Holos: A Web-Scale LLM-Based Multi-Agent System for the Agentic Web
A 23-author team releases Holos, a massive AI agent ecosystem designed for long-term autonomous evolution.
A large research team of 23 authors has introduced Holos, a foundational system for the 'Agentic Web'—a proposed future internet where persistent, heterogeneous AI agents autonomously interact and co-evolve. The system is a web-scale LLM-based multi-agent system (LaMAS) designed to overcome critical scaling issues like coordination breakdown and value dissipation that plague current multi-agent AI. Its core innovation is a five-layer architecture built for long-term ecological persistence, aiming to bridge the gap between individual agent collaboration and large-scale emergent behavior.
At its heart, Holos features three key modules: the 'Nuwa' engine for high-efficiency agent generation and hosting, a market-driven 'Orchestrator' for resilient coordination between agents, and an endogenous value cycle to ensure incentive compatibility. This structure allows the system to manage complexity at a massive scale, moving beyond isolated task-solving agents toward a continuously evolving digital ecosystem. The team has publicly released Holos, providing both a practical resource for the AI community and a crucial testbed for future research into self-organizing agentic systems, which they see as a pivotal step toward more general artificial intelligence.
- Designed for the 'Agentic Web', an ecosystem where AI agents autonomously interact and co-evolve persistently.
- Features a five-layer architecture with a 'Nuwa' engine for agent generation and a market-based Orchestrator for coordination.
- Publicly released as a testbed for large-scale multi-agent system research, addressing scaling and coordination challenges.
Why It Matters
Provides a foundational framework and testbed for building the next generation of scalable, self-organizing AI agent ecosystems.