Compass Docs
Observability

Observability overview

See every run across your project — health metrics, per-agent and per-workflow rollups, and what's running now.

Observability is where you watch and understand everything your AI does at runtime. Every execution — deployed or not, triggered any way — produces a run, and a run is a trace: a complete, step-by-step record of what happened. Observability gathers all of them in one place, under Console → Observability.

Project metrics

The top of the page summarizes the whole project's recent health:

  • Runs (30d) — how many runs in the last 30 days.
  • Success rate — the share that completed successfully.
  • p95 latency — how long runs take at the 95th percentile, with the average alongside.
  • Active now — how many runs are executing right this moment.

By agent and workflow

Below the metrics, a table rolls up activity per agent and per workflow that has run:

  • total runs,
  • success rate (highlighted when it dips below 90%),
  • average duration,
  • total tokens used,
  • and last run time.

The table is sortable, so you can quickly find the busiest, slowest, or most error-prone piece of work. Select any row to drill in.

Drilling into one agent or workflow

Opening a row focuses on that single agent or workflow: a header with its rollup metrics (total runs, success rate, average run time, total tokens, last run) and how many are running right now, followed by a runs list filtered to just its executions.

Watching live

Runs in progress show up immediately — the Active now count and a live indicator mark anything currently executing, and you can open a running run to watch it stream in real time.

Next

On this page