Show HN: FERNme – agent memory that updates with ~zero LLM calls

Category: library

Tags: agent-memory, ai-agents, hebbian-learning

Score: 6.3/10 (Innovation: 6, Technical: 7, Documentation: 6, Utility: 6)

FERNme is a memory layer for AI agents that uses a fuzzy preference graph with Hebbian co-occurrence and ACT-R decay instead of LLM calls for memory writes, achieving near-zero-cost updates. It compiles memories into a compact ~40-token card and uses spreading activation for retrieval, showing promising results on synthetic benchmarks. The project is interesting because it explores a more efficient, brain-inspired alternative to LLM-heavy memory systems, though it admits the mechanism is not novel and needs real-world comparisons.

Target audience: backend devs

View on Hacker News