Hive

Hive studies how to make multi-agent systems programmable and controllable, not just runnable. It provides an orchestration substrate where workflows, state transitions, and serving boundaries are explicit, so multi-agent behavior can be designed as a system rather than improvised through prompts alone.

programming surface
        |
        v
workflow and state control
        |
        v
agent serving and execution

The central idea is to separate what the agents should do from how the system coordinates them. That separation makes complex multi-agent applications easier to understand, extend, and operate.