EvolveR: Self-Evolving LLM Agents through an Experience-Driven Lifecycle

In three linesEvolveR is a framework enabling LLM agents to learn from their own experiences through a closed-loop lifecycle. It combines offline self-distillation (extracting strategic principles from interaction trajectories) and online interaction (retrieving principles to guide decisions). Tested on multi-hop QA benchmarks, it outperforms existing agentic baselines.

Summary generated by Claude — human-verified