Reports

Reports are ready-made dashboards over the usage, quality, safety and infrastructure data of your services. A summary page answers "what changed?" in one screen, and each report goes into one subject: model spend, agent sessions, guardrail blocks, crawler jobs and so on. Every figure can be set against the previous period, and every report can be downloaded as CSV.

Enterprise module

Reports is an enterprise module (reports). The Reports row of the left menu appears only when the reports API lists something you may read: in the community build, without an active enterprise license, or when none of your permissions cover a report, the row is not shown. See Licensing.

The summary page

Choose Reports in the left menu. The summary page shows, for the period and scope you pick:

• Executive summary. One tile each for Model requests, Spend, Error rate, p95 latency, Agent sessions and Guardrail blocks, with the change against the previous period and a small trend line. A tile is shown only when you may open the report behind it, and a tile with no figure for the period is left out. Spend and requests follow the Cost & Spend and Model Operations reports that their tiles open.
• What changed. With Compare on, the biggest moves against the previous period in plain sentences, for example which model accounts for most of a rise in spend, each with an Open report link.
• Top movers. The largest absolute changes in Spend, Requests and Errors, by model, agent, API token or project.
• Requests and spend. A chart with requests on the left axis and spend (USD) on the right. With Compare on, the previous period is drawn as dashed lines.
• All reports. A card per report you may open, grouped as in the menu.

The controls at the top right are the scope (This project or All projects, see Permissions), the period (last 24 hours, last 7 days, last 30 days, all time or a custom range), the Compare switch (on by default) and a refresh button that bypasses the cache.

Reports and groups

The left menu lists the reports under Reports, below the Summary entry, in four groups. A report is listed only if you may read the service that owns its data.

Group	Reports
Usage & cost	Cost & Spend, Users & Tokens, Model Operations, Batch Jobs
Quality & reliability	Agent Health, Model Health, Knowledge Base Health, MCP Server Health, Tool Health, Agent Sessions, RAG / Knowledge, MCP Servers, Tools
Safety & compliance	Guardrails & Safety, Audit & Security
Infrastructure	Web Search, Browser Sessions, Crawler, Sandboxes, Realtime Voice

There is also an Overview report (cross-service usage at a glance) that feeds the summary page. The five health reports chart the same rollups as Entity health.

Reading a report

Open a report from the menu or from a card. The page has the same scope, period, Compare and refresh controls as the summary, and:

• Filters. Where a report has them (model, agent, user, API token), pick a value to narrow every widget. Clicking a row of a breakdown sets the matching filter; Clear filters removes them.
• Previous period. With Compare on, tiles show the change and charts overlay the previous period. Switch it off to skip the comparison queries.
• Split by dimension. Time charts can be split into their biggest contributors (for example spend per model), with the rest grouped as Other.
• Breakdown tabs. A Breakdown by row of tabs (for example Model, Agent, Route, User, API token, Service, Project) switches the ranked table under the chart.
• Sample records. Tables of individual records list recent entries; where a record has a trace, Open trace opens it in Tracing.

A report with no data in the period says so; try a longer period or clear the filters.

Exporting CSV

The CSV button downloads the report's widget results, for the current scope, period and filters, as one CSV file. The button is shown only to people who hold the export permission. Every download is written to the audit log as a reports.export event with the report, the projects and the period.

Who can open what

Seeing a report takes two things: read access to the service that owns its data (for example Guardrails for Guardrails & Safety), and the per-report actions below. Cost & Spend, Users & Tokens, Model Operations and Batch Jobs are held to Model Hub.

Action	Level	What it allows
report.view	Read	Opening the report.
report.export	Read	Downloading it as CSV.
report.view_all_projects	Admin	The All projects scope and drilling into another project. Without it you see the report for your own project only.
report.view_people	Admin	The parts that name people or API tokens: breakdowns, splits, columns and filters by user or token. Without it those parts are left out, and the Users & Tokens report is not shown at all.

Readers of the reports service hold the two Read actions; the two Admin actions are held by owners and administrators until a policy grants them to someone else. A report is a resource Report::"<key>" inside a group (usage, quality, safety, infrastructure), so one rule can cover a whole group. See Permissions and Policies. A refusal is a 403 that names the action and carries a decision id.

API

Method and path	Purpose
GET /api/reports/summary	The summary page data.
GET /api/reports/dashboards	The reports the caller may open (what the menu lists).
GET /api/reports/dashboards/:key	One report. Query: period (or from and to), scope=tenant, compare=0, fresh=1, filters.
GET /api/reports/dashboards/:key/export	The report as CSV.

These are dashboard (session) routes; they answer 402 without an enterprise license.

Reports by email or chat

Reports are read in the Console; they are not sent on a schedule. For a summary delivered to people, use health digests, and for model-written reports on your busiest agents, scheduled prescription reports.