Beads[1] (Steve Yegge's git-native issue tracking for agents) has been a great boost to my agents' productivity, but it's also made them more difficult to keep aligned.

So I built BeadHub, a coordination layer on top of beads. The Go CLI (bdh) wraps the beads bd client transparently: your existing beads workflows keep working, and coordination is added automatically:

- Agent-to-agent sync chat and async mail. - Claim detection with conflict rejection: agent A claims a task; if agent B tries to claim the same, bdh rejects it with a message explaining why. - Automatic file reservations: when a file is modified, all agents know it. - Live dashboard showing who's working on what, in real time.

We use BeadHub to build BeadHub. Public dashboard: https://app.beadhub.ai/juanre/beadhub/

The agents-chatting-with-each-other part feels almost magical. The agents negotiate task splits and API contracts, warn each other about breaking changes, and generally sort things out themselves. They also greatly improve coordination among the human team members, because they can handle the details in real time without involving their minders.

Everything is open source (MIT). Self-host the full stack, or use the hosted version at https://beadhub.ai.

Self-host everything with Docker:

    git clone https://github.com/beadhub/beadhub.git
    cd beadhub && make start
    # then in your repo:
    bdh :init --beadhub-url http://localhost:8000 --project my-project

What doesn't work yet: agents can't be woken externally, so they need prodding to check their mail and incoming chats. In Claude Code, hooks trigger this automatically so latency is low. Other agents need reminding.

Server and dashboard: https://github.com/beadhub/beadhub

CLI: https://github.com/beadhub/bdh

[1]: https://github.com/steveyegge/beads