Dashboard

Home#

One question: is everything OK?

Home is the page you leave open. It polls on a short interval and is designed so that nothing needing your attention is more than one glance away.

What is on it#

Live status. How many sessions are active right now, on which runtimes, and what they are doing. "Active" here means recent event activity, not a heartbeat — a heartbeat tells you a process is alive, which is a different and much less interesting fact than whether the agent is working.

Spend. Today, this week, this month, with the trend. The number is derived from the same rollups the Cost page uses, so the two never disagree.

Session activity timeline. Thirty days of session starts, so a quiet stretch or an unexpected burst is visible without opening anything.

Health. Disk, memory, uptime, GPU where present, plus whether the daemon is keeping up with ingest.

Channels. Which input channels are live and whether messages are being delivered.

Crons. Scheduled jobs, their last run, and whether anything failed.

The badges#

The navigation carries counts, and they are the fastest signal on the page:

A number on Guard is the one worth reacting to immediately — it means an agent is, right now, doing something a detector flagged, and the sessions behind it are ranked by spend at risk rather than severity.

Where the data comes from#

curl -s localhost:8900/api/overview   | jq
curl -s localhost:8900/api/timeline   | jq '.days[-7:]'
curl -s localhost:8900/api/system-health | jq
curl -s localhost:8900/api/channels   | jq

/api/overview is the one the page polls. It is composed from the daemon's rollups, so it is cheap even with a large store — it is not scanning your event table on every poll.

If the page is empty#

In order:

1. clawmetry status — is the daemon running and ingesting?
2. clawmetry runtimes — was anything detected?
3. Troubleshooting for the rest of the chain.