Raven is an open-source multi-agent ecosystem that automatically constructs and evolves modular ‘harnesses’ for specific models and domains, treating each model-harness pair as a composable unit of intelligence. A Host Agent decomposes goals, routes subtasks to specialized agents, manages execution dependencies, and integrates results, while a host archive, EverOS, and Skill Forge preserve and reuse experience across tasks. The paper provides theoretical sufficient conditions under which this composition expands reliable task coverage beyond any individual agent, and shows substantial gains over existing agent systems on long-horizon, cross-domain tasks.
