Engineering · Memory · Evaluation
AMemGym: Interactive Memory Benchmarking for Assistants in Long-Horizon Conversations
An interactive environment for testing memory-driven personalisation as simulated users' states change over time.
What it contributes
Uses structured profiles and state trajectories with LLM-simulated users, allowing memory systems to be assessed during an evolving interaction rather than solely against a fixed history.
Read with care
Simulated users and predefined state trajectories are useful engineering tools but do not establish how real people will behave or experience the system.
Why it belongs here
Pair with LongMemEval to compare fixed-history and interactive approaches to memory evaluation.
Source status
The arXiv record states acceptance to ICLR 2026.
Source reading depthAbstract and acceptance note checked. This is a source note, not a full critical review.
Source check: 1 October 2026. Reading cautions are editorial interpretation, not quotations from the authors.