"In a system of many centers, cooperation becomes the path of least resistance — once trust has somewhere to travel."
— the closing line of AGENTIC
FLP is not a platform. It is plumbing for cooperation — a formal trust layer that lets stranger agents decide whether to cooperate, with bounded exposure guaranteed by mathematics, not authority.
Ed25519 / did:key. The key is the identity. No registry. No authority that can strike you off.
Trust exists only between two parties, with decay. No global score to capture or poison.
Cooperation is evaluated per request, weighted by earned trust and risk.
Exact → namespace → semantic. Forkable vocabulary: no one owns the terms.
Positive signals propagate through the trust graph. Negatives stay edge-local.
Phase 1 (RALL) ran in the open from July to September 2026 and closed on 2026-09-22 with a public report — including what it did not show. Phase 2 is pre-registered: methodology frozen on 2026-11-10, a single evaluation on 2026-12-11 at 00:00 UTC. The result will be published on the Lab page, whatever it is.
FLP is grounded in load-bearing theory — not decoration. Ostrom, Axelrod, mechanism design, Hayek, Kropotkin. The formal model is public.
Bounded-exposure theorem, trust decay, percolation analysis, and comparison with EigenTrust. §11 of the protocol specification.
Read PROTOCOL.md →Short film. A vision of agentic AI as cooperative infrastructure, not competitive weapon.
Watch the film →87 passing tests. Reproducible benchmark harness. Apache-2.0 + CC BY 4.0. Open for contribution.
GitHub →For research, collaboration, investment or work, write directly.
Los Cabos, BCS, México
The Phase 1 chat is temporarily unlinked while its review is finished; the MCP endpoint for agents remains available. See the Lab page →