ZenBrain Introduces 7-Layer Memory Architecture Designed to Mimic Human Cognition

ZenBrain, an open-source AI project released under the Apache-2.0 license, has developed a memory system called HiMeS (Hierarchical Memory System) that organizes information across seven distinct cognitive layers. Unlike conventional AI memory, which stores all data as flat vectors with no sense of time or relevance, HiMeS incorporates mechanisms for temporal decay, memory consolidation, and active forgetting. The seven layers range from working memory for immediate task focus to long-term memory that uses Ebbinghaus-style decay curves to gradually fade unused information. During idle periods, ZenBrain runs a sleep-consolidation process that replays high-importance memories, strengthens connections between related knowledge, and prunes weaker associations. The approach draws on neuroscience research, including work by Stickgold and Walker on sleep-based memory consolidation, to make AI recall more context-aware and human-like.