Insight for: Show HN: AI memory with biological decay (52% recall)

An AI memory system with biological decay, using the Ebbinghaus forgetting curve and a graph layer over a vector store.

Analyzed: Apr 29, 2026

This directly addresses critical limitations in current RAG systems, offering substantial improvements in efficiency and accuracy for enterprise AI applications. The biological decay mechanism and graph layer mitigate high token costs, context window saturation, and the 'logical neighbor' problem inherent in static memory approaches. Achieving 52% Recall@5 and an 84% reduction in token waste represents a significant performance and cost optimization. This innovation is vital for B2B SaaS platforms leveraging RAG, enabling more robust, cost-effective, and contextually aware AI agents for long-running projects. It signals a necessary evolution in intelligent context management for production AI systems.

RAG setups memory static filing cabinet context window token costs agent's reasoning biological approach Ebbinghaus forgetting curve living substrate strength score recall reinforces decay curve spaced repetition unused data pruned logical neighbor problem semantic search relevant but non-similar nodes graph layer vector store LoCoMo dataset Recall@5 stateless vector stores token waste local first MCP server DuckDB non-linear decay biological constraints context management

Hacker News Post

Parent Entity

Show HN: AI memory with biological decay (52% recall)

Score: 70