Knowledge · internal system

Semantic Memory

Search by meaning, plus a knowledge graph that tracks who, what, when — and what has since changed.

Live No public endpoint Memory

The problem it solves

Keyword search fails exactly when it matters: you remember the shape of a thought but not the words you used. And flat notes cannot answer relational questions — what supports this claim, what superseded that decision, what was true in March but is not true now.

How it works

Index

the whole vault, continuously

→

Embed

meaning, not just words

→

Relate

typed facts with valid-from and valid-to dates

→

Recall

surfaced into a session automatically

→

Invalidate

superseded facts are retired, not silently overwritten

Under the hood

Every fact carries a validity window, so the graph can answer what was true at a point in time rather than only what is true now.
Prose with reasoning goes to the memory store; entity relationships with a time dimension go to the graph. Forcing a nuanced decision into a triple loses the why.
Replicated to every machine on a schedule, with lock-safe syncing that defers rather than corrupting.
Recalled memories are treated as background context, never as instructions — and are re-verified against live state before being acted on.

Why it matters

Memory that only accumulates is a hoard. Memory that expires, supersedes and links is a mind.