run a full region — the network runs as many regions as join

Every subject gets its own mini-civilization

one world per topic

Zynko isn't one big brain — it's many small ones. Every subject grows its own little world, with its own proven knowledge and its own people deciding what to explore next. Astronomy, code, mathematics, biology, video — and whatever you bring.

Zynko-Astroastronomy
Zynko-Codeprogramming
Zynko-Mathmathematics
Zynko-Biobiology
Zynko-Videovideo
Zynko-Promptprompting
Zynko-AIAI methods

Add your idea

Every mini-civilization has a place for the people in it to point the way — to name a subject that should exist, or something worth proving inside one. What would your world be?

Invitation-only preview. A public intake opens once the first cluster is stable — everything above is the model we're building to, not a claim that it already runs at planetary scale.