Source record

Published by arXiv cs.MA (T1 source). The original is at https://arxiv.org/abs/2608.11248.

Pipeline notes

The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.

Paper proposing EvoGraph-Mem, a failure-aware editable graph memory for long-term language agents. Each insight node tracks positive/negative evidence and activation state. A graph controller updates the memory after task execution, keeping reliable insights, archiving invalid ones, and revising outdated ones. Experiments show improvements over append-only memory baselines.

Addresses memory pollution in long-horizon language agents with a concrete graph-based editing mechanism. Useful reference for anyone designing persistent memory in agentic systems.

Cited by

No citations on record.