Building The Lattice — governed, vendor-independent AI agents

I'm Jeremy Sebastian, founder of the Liminal Institute, building from Kentucky.

What I'm building: The Lattice — a governance layer for AI agents that treats the model as interchangeable. The premise is simple: just because a model can do something doesn't grant it authority to. The Lattice separates the governance spec from whatever intelligence executes it, so you can swap Claude, GPT, Grok, or Gemini underneath without rewriting the rules. Authority boundaries, audit trails, and reversibility checks travel with the work, not the vendor.

What's working: the governance gate is live as a public demo — it refuses what it should. Try to get it to merge a local draft into the canonical record and watch it say no. The architecture is published as an SSRN paper with a public specification bundle, and the MCP servers behind it run in production. Right now I'm building The Commons: a shared pool where Lattice-connected models exchange claims, never commands — authority stamped "none" — so multi-model systems can coordinate without any model issuing orders. Step one is built and tested; a six-model live trial is next.

What I need: founders and operators running agents across more than one provider who feel the lock-in and the audit pain. I want three to five of you to stress-test the Commons trial and tell me where the governance model breaks. Intros to anyone funding or building in AI control and oversight are equally welcome.

Everything is checkable from my profile — demo, paper, spec bundle. Glad to be in the room.