What's actually different here

Most "personal AI" is a chat window glued to a vector database.

We rebuilt the whole stack. Atomic typed records. Cryptographic provenance signed at ingestion. Per-tenant row isolation enforced by the Postgres engine itself, not by a WHERE clause. Dozens of schema-aware parsers — the parser is the hard part, not the OAuth. An entity graph that lives in the same database as the records it indexes. An Agent API that asks the user's permission for every scope it reads. None of this is novel on its own. All of it together is the difference between an answer that sounds right and an answer that survives discovery.