How xMemory cuts token costs and context bloat in AI agents

XMemory was developed by researchers xMemory was developed by researchers at King’s College London and The Alan Turing Institute.

Organizes conversations into a four-level hierarchical structure xMemory organizes conversations into a searchable four-level hierarchy of themes, semantics, episodes, and raw messages.

Performs top-down retrieval through the hierarchy xMemory performs top-down retrieval starting from themes and semantics, then drills to episodes or messages when needed.

Reduced token usage on some tasks from over 9,000 to about 4,700 tokens per query, researchers report xMemory reduced token usage from over 9,000 to about 4,700 tokens per query on some tasks compared to existing systems.

Improved answer quality and long-range reasoning across various LLMs while cutting inference costs, experiments showed xMemory improved answer quality and long-range reasoning across various large language models while lowering inference costs.