Semantic Memory

Overview

Semantic memory stores embeddings derived from text so agents can later retrieve related facts by meaning. A query like “recurring billing” can surface memories that never used those exact words.

Memory lifecycle

1. remember() embeds and stores content
2. recall() embeds a query and ranks neighbors
3. compress() optionally summarizes and archives clusters
4. forget() / clear() remove data

Content model

v1 content is text-only. The original content.text is never rewritten by the SDK.

content: {
  text: "Stripe supports recurring invoices."
}