Last updated: 2026-09-06
orca stats reports a snapshot of activity for the tenant over a look-back window: totals, a per-agent breakdown, and hotspots. orca usage reports token, cost, and run usage over time, plus tool-call and sandbox meters; spend lives only there, because the stats totals carry no cost field. See Billing for how usage is priced.
Both commands accept the global flags and take no positional arguments. A --window <w> value is a positive integer followed by m (minutes), h (hours), d (days), or w (weeks); the help text advertises 1h, 24h, 7d, and 30d. An invalid window fails with exit code 2 before any request is made. The default is 24h for every orca stats view and 7d for orca usage.
orca stats
--json it also fetches the per-agent and hotspot views; if either of those requests fails the summary still prints. When stdout is a pipe only the summary is fetched.
window, agents, sessions, liveSessions, runs, runningRuns, failedRuns, cancelledRuns, tokens, errorRate (a fraction with four decimals), p95RunDurationMs, runners, healthyRunners. --json prints { "window", "summary", "agents", "hotspots" }: summary is the server’s summary record (window, since, until, totals, sessionStatusCounts, runStatusCounts, runnerPool), agents is the per-agent array (empty when unavailable), and hotspots is the hotspots record (null when unavailable). The terminal view shows a Stats panel (agents, live and total sessions, runs, running, failed, and cancelled counts when non-zero, tokens, error rate, p95 run duration, healthy runners out of total and active sessions when the runner pool reports), then the Agents table and the Hotspots panel described below.
Wherever a token count appears, it is the sum of input, output, cache-read, and cache-create tokens.
orca stats agents
name, runtime (- when empty), runs, failedRuns, tokens, lastActivityAt (- when absent). --json prints the server response (window, since, until, total, limit, offset, sort, agents); each agent record carries name, runtime, sessions, liveSessions, runningSessions, erroredSessions, runs, runningRuns, failedRuns, cancelledRuns, tokens, lastActivityAt, and p95RunDurationMs. The terminal view is a table with agent, runtime, runs, failed, tokens, and last active; with no activity in the window it prints No agent activity in this window. on stderr.
orca stats hotspots
tokenConsumers (value in tokens), failingAgents (value in failures), busyRunners (value in sessions), and longSessions (value in minutes).
kind, name, value. --json prints the server response (window, since, until, and the four arrays; each entry has kind, name, value, and optionally runtime, tokens, lastEventAt). The terminal view is a Hotspots panel with a section per non-empty list; when every list is empty it prints No hotspots in this window. on stderr.
orca usage
--meter picks which series is plotted and listed. Tool-call and sandbox meters are aggregate-only, so they appear as totals and cannot be chosen with --meter (exit code 2).
YYYY-MM-DD HH:MM), the meter name, the meter value (total tokens, cost in dollars, or run count), and the bucket’s cost in dollars to two decimals (empty when zero). --json prints { "window", "meter", "timeseries", "meters" }: timeseries has window, bucket, since, until, and buckets, each with start, end, runs, failedRuns, runningRuns, tokens, and costCents; meters has window, totals (toolCalls, sandboxSeconds), and byProfile, or is null when unavailable. The terminal view is a Usage panel with tokens, runs, failed runs and spend when non-zero, tool calls and sandbox seconds (or a tool-call / sandbox unavailable note), and a line chart of the chosen meter per bucket.
Verified against orca-cli
cli-v0.5.0.