If you run several Claude Code sessions at once, especially across more than one computer, it’s easy to lose track of which ones are doing what. So I put them all on a Home Assistant dashboard with a bunch of other fun and interesting metrics.
Key features
- Every session live: working, waiting on you, idle or stale, with its model, context use, tokens and a one-line “working on” summary
- One row per computer, and the day so far: sessions, prompts, PRs, time spent waiting on you
- Your Claude plan limits, and whether you’re on course to hit them before they reset (with Clawdmeter, Claude Usage or similar)
- Tokens by model and kind, cache hit rate, and the same by day
- Phone alerts when a session is waiting on you or a limit is running hot: two blueprints, a couple of clicks each
- Personal records, a daily recap, and optionally a status light (an AWTRIX pixel clock or any smart bulb)
How it works
No custom integration: a small Claude Code hook publishes each session’s state over MQTT, and a package of template sensors and a YAML dashboard do the rest. Install the hook as a Claude Code plugin, copy two files into Home Assistant, and restart; the README has the steps. Each session’s last prompt and a short summary go to your own broker, and the README lists everything that’s sent.
I’ve run it daily across three Macs for several weeks, and it’s tested automatically on Linux and against each Home Assistant beta. It’s my first project shared here, so I’d love to hear what you think, what breaks, and what you’d add.
Not made by or affiliated with Anthropic.