ActionRAG enhances LLM agent long-term memory with graph diffusion

Researchers have introduced ActionRAG, a novel graph memory architecture designed to enhance the long-term memory capabilities of LLM agents. This system addresses limitations in current memory pipelines by employing cautious indexing and query-time graph retroevaluation to refine evidence and ensure consistency. ActionRAG's pipeline reconciles type labels and applies tiered entity resolution, and at query time, it uses diffusion to refine triple scores based on local path structure. Evaluations on the MemoryAgentBench benchmark show ActionRAG achieving an average accuracy of 55.1, surpassing existing retrieval-based memory baselines by 5.4 points, while maintaining low synthesis costs. AI

IMPACT This research could lead to more capable and reliable LLM agents by improving their ability to recall and synthesize information over extended interactions.

RANK_REASON The item is an academic paper detailing a new architecture for LLM agents. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• ActionRAG
• alphaXiv
• Apple Silicon
• arXiv
• CatalyzeX Code Finder for Papers
• Connected Papers
• CORE Recommender
• DagsHub
• Gotit.pub
• Hugging Face
• Influence Flower
• Litmaps
• LLM Agents
• MemoryAgentBench
• ScienceCast
• scite Smart Citations

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →