agentscope

Why we included this project

AgentScope is aimed at teams that want to see what their agents actually do before letting them loose on real systems. Every reasoning step, tool call, and message flows through a unified event bus, so you can watch an agent work and step in mid-run when something looks off. A configurable permission system gives you fine-grained control over which tools and resources an agent can touch, which matters once agents start executing shell commands or editing files. The framework also covers deployment concerns directly: multi-tenancy, multi-session services, isolated workspaces and sandboxes across Docker, Kubernetes, and remote providers, plus built-in support for MCP servers and skills. If you are moving from a working prototype to a deployed multi-agent service, the observability, safety, and deployment pieces are already there instead of something you have to bolt on later.