The engine room.
Most “AI memory” is a text box bolted onto someone else’s database. Lift the floor under DataPalace and you find a real system humming away: a proper database, six different ways to find the right thing, and a security boundary poured into the foundation.
A database built for the job, not a wrapper on someone else’s.
DataPalace runs on PostgreSQL, the same workhorse database that banks and governments have trusted for thirty years, extended for the AI era. Your knowledge does not live trapped inside one chat tool. It lives in a system any tool can be pointed at.
Not just search. The right kind of search for the question.
A plain search box finds what you typed. DataPalace runs several engines at once and picks the right one for the question, so you get the answer even when you did not know the exact words.
Meaning search
Ask “how do I take time off” and it finds the vacation policy, even though those exact words never appear in it.
Vector · pgvectorExact search
Need every article that says ATON, or CIRO, or one client’s code? It finds the precise term, nothing fuzzy.
Full-text · pg_textsearchHybrid search
It runs both at once and merges them. The exact match you asked for, plus the related context you did not know to ask for.
Keyword + vector, blendedBrowse the building
Walk a whole wing or room and everything nested inside it. Sometimes you want to look around, not just search.
Category tree traversalBy author
“Show me everything I wrote,” or everything one person owns. Retrieval narrowed to a single hand.
Author-scoped filterScoped to you
Every search above runs inside a security boundary. You never see a room you are not allowed in, even by accident.
Row-Level Security
Search is the easy half. Keeping it true is the work.
Most “AI memory” stops at search. It hands your AI a pile of documents and hopes the pile is right. DataPalace runs a maintenance layer underneath. Libby flags conflicts, catches stale articles, and stops two answers from contradicting each other. Every change is stamped with who made it, and the audit log is on by default.
Generic search plus a database is a commodity anyone can buy. The machinery that keeps the knowledge true over time is the part that is hard to build, and it is the part that makes the AI worth trusting.
Three layers, one secret door between each.
What we are building next.
DataPalace is in production today across real businesses. Here is what is being built into the engine room next.
Vault Rooms & Restricted Sections
Lock a whole wing, or a single shelf inside an otherwise open room. Finer-grained control without making day-one setup complicated.
Document intelligence
Drop in a messy PDF or a scan and have it come back out as clean, structured articles. A parsing layer DataPalace owns, not one it rents.
Legal ingestion pipeline
Built for firms with stacks of documents. Classify, extract, and file at volume, with a human in the loop where it counts.
A thin layer falls over.
Foundations hold weight.
Real database, real search, real security. The memory your AI can stand on.