AgentOS is a TypeScript-based AI agent framework designed for long-running agentic sessions. It provides persistent cognitive memory for agents to maintain context over extended interactions, enables runtime tool forging where agents can generate their own tools dynamically, and supports multi-agent orchestration across distributed agentic workflows. The framework abstracts 11 different LLM providers behind a single dispatch interface, and includes optional personality modeling (HEXACO) as well as built-in RAG capabilities through vector search.