Infini Memory: Maintainable Topic Documents for Long-Term LLM Agent Memory

Infini Memory is a new persistent memory architecture for long-term LLM agents that organizes memory as topic-structured documents, facilitating evidence aggregation, fact revision, and memory maintenance. It employs a retrieval procedure allowing iterative tool calls during inference, leading to a 64.7% score on the MemoryAgentBench benchmark. This architecture addresses the limitations of existing memory systems, making it significant for practitioners aiming to enhance the maintainability and efficacy of LLMs in dynamic environments.