Overview

#/overview is the console's control center: what needs you, what is running, whether the system is healthy, and how runs have gone lately. It sits above the Inbox in the rail. The Inbox stays the home page and the place you answer things; Overview gives you the whole picture first.

Like Runs and the Inbox, it covers every project on the machine, because the run store is global. See Data directory.

What it shows

Section	What it shows	Source
Headline	"N things need you", plus runs in flight and runs that failed in the last hour	Run catalog
Needs you	Pipeline gates and agent approvals, longest waiting first, then failed runs. Accept / Reject answer an agent approval in place; Retry retries a failed run's stage. A gate opens its run, because answering needs the evidence.	GET /api/runs, GET /api/agents/approvals
System health	Agent slots (running, held by a gate, free), providers, connections that need attention, triggers armed, run store size and free disk	GET /api/health, GET /api/connections, GET /api/triggers
KPIs	Runs, success rate, longest waiting on you, pipeline spend. Runs, success rate and spend compare with the previous window of the same length	GET /api/overview
Live now	Runs in flight with their stage track and elapsed time, then running or waiting agent sessions with their 7-day turn stats	Run catalog, GET /api/agents/sessions
Activity	Recent finished runs and trigger fires	Run catalog, GET /api/triggers
Runs by outcome	Succeeded, failed and cancelled runs per bucket	GET /api/overview
Pipelines	Per pipeline in the window: runs, success rate, median time of finished runs, cost, last run. Pipelines under 70% success (two or more finished runs) are tagged flaky	GET /api/overview
Triggers, Connections	Next fire or last fire per trigger; channels and email accounts with their last check	GET /api/triggers, GET /api/connections

The 24h / 7d / 30d control changes the KPIs, chart and pipeline table. It is remembered for the browser session. The page refreshes about every 15 seconds while the tab is visible; the run catalog refreshes more often.

A section whose API fails shows what it can and the rest stays live.

Definitions

• Success rate = succeeded / (succeeded + failed). Cancelled runs count as neither. It is "—" until a run has finished.
• Window: runs counted by the time they were created. Buckets are rolling segments ending now: 24h is 8 buckets of 3 hours, 7d is 7 days, 30d is 10 buckets of 3 days.
• Spend is the sum of each stage attempt's cost_usd. A run whose provider reports no cost adds nothing.
• Longest waiting on you is the oldest gate or agent approval in Needs you.

GET /api/overview

Query: range is 24h, 7d (default) or 30d; anything else is 400 with code: "invalid_range".

{
  "range": "7d",
  "generated_at": "2026-10-17T12:00:00.000Z",
  "since": "2026-10-10T12:00:00.000Z",
  "totals":   { "runs": 48, "succeeded": 39, "failed": 4, "cancelled": 1, "in_flight": 4, "finished": 43, "success_rate": 0.907, "cost_usd": 18.4 },
  "previous": { "runs": 40, "succeeded": 31, "failed": 6, "cancelled": 0, "in_flight": 3, "finished": 37, "success_rate": 0.838, "cost_usd": 16.3 },
  "buckets":  [{ "start": "2026-10-10T12:00:00.000Z", "succeeded": 3, "failed": 0, "cancelled": 0 }],
  "pipelines": [{ "pipeline_id": "feature-ship", "runs": 19, "succeeded": 18, "failed": 1, "cancelled": 0, "success_rate": 0.947, "median_duration_ms": 860000, "cost_usd": 6.8, "last_run_at": "2026-10-17T11:54:00.000Z" }]
}

It reads one light row per run (id, pipeline, status, timestamps, summed cost) and aggregates in memory; it does not load stages. It needs the same access as GET /api/runs.

Not on the page

Median gate wait time, a global event feed and mailbox sign-in health are not shown: the run store does not keep them in a form the page can read cheaply.