dwrld is a capability-gated application platform and workstation. It boots its own OS layer in your browser and serves a whole catalog from a tiered content-addressed store — identity, on-demand services, and GPU offload, all behind one front door. Held surfaces sit behind a passphrase and object-and-relation capability grants; the wallet is not what opens them yet.
Every node is a real surface, and every node that can be measured is — a live round trip, refreshed every 30 seconds. The hollow ones are held behind a capability: counted, never named, and deliberately never probed, because a latency is itself a signal. Drag to turn it.
Seven further surfaces are held behind a capability — the vault, the council, the workstation, the machine room. They are not linked here on purpose: a door you can see is honest, a door you click into a password is not.
The operator's product: turn machines, apps, storage and identities into one capability-governed environment. Deploy an app as a signed manifest; serve hundreds of dormant apps on demand; recover a dead app from cold backup and make it live.
The user's product: enter with a wallet, get a scoped account + vault + the apps you're granted, and use every project as one coherent workstation — boot screen, orb menu, live console, 2D map (3D later).
Identity control plane (Ed25519 capability grants) · Zanzibar-style authz · content-addressed tiered store (RAM → SSD → 1 TB cold) · on-demand socket-activated services · reverse-tunnel GPU offload · one nginx front door. Backups, monitoring, one-mutation rollback — run like production.
The protocol is dumb mail; the server is the gated dispersion centre. Capability lives in the server, never the transport — so authority is one auditable layer, revocation is a single epoch bump, and the transport stays reusable.
Unaided data structures & algorithms — a daily no-AI hour, so I can defend the reasoning as well as the systems. The goal isn't to lean on agents; it's to out-think the problem first, then direct the fleet.
Distributed systems, capability security, and the parts of my own stack I want to defend line-by-line — the netcode tick loop and the capability-grant verify path. I built these with agents; I'm learning to own every decision in them.