Show HN: FERNme – agent memory that updates with ~zero LLM calls (github.com)

FERNme has introduced a revolutionary memory layer for AI agents that requires virtually no calls to large language models (LLMs). This user-owned memory system is designed to learn from individual interactions, tracking not just facts but nuances such as communication style and emotional state. Unlike traditional memory systems, which rely heavily on LLMs for updates and can lead to high costs and inaccuracies, FERNme operates entirely on arithmetic updates to a graph structure. As a result, it maintains a flat cost per interaction, regardless of how much data is stored, allowing for efficient, user-controlled memory management across various platforms. This development is significant for the AI/ML community as it shifts the paradigm towards more accessible and interpretable AI systems. FERNme’s architecture supports privacy and transparency by providing users with control over their memories, enabling them to see, edit, and manage their own data. The memory is also designed to enhance outcomes, as it reinforces successful interactions while adapting to changing preferences. By utilizing techniques like Hebbian learning and spreading activation, FERNme not only optimizes for cost but also ensures effective learning and responsiveness in diverse applications, potentially elevating user experiences across industries.