Skip to content
Agentometry

Reporting and exports

/me/report, /team/[slug]/report, and /org/report render a trailing-period agent digest with readout highlights, daily spend/session trends, and a 90-day activity calendar. They also include a cost-versus-duration scatter, concurrency signals, a weekday/hour heatmap, and a complete-week digest table. They expose the same aggregate payload as Markdown, CSV, JSON, or a portable report bundle.

The corresponding scoped trend pages are /me/trends, /team/[slug]/trends, and /org/trends. They use the same visibility boundaries as their dashboards and are intended for exploration; reports are the compact planning/readout surface.

Trend and report charts include exact-data tables. The scatter shows session cost against duration (with aggregate scopes bucketed to preserve privacy), concurrency reports peak overlap and the share of sessions that ran in parallel, and the heatmap groups starts by the selected timezone’s weekday and hour. Weekly digest rows use complete Monday–Sunday UTC weeks, excluding the current partial week.

All report and trend surfaces accept repo=owner/name to scope the analysis to a repository. They also accept from=YYYY-MM-DD, to=YYYY-MM-DD, and tz=Area/City for a custom (maximum one-year) window and local-time heatmap grouping. Invalid dates/timezones fall back safely, and the default remains a trailing 30-day UTC window. Analytics are included in every JSON, Markdown, CSV, and bundle export; aggregate team and organization exports remain bucketed and omit member identity. Chart surfaces provide an accessible exact-data table plus a link to matching session records for drill-down.

Each report compares a trailing 7-, 30-, or 90-day period with the immediately preceding period of equal length. A zero prior value is rendered as new, not as an invented percentage. The report payload and renderers live in apps/web/src/lib/reporting.ts, so every representation has identical metrics, periods, and caveats.

Scope and privacy

  • My report contains only the authenticated developer’s interactive sessions.
  • Team report requires team-lead access and uses the same team metadata-sharing filter as the dashboard. It contains no member-level rows.
  • Organization report requires aggregate org access and uses the organization metadata-sharing filter. It contains no member-level rows.

Reports never include transcript text. Their notes state that costs are derived from captured telemetry rather than an invoice, and that only interactive runs appear. CSV values are guarded against spreadsheet formula injection.

The bundle format is a JSON container with report.json, report.md, report.csv, and an escaped standalone report.html, plus a manifest recording the period, scope, schema version, and visibility policy. It deliberately contains no member-level rows or transcripts.

Use ?range=7, ?range=30, or ?range=90 to select a standard window, or use the custom from, to, and tz parameters. Download links preserve the active scope and window; an omitted or invalid range defaults to 30 days.

Individual session pages also include owner-scoped visual analysis for per-turn cost, cumulative spend, cache efficiency, tool activity, model mix, and subagent bursts. These visuals include exact data tables for accessible review and never expose transcript content or owner-only evaluation data to aggregate views.

Scheduled digests

The ingest scheduler includes an opt-in send-report-digest job. It computes a seven-day aggregate (interactive session count and telemetry-derived spend) and delivers it through enabled alert channels (webhook, Slack webhook, or SMTP email), reusing the existing retry and delivery audit log. The default schedule is 08:00 UTC and can be changed from /admin/jobs. The notification links to the organization report and contains no member identifiers or transcript content.