Sovereign Knowledge Federation: A Boundary-and-Capability Model for Sharing Second Brains Between People and Their AIs

Personal knowledge management has matured into a stable practice. People now keep structured second brains in tools like Obsidian, and AI assistants can read and reason over those vaults directly through emerging connector standards. What does not yet exist is a coherent model for federating a personal knowledge base: sharing parts of one person's vault, with governed permissions, into another person's vault or another person's AI, so that knowledge can flow between sovereign individuals and their machines without being surrendered to a central platform. This paper develops such a model from first principles, as a derivation that peels the problem layer by layer: structure, then access, then federation, then trust, then identity, then governance. The central architectural claim is that structure, access control, and federation can be unified under a single primitive, the securable boundary, once the boundary is cleanly separated from the capabilities that grant the right to cross it. The central trust claim is that knowledge crossing a boundary should travel as a provenance-bearing claim envelope attached at crossing time, with the receiver computing fit-for-purpose assurance over that envelope rather than expecting the network to deliver a guarantee. Two of the paper's claims carry first-party evidence. A retrieval substrate built over a real cosmos-structured vault shows that sparse-first retrieval holds and that serving from a durable substrate is retrieval-neutral. A pre-registered label-conditioning study shows that the local models tested do not, in practice, down-weight a claim because it carries a low-trust label, which means trust in this architecture has to act as admission control over what enters the reasoning context, not as graduated weighting the model applies once the claim is already in front of it.