hapi-monitor: a TUI dashboard for your hub + runners

[heavygee]

Been running my HAPI hub on a Tailscale box and wanted an always-on display of what every connected agent was doing - which Cursor / Claude / Codex sessions are thinking, which are stuck, which machines are still alive. Built it for myself while hacking on this setup, polished it into a standalone tool. Figured someone else might want to drop it into a tmux pane somewhere.

Install (no clone needed):

npx hapi-monitor --watch

Picks up ~/.hapi/settings.json automatically. Override the hub with HAPI_HUB_URL=https://your-hub.tailnet.ts.net. Runtime deps: Node + Python 3 (and a C compiler if you want the crisper native chart - falls back to a pure-Python plot if not).

What's on screen:

• nvtop-style step-plot of working-session count + all-time peak
• Per-session table grouped by status: WORKING / STUCK? / ZOMBIE / OK / INACTIVE
• Model tier per session ($ premium / · eco / ? auto)
• Live CPU + RAM for each agent process
• Hub + runner systemd states alongside app / protocol / git rev
• Hotkey nav: j/k scroll, TAB next alert, i toggle inactive, g/G top/bot, q quit

The anatomy doc walks every line of the screen for anyone who wants to know what each glyph means.

• Repo: https://github.com/heavygee/hapi-monitor
• npm: https://www.npmjs.com/package/hapi-monitor
• MIT licence; PRs welcome.

Thanks @tiann for building HAPI - it's been a fantastic substrate to work on. More than that, it feels like the first interface that gets me closer to what I am doing rather than further away.