AI agents forget — and when they do remember, nobody can prove what they stored, who could read it, or whether it was really deleted. Memory You Can Govern fixes that. Our agent stores every memory in MongoDB Atlas with a governing card attached: who may read it — and who may change or forget it — when it expires, and a receipt for every operation. You don't operate it with commands — you talk to it in plain words, and the box enforces every rule underneath; the model that phrases the answers is never shown a memory the asker may not read, and questions about the model itself are deflected as outside its scope, by design. Ask in a new run and it answers from the box. Correct a fact and the old version is superseded, on the record. A visitor is refused a Keeper-only memory for reading and for changing, and the refusal itself writes a receipt. Forget a memory and the content is hard-deleted across every version of it: chat, direct search and vector search all return nothing, and even the receipt trail is scrubbed. Memory you can govern — and proof when it's forgotten.
Team
