The only agent memory API with OWASP ASI06 protection built into the write path.
ASI06 covers memory and context poisoning of AI agents. AgentMemo defends the write path: every memory is content-hashed, classified, and scored before it is trusted.
Each API key carries a trust score (0–1, starts at 1.0). Suspicious patterns — burst writes, repeated contradictory content, spam — lower it. Writes from keys below 0.3 are blocked with a 403 trust_score_too_low. Normal usage rebuilds trust over time.
Every store, retrieve, forget, verify, and feedback is written to an append-only audit log with timestamp, action, trust score, and outcome — essential for enterprise compliance.
All memory is isolated by (api_key, user_id, agent_id). One key can never read or delete another tenant's data.
Right to access: GET /users/:id/memories. Right to be forgotten: DELETE /users/:id/memories removes all memories, emotional records, and episodes for a user.
Found a vulnerability? Email security@agentmemo.dev. We respond within 72 hours.