Heartwood Memory blog
How governed agent memory works in practice: policy before ranking, what leaves the store when a model ranks recall, the audit record, and erasure with proof.
Subscribe by RSSJev shadow mode still sends your data
In Jev shadow mode your local ranking is kept and Jev's scores are only recorded. The query and candidate memories still go out to TypeSafe AI.
What to record when a model ranks memory
An agent memory audit trail for ranked recall: which ranker ordered the results, the egress decision, a hash of the request, and no query or memory text.
What Jev sees when it ranks agent memory
What a memory tool sends to TypeSafe AI when Jev ranks recall, what a trial still sends, and what Heartwood Memory holds back. Eight questions to ask.
Policy before ranking: agent memory and Jev
In permission-aware RAG, access rules run before ranking, because a reranker reads what it scores. How Heartwood Memory reranks recall with Jev.