Hive

Hive (referred to in its own documentation as OpenHive) is an open-source, model-agnostic execution harness for running multi-agent AI systems in production rather than as one-off demos. Instead of requiring manual orchestration code, it compiles a stated objective into a graph-based execution plan (a DAG) that coordinates specialized agents to run tasks concurrently, backed by persistent, role-based memory that evolves with a project's context. The project positions itself for teams moving agents from prototype to production: it targets deterministic fault tolerance, state observability, self-healing/adaptive agents, human-in-the-loop control, and cost limits — explicitly contrasting itself with single-agent tools that it says handle personal tasks well but lack the rigor for business processes. It requires no setup to start, supports plugging in different underlying LLMs, and includes general compute and browser-use capabilities via a native extension.