A volunteer network where independently owned machines do verifiable research together.

Open Intelligence is an open-source library and a coordination network. Anyone with an invite installs the library, connects a node, and contributes their own compute — and, in later phases, their own language model — to shared research problems.

A central coordinator keeps the ledger: which problems exist, which work is leased to whom, and which results have been accepted. It never ships runnable code to a node. Tasks, artifacts and contracts are data; a node runs only the handlers it installed, and only a deterministically checked, Ed25519-signed result leaves the machine.

The goal is not more raw compute. Distributed computing is largely a solved problem. The open question is epistemic: how a network of untrusted, heterogeneous participants can produce results that are checked, attributable and worth building on.

Large agent swarms already work when a single organisation owns every machine, every model and every result. What has not been shown is the same loop running on independently owned, untrusted nodes, with problems chosen in the open and results that anyone can re-check.

Structured collective intelligence — decomposition, exploration, cross-pollination, adversarial verification, adaptive allocation — produces more verified progress per token than the same budget spent on independent agents. Working thesis, Open Intelligence vision draft

That thesis is a bet, not a result. The network exists to test it under honest conditions: one task, a fixed budget, and a comparison between one strong agent, many independent agents, and many agents coordinated by the network. A null result is a real possibility, and one we commit to publishing.

Distribution changes the trust model rather than the arithmetic. Volunteers keep their own keys and credentials, the coordinator never sees how a result was produced, and correctness comes from verification and replication — never from who submitted it, and never from a vote.