Microkernel Architecture
Core is LLM-ignorant. Everything useful is a Paw or Skill. Zero opinions, maximum flexibility.
OpenVoleMicrokernel AI Agent Framework
The minimal loop that makes agentic behavior possible — nothing more.
Skip to content
27 Official Paws
5 Brain providers, 6 channels, 11 tools, 4 infrastructure — all subprocess-isolated with capability-based permissions.
OpenClaw Compatible
Same SKILL.md format, same ClawHub marketplace. Skills written for one work on the other.
Filesystem Sandbox
Every subprocess Paw runs with Node.js permission model restrictions. Network, filesystem, and child process access controlled per-paw.