Lists executions and opens one to show its event log and steps side by side. Creates a new execution against any registered agent, or cancels a running one.
Agents
Lists registered agents, registers a new one, deletes one, and edits its policy bundle.
Approvals
Lists approvals still pending, soonest timeout first, and grants or denies each one.
Metrics
Kernel counters and latency quantiles: executions, steps, dispatch outcomes, policy decisions, replay hits, and queue depth.
The browser never calls the kernel. Every request goes to the dashboard’s own
/api/v0/* route. That route forwards it to REBUNO_URL and attaches
Authorization: Bearer $REBUNO_API_KEY. The token never reaches the browser.
Give it a token with the access you want the dashboard to have. Anyone who can
reach the dashboard gets that access. Put it behind your own authentication
before exposing it.
Set PROMETHEUS_URL and the metrics page queries Prometheus over a time range
you pick on the page.With it unset the page scrapes the kernel’s /metrics endpoint instead. Those
are one process’s in-memory counters. They are cumulative since that replica
started, so no time range applies. Across replicas you see whichever one
answered. The page labels which source it used. Point a real Prometheus at the
kernel for anything beyond local development.
The image is built from
deploy/Dockerfile
and published to ghcr.io/rebuno/dashboard on tagged releases. It serves a
standalone Next.js build on port 3000:
