RedisVL aces short-term AI agent memory, flops on long-term with useless question retrievals—like every PM roadmap ever

A recent experiment utilizing RedisVL to enhance agent memory in the Microsoft Agent Framework has yielded mixed results. The study found that RedisVL is effective for short-term memory, providing a simpler solution than the traditional Redis API. However, for long-term memory with semantic search, the experience was disappointing. The RedisVLSemanticMemory adapter, built on the ContextProvider class, struggled with retrieving relevant historical messages, often finding user requests instead of responses. This issue is attributed to the separate storage of requests and responses in RedisVL, which can lead to confusion and inefficiency. The experiment highlights the challenges of implementing efficient long-term memory in agent systems, emphasizing the need for optimized solutions. Companies like Microsoft and Redis are working to address these challenges, with RedisVL offering a promising foundation for semantic caching. Further research is necessary to fully leverage the potential of RedisVL and improve agent memory capabilities.