Cellium Agent is an open-source Python framework designed for developers to build autonomous AI agents with sophisticated, adaptive decision-making capabilities. It stands out for its lightweight, local-first architecture, avoiding heavy external dependencies and complex abstractions, which makes it easy to understand and extend. Key features include an adaptive decision loop (Plan-Execute-Observe-Replan) for continuous learning, Bayesian bandit-based strategy optimization for efficient task execution, and a three-layer memory system (persona, session, long-term) for robust context management. Its event-driven microkernel design further enhances modularity, and notably, it requires no external vector database, streamlining development for those seeking a lean, self-contained agent solution.