Large-Scale Agent Orchestration Framework

MultiClaw is an open-source framework for building, orchestrating, and running large-scale agent clusters. It supports multi-agent cluster creation, multi-agent collaboration, agent self-evolution, and enterprise-grade security. It is the engine behind Beyond Skyrim experiments and the foundation of enterprise agent infrastructure.

Star on GitHub View docs

Core features

Orchestration, collaboration, and control for large-scale agent clusters

Large-scale agent cluster orchestration Multi-instance and single-instance multi-entity; smart scaling. Single node supports thousands of instances and millions of agents.
Multi-agent autonomous collaboration Admin global instance management and CEO-style collaboration; agents explore and complete tasks on their own.
Co-evolution Agents extend capabilities and accumulate experience and skills; global, company, and instance-level co-evolution networks.
Skill extension Seamless global skill library: desktop and browser control, video generation, copywriting, embodied AI, and more.
Enterprise control Agent identity design, resource and permission control, with human approval and audit where needed.
Enterprise availability Full observability from business to single task; failure recovery and continuous checks for seamless operation.

Quick start

Three commands to experience MultiClaw

pip install multiclaw
multiclaw init my-first-cluster
multiclaw run examples/social_simulation.yaml
View full docs

Contribute

MultiClaw is open source. We welcome bug reports, feature ideas, docs, and code. See the contributing guide to get started.

MultiClaw is derived from the open-source project ZeroClaw, inheriting its core design, with major upgrades in large-scale scheduling and security.

View CONTRIBUTING.md

Contributors

Thanks to everyone who contributes to MultiClaw