Database
Browse and edit tables of the data layer directly.
showing 2 of 2 rows (first 500)
The database — exact facts, compared with vector memory
What it holds. Rows. Orders, customers, settings, prices — facts with a shape, where every value is exactly itself and nothing is approximate.
How it differs from vector memory. A database answers questions you can state precisely: which orders shipped last week, how many are unpaid, what this customer spent. Vector memory answers questions you can only state loosely, by meaning. One is exactness over facts; the other is similarity over language. Asking the database "something about returns" gets you nothing; asking vector memory "sum of March invoices" gets you a paragraph that mentions March.
What it costs. Nothing per query, and no model is involved. Filtering, counting and sorting happen in milliseconds and stay exact however large the table grows.
Where it is weak. It only answers what can be written as a query. It cannot find the passage that sounds like your question, and it has no opinion about meaning.
They work together. A vector record can point back at a row — its table and id travel with it. So a meaning-search over descriptions can end at the exact row, and the exact row can carry the text that made it findable. Same SQLite file, one backup, one wipe.