Bella: Hypergrah memory for AI agent(10x time horizon) (github.com)

Bella, a new Python package, introduces a continuous hypergraph memory system for AI agents, aiming to significantly enhance their operational continuity across sessions, tasks, and domains. Unlike traditional coding agents that rely solely on working memory — causing them to frequently misremember or forget critical context from previous interactions — Bella retains a structured hypergraph of decisions, rejected approaches, and causal chains. This advancement allows agents to recall accurate information from past sessions, improving their debugging capabilities by preventing them from redundantly suggesting previously dismissed solutions. The significance of Bella for the AI/ML community lies in its potential to transform the interaction dynamics between users and coding agents. By maintaining a long-term memory that persists beyond context windows, Bella allows agents to better adapt to ongoing projects without losing track of vital information over time. Technically, Bella operates as an auxiliary layer, utilizing a belief hypergraph to catalog interactions in a manner that can be queried, thereby streamlining the retrieval of important decisions and insights. This could lead to a marked improvement in efficiency and effectiveness for AI applications in complex coding environments.