Agent Bus Dashboard
Agent Bus Dashboard is a dependency-free local switchboard for an existing AgentBus or Agent Coordinator installation. It shows projects, attached agents, live status, current work, usage, tasks, and complete conversation history, with explicit controls for starting, stopping, opening, and messaging agents.
It does not install or start a broker. If no service is running, the dashboard reports that state.
Install
codex plugin marketplace add anon5376/codex-agent-plugins
codex plugin add agent-bus-dashboard@codex-agent-plugins
Restart the ChatGPT desktop app, then ask:
Open my local agent dashboard.
Run directly
From this plugin directory:
python3 scripts/dashboard_server.py
Open http://127.0.0.1:8788, choose Local setup, and enter the directory containing your projects. Every direct child plus nested Git repositories becomes a project entry.
Configuration is stored with mode 0600 at ~/.agent-bus/dashboard.json. CLI flags and AGENT_DASHBOARD_* environment variables override the file. The dashboard reads the live broker at http://127.0.0.1:7717 by default and can also read an optional coordinator database and AgentBus history database.
Requirements
- Python 3.9 or newer
- No third-party Python or frontend dependencies
- An existing AgentBus or Agent Coordinator installation for live data and MCP tools
- macOS only for the Open session in Terminal control; the rest of the dashboard is portable
The server binds to loopback and has no built-in sign-in. Do not bind it to a network interface without placing authentication and transport security in front of it.
Verify
python3 scripts/dashboard_server.py --check
The check renders every configured source and runs role, CSRF, folder-state, configuration, and broker-loopback tests against throwaway state.
Data boundaries
The dashboard reads configured services and databases in place. It does not migrate or rewrite them. Editing a registered agent role writes only that field back to agents.json; pins, hidden projects, and conversation archive/trash state live in the dashboard's own state file.