Vendor-agnostic orchestration OS for autonomous AI agents. Coordinate Claude, GPT, Gemini, and Grok using Markdown files as shared memory. Human-in-the-loop, Git-native, runs on GitHub Actions free tier.

Agent Hive

Agent Hive is a repo-native control plane for autonomous work. In v2.5, that means you can keep existing worker flows like Codex, Claude Code, and local/manual execution while operating them through a browser-first command center and, if you want, a thin desktop beta shell.

Keep your agent. Add a control plane.

If you are already using coding agents and feeling the limits of isolated sessions, Agent Hive is the layer that makes them feel like part of one real system:

• one command center above many runs and many projects
• one governed loop instead of “the agent said it was done”
• one inspectable substrate instead of hidden agent state
• one place to steer Pi, OpenClaw, Hermes, Codex, Claude Code, local execution, and manual handoffs

The center of gravity in this repository is the Hive v2 substrate. v2.3 established the truthful operator surface; v2.4 extends it into native Pi, OpenClaw, and Hermes companion flows; v2.5 turns that foundation into the first command-center release.

Why It Feels Different

• The console is not a toy dashboard. The browser command center is a real operator surface with active runs, inbox items, notifications, campaign reasoning, retrieval traces, and run detail in one place.
• Hive does not replace your worker harness. It sits above it, so you can keep using Pi, OpenClaw, Hermes, Codex, Claude Code, local execution, or manual loops.
• Native companion paths are real now: Pi can open or attach, OpenClaw can