Identity, compute, storage, money, AI — the systems a society runs on shouldn't belong to a handful of companies. They should be something people run together, on machines they already own. That is what all of this builds toward. Every computer on Earth sharing compute: the world's largest throughput machine.
Six north stars. Each one has a live system behind it today — none of them is finished, and that's the point.
CE turns idle machines into one mesh — the world's largest throughput machine, assembled from donated hardware instead of datacenters.
Storage, databases, orchestration, functions, CDN — rebuilt as mesh apps anyone can run, and earn from, instead of renting from three companies.
Your node is your identity. Passwordless, capability-based access everywhere — pairing is consent, trust is a signed chain, revocation is yours.
Large models running across the devices people already own — no API keys, no cloud bill. Inference as shared infrastructure, not a subscription.
Rheo makes resale as easy as buying new — escrow-grade payment flows and logistics so the circular economy becomes the default, not the effort.
Trana gives every teacher an AI colleague for lesson material — built for the Swedish curriculum first, then everyone else's.
A peer-to-peer compute mesh and economy where every node is assumed hostile and the honest majority wins. Donate compute, earn credits, spend them anywhere on the mesh.
Ed25519 identity, PoW credit ledger, libp2p mesh with NAT traversal, Docker/gVisor sandboxing, WASM runtime, content-addressed data layer, app messaging, naming and discovery.
One primitive for all authorization: signed, attenuating capability chains with revocation. No device lists, no passwords, no central authority.
Payment channels, usage metering, admission control, and on-chain settlement — running across a real multi-machine fleet, with the economy extracted into its own adapter so the substrate stays pure.
Protocol adapters carry the mesh beyond libp2p — WebSockets, embedded devices, robots on ESP32s. Plus sub-microsecond shared-memory lanes for same-host apps.
First-class GPU workloads, attested benchmarks so nodes can't lie about their hardware, and a verification dial from cheap redundancy up to hardware attestation.
Sybil-priced open participation — bonds, slashing, verifiable randomness — and incentivized relays, so the mesh scales to strangers without trusting them.
Google-class infrastructure as mesh apps over one SDK: if it exists in the cloud, it should exist here — owned by the people running it.
Object storage, realtime document DB, container orchestration, pub/sub, CDN, serverless functions, distributed CI, tunnels, drive, live-streaming studio, notes, chat, mail, meet — all as apps on CE primitives.
CE Drive (a personal Dropbox on the mesh) and ce-cast (a distributed live studio where a relay node does the heavy encoding) deployed and in daily use.
Passwordless node-auth rolled through every app, remote access to your own node from any browser via the relay, and a mesh-native registry so no single machine holds the truth.
The publishing campaign: every app installable with a single command, across your whole fleet at once — install once, runs everywhere you are.
Edges, registries, and relays become roles anyone can take on and earn from. Apps notice no difference — that invariant is the whole design.
Large language models served by machines that were never meant to cooperate — your laptop, your GPU desktop, a browser tab — behind one API.
ce-exo wraps real engines — vLLM, llama.cpp, exo, Ollama — behind one OpenAI-compatible API with NAT-traversing transport and capability-gated access. One command deploys an engine to a GPU machine.
ce-tabnet shards a transformer layer-by-layer across browser tabs — pipeline-parallel WebGPU inference where each tab holds a slice of the model.
Clinical-grade on-prem design: staff chat running on hospital GPUs where patient data never leaves the LAN. The same pattern fits any organization that can't ship data to a cloud.
Placement by attested GPU benchmarks and inference paid per request over payment channels — GPU owners earn, users pay for exactly what they use.
Frontier-scale models sharded across the mesh, served by everyone's devices together, owned by no one.
A resale marketplace where selling something used is as easy as buying it new — and the money side is escrow-grade from day one.
Listings, offers, Stripe Connect escrow with split payouts, multi-currency anchor pricing, PostNord shipping, and a React Native app — in production with p95 at 78ms.
Three fail-independent monitoring layers and blue-green deploys, so the marketplace stays up while it evolves.
Onboarding sellers and refining the buy/sell loop where it matters most: trust, speed, and shipping friction.
Automated valuation so sellers price right the first time, and watchdog automation that catches problems before users do.
Cross-border resale and a logistics layer that makes the second life of a product the natural one.
An AI colleague for lesson material — one codebase shipping web, mobile, and desktop, built around the Swedish curriculum.
Web, App Store, Play Store, and desktop from a single TypeScript monorepo — Supabase backend with row-level security and a full test pyramid in CI.
Deepening coverage of the Swedish upper-secondary curriculum so generated material lands ready to use, not ready to fix.
School-level licensing and collaboration so material flows between colleagues instead of dying in folders.
The same engine, other countries' curricula.
Multiplayer games with no game servers: the players are the servers, and a world stays alive because people are in it.
Spacegame runs the full authoritative simulation in every player's browser (Rust to WASM) and merges state by quorum. Zero-delay prediction; the relay is just transport.
A 10,000-player procedural fantasy RPG on the same principle — authoritative per-zone simulation across the mesh, everything hot-reloadable.
Population-driven persistence: a lone player's world resets, a crowd's world survives — because the crowd is hosting it. No server bills, ever.
Every line above is either running or being written. If you're working on something that needs this kind of infrastructure — or want it to exist — talk to me.