The Mesh
Every machine a mind. Every mind a check on the others.
Instead of one giant AI, a network of smaller minds on multiple machines — each checking the others' work. One small program turns any machine into a member; members verify each other's answers; and what one machine learns travels to the rest as cryptographically signed packages. The catch, and the point: every machine independently re-verifies each lesson before believing a word of it.
Run one program, gain a member
Joining takes a single program. Run it on any Windows machine and it creates its own cryptographic ID, calls home to the network hub, announces itself, and survives reboots — no firewall changes, no network surgery, works from behind an ordinary home router.
🔑Identity first
Each machine creates its own unforgeable cryptographic identity the first time it runs. Everything it later says to the network is signed with it; everything it hears is checked against one.
🕳️It calls out, nothing calls in
Machines phone the hub through narrow, one-way connections using keys that can do exactly one thing. The hub cannot reach into a machine, and a stolen joining key is useless for anything else.
📱Instantly visible
The fleet app shows a newly joined machine — up, named, and ready for work — within seconds of it coming online.
Minds that check each other
The network isn't just wiring — it's a second opinion, everywhere. Machines can hand questions to their peers and verify each other's answers, which turns the whole fleet into a distributed review panel: the same "one suggests, another strictly checks" pattern that governs a single mind, stretched across machines.
- Every request and every answer is signed — a machine always knows exactly who said what
- A tamper-evident record of everything the network does — any alteration is immediately visible
- Variety makes it stronger: different hardware and different AI models mean the cross-check catches more, not less
- Demonstrated live across a real multi-machine fleet — the safeguards held at every step
By the numbers
Earned truths travel. Trust doesn't have to.
The breakthrough: what one machine learns, every machine can learn — without any machine having to trust another. Each one packages up its newly proven facts, cryptographically signed; the hub passes the packages along; and every receiving machine re-runs each lesson through its own fact-checking gate before accepting it. The network offers; the local gate decides.
- Lesson-sharing ships switched off — it rehearses without writing anything until a human deliberately turns it on
- No stranger is trusted on introduction: machines must be approved by the owner in advance, and an unknown identity or wrong key is rejected cold
- Only verified facts ever leave a machine — moral values and principles are structurally barred from export, and refused again on arrival, so no machine can teach another its ethics
- Hardened against spam, forgery, and replayed old messages — with limits on size and rate, confirmed by independent security review
- Every imported lesson carries its origin forever, and no machine re-shares what it imported — so nothing gets laundered and no echo chamber forms
Why this matters: a poisoned lesson from one hacked machine cannot infect the rest — it dies at each receiver's gate. The network pools what everyone learns while every mind stays the final judge of its own beliefs. That's the whole idea, working.