Show HN: SOTA long memory eval with open source models (ensue.dev)

The Ensue team has developed a state-of-the-art memory retrieval system that achieves leading performance on the LongMemEval benchmark using only open-source models. By implementing a multi-stage retrieval architecture, the system achieves 96-100% accuracy in single-session categories, which is among the highest recorded. The innovative approach involves separating the retrieval process into ingestion and querying phases, leveraging fine-tuned classifiers and the Chain-of-Verification technique to reason over structured memory rather than relying solely on similarity matching. This advancement is significant for the AI/ML community as it demonstrates that open-source models can compete with proprietary systems, pushing the boundaries of what is possible with accessible technology while enhancing privacy and control over data. The architecture's ability to adapt and improve as open-source models evolve means that developers can build robust agents capable of efficiently managing and retrieving persistent memories, addressing critical challenges associated with memory degradation in long-term AI systems. This marks a pivotal step towards more efficient and personalized AI interactions across diverse applications.