본문으로 건너뛰기
이 페이지는 아직 사용자의 언어로 번역되지 않았거나 번역이 기술 검토를 기다리고 있습니다. 아래에는 영어 원문이 표시됩니다. 영어 페이지 열기

Analytics

Supported modes: Hosted Available in: Free Solo Teams (cost analytics: Solo Teams)

The Analytics area groups two pages: Decisions (/analytics) and Costs (/analytics/costs). Both are scoped to the project selected in the nav.

Decisions (/analytics)​

Metrics shown​

  • Total decisions over the selected window.
  • Allow / transform / deny split as stacked counts and percentages.
  • Timeline chart — decisions per bucket (hour or day, depending on the window).
  • Top tools — the tools that fired the most policies.
  • Top policies — the policies that evaluated the most requests.
  • Tamper alerts feed — recent tamper-detection signals from SDKs, with machine ID, reason, and timestamp.

Time range selector​

Top-right of the page. Options: last 1h, 24h, 7d, 30d, custom. Custom lets you pick start and end, subject to your org's audit retention setting (see Governance Home for raising retention).

How to interpret the timeline​

  • A flat horizontal line means decisions are steady.
  • Sudden deny spikes usually correlate with a rule publish or an attacker probe. Cross-reference with the DLP rule history.
  • Silence (no decisions at all) on a previously active project means the SDK is no longer reporting. Check fleet status.

Tamper alerts feed​

Each row shows the machine ID, the reason string reported by the SDK (for example bundle_signature_mismatch, key_rotation_observed), and the bundle hash. Click a row to pin the machine in the fleet view.

Costs (/analytics/costs)​

Cost analytics requires that your SDK is forwarding per-call token counts. Without that, cost numbers default to zero. See Gateway for per-call cost capture.

Metrics shown​

  • Total spend over the selected window.
  • Per-model spend — a table of model, calls, input tokens, output tokens, and cost.
  • Per-day spend — a bar chart of daily totals, overlaid with a rolling average.

Time range selector​

Same as Decisions.

How to interpret​

  • Sudden cost jumps usually mean a new agent, a looping caller, or a model switch to a more expensive model. Drill down with the per-model view.
  • A zero-spend model in the breakdown means the model was called but your SDK did not emit token counts; update the SDK.

Backing APIs​

  • Decisions: GET /api/projects/{projectID}/analytics.
  • Cost summary: GET /api/projects/{projectID}/analytics/costs.
  • Per-model: GET /api/projects/{projectID}/analytics/costs/by-model.
  • Per-day: GET /api/projects/{projectID}/analytics/costs/by-day.