Open research, openly published.

The umbrella track. What open, non-proprietary infrastructure for AI looks like across compute, coordination, and economic layers — and how to keep the substrate that powers AI from collapsing into a small number of providers.

Standards for agent identity, scoped delegation, capability boundaries, and trust establishment between agents and the humans, institutions, and other agents they act on behalf of.

Architectures and protocols for AI that isn't bound to hyperscaler concentration — distributed training, inference, and hosting, along with the economic rails that make participation possible.

Attestation, audit trails, reproducibility, and cryptographic verification of what agents and models actually did — the technical foundation for accountability and oversight.

Local-first AI, edge inference, and the coordination layer that lets distributed agents work together without a central broker.

The standards work itself — protocols, specifications, and interoperability for the open agentic web, developed in the open through working groups and public review.