Agentic memory system shares private and public data across users
AIM: A Privacy-Aware Interoperable Memory Framework for Multi-Agent Multi-User LLM Systems
Artificial IntelligenceMachine Learning
Summary
Large language models usually forget what you talked about once the chat ends and don’t remember your likes over time. The authors developed AIM, a system that helps many users and AI agents share memory safely by separating private info only a user can see from public info everyone can use. AIM makes sure private memories are kept secret while useful shared memories help AI agents work better. They also created MUMBench, a test set to check how well these systems remember, update, and remove info when many people interact.
What this means in practice
- •For chatbot developers: Integrate AIM to manage and protect multiple users’ private and public conversations for more personalized and consistent responses.
- •For customer support teams: Use AIM-enabled systems to share common knowledge while keeping individual customer data private, improving teamwork across support agents.
Authors
Zachary Johnson, Nigel Boachie Kumankumah, Somya Chatterjee, Tejas Sathyamurthi, Min Chen, Xinyi Alice Li, Xiao Wang, Emily Morgan Gelchie, Jessica Lin, Sadid A. Hasan, Sulaiman Vesal
Abstract
Traditional large language models (LLMs) are scoped to individual user sessions, limiting their knowledge to a single conversation and preventing them from learning user preferences that evolve over time. Existing agentic memory systems address this limitation but generally operate at the individual-user level, restricting the public knowledge that could be shared across users to improve downstream responses. We introduce AIM (Agentic Interoperable Memory), a unified, privacy-aware memory framework that enables multi-agent, multi-user LLM systems to persistently manage private and shared memory. AIM dynamically classifies information as private, scoped to one user and inaccessible to others, or public, accessible to all users. It enforces index-level access controls so that private memories are retrievable only by their owner, protecting sensitive data while allowing beneficial shared knowledge to improve coordination and consistency. We also introduce MUMBench (Multi-User Memory Benchmark), a dataset of multi-user interactions containing private and shareable information across four domains. To our knowledge, MUMBench is the first public dataset designed to evaluate multiple memory operations, including retrieval, creation, update, and deletion, in a multi-user environment. Across three independent runs on MUMBench, AIM achieves 96.0% visibility classification accuracy, 58.8% strict operation accuracy, and 70.5% state-aware operation accuracy.