Can your AI agent remember your secrets without the cloud ever seeing them?

MemPrivacy: Privacy-Preserving Personalized Memory Management for Edge-Cloud Agents

As LLM-powered agents move to edge devices, they face an unexpected constraint. These systems live on your phone or your company’s server, but they need the cloud to do anything sophisticated: form long-term memories, retrieve past interactions, reason over complex context. The problem is that sensitive information keeps flowing upward. A healthcare app remembers “patient has diabetes and anxiety, lives with partner who works in cybersecurity, concerned about medication costs.” An e-commerce system tracks “allergic to shellfish, recovering from divorce, buying gifts for new partner.” All of this is task-relevant for personalization. All of it is deeply personal.

The obvious solution is masking. Replace specific details with generic placeholders. Diabetes becomes [MEDICAL_CONDITION]. $200 monthly becomes [FINANCIAL_METRIC]. The cloud never sees the actual values, so privacy is protected.