Get the guardrail coverage report

GET/v1/trust/guardrail-coverage

Get the guardrail coverage report.

How much of your organization's traffic its guardrails checked in the window and what they caught: totals (requests, requests checked, blocked, redacted, monitored, streamed requests, MCP Gateway tool calls), each guardrail that ran in the window or is enabled now with its checks and outcomes, the personal data types an enabled check redacts or blocks, and the tool-result scan. statements restates the main figures as plain sentences.

The window defaults to the 90 days up to now and can be at most 366 days long. Without workspace_id it requires an organization owner or admin; with it, any member of that workspace.

Errors: 400 when start_time is not before end_time or the window is longer than 366 days; 402 when the organization's plan does not include the AI Trust Pack; 403 without the access above.

Authorization

AuthorizationBearer va_
A platform API key (va_...), created under Settings > API keys. Scoped to an organization or one workspace, with an admin or member role. Used by the management API.

Query parameters

organizationstringrequired
the organization slug
workspace_idstring<uuid>
limit the report to this workspace (UUID); omit for the whole organization
start_timestring<date-time>
start of the window, ISO 8601 date-time; default 90 days before end_time
end_timestring<date-time>
end of the window, ISO 8601 date-time; default now

Responses

200application/json

organizationstring
The organization slug the report is for.
workspace_idstring<uuid>
The workspace the report is limited to (UUID); absent for an organization-wide report.
start_timestring<date-time>
Start of the report window (ISO 8601 date-time).
end_timestring<date-time>
End of the report window (ISO 8601 date-time).
generated_atstring<date-time>
When the report was generated (ISO 8601 date-time).
summaryobject
Guardrail coverage of the organization's traffic in the window.
Child attributes
- requestsinteger<int64>
  Requests through the gateway in the window.
- requests_checkedinteger<int64>
  Requests at least one guardrail ran on.
- checked_percentnumber<double>
  requests_checked / requests, 0 to 100, one decimal; null with no requests.
- streamed_requestsinteger<int64>
  Streamed requests in the window.
- streamed_requests_checkedinteger<int64>
  Streamed requests at least one guardrail ran on.
- blockedinteger<int64>
  Requests whose combined guardrail outcome was BLOCK.
- redactedinteger<int64>
  Requests whose combined guardrail outcome was REDACT.
- monitoredinteger<int64>
  Requests whose combined guardrail outcome was MONITOR (flagged, passed through unchanged).
- tool_results_scannedinteger<int64>
  Tool results in requests, scanned by the always-on tool-result scan.
- tool_results_flagged_requestsinteger<int64>
  Requests where the tool-result scan flagged something (MONITOR, REDACT or BLOCK).
- mcp_callsinteger<int64>
  MCP Gateway tool calls in the window.
- mcp_calls_checkedinteger<int64>
  MCP Gateway tool calls a guardrail checked.
- mcp_calls_blockedinteger<int64>
  MCP Gateway tool calls a guardrail blocked.
guardrailsobject[]
Guardrails that ran in the window or are enabled now, most used first.
Child attributes
- slugstring
  The guardrail's slug (gr-...).
- namestring
  The guardrail's name; absent when it ran in the window but has since been deleted.
- enabledboolean
  Whether the guardrail is enabled now; absent when it has been deleted.
- workspace_idstring<uuid>
  The workspace the guardrail belongs to (UUID); absent for an organization-wide guardrail or a deleted one.
- checksobject[]
  Its checks, as configured now; absent when it has been deleted.
  Child attributes
  - typestring
    The validator id (for example PII_DETECTOR, prompt_injection), as GET /v1/guardrail-validators lists it.
  - namestring
    The validator's display name; absent for a validator id the gateway no longer knows.
  - kindstring
    DETERMINISTIC (a pattern or rule check run by the gateway) or LLM_JUDGE (an AI check, run by TypeSafe AI).
  - targetstring
    What the check runs on: INPUT (the request), OUTPUT (the response) or BOTH.
  - actionstring
    What the check does on a finding: BLOCK, REDACT or MONITOR.
  - pii_typesstring[]
    PII checks only: the personal data types it looks for (EMAIL, PHONE, SSN, CREDIT_CARD, IP_ADDRESS).
- virtual_keysinteger<int32>
  Virtual keys that apply it to every request.
- mcp_bindingsinteger<int32>
  MCP connectors and endpoints that apply it to tool calls.
- requestsinteger<int64>
  Requests in the window it ran on.
- passedinteger<int64>
  Of those requests, how many passed every guardrail applied to them.
- blockedinteger<int64>
  Of those requests, how many were blocked (by this or another guardrail on the same request).
- redactedinteger<int64>
  Of those requests, how many were redacted (by this or another guardrail on the same request).
- monitoredinteger<int64>
  Of those requests, how many were flagged without being changed.
pii_types_coveredstring[]
Personal data types (EMAIL, PHONE, SSN, CREDIT_CARD, IP_ADDRESS) that an enabled PII check with the REDACT or BLOCK action looks for; empty when none does.
tool_scanobject
The scan of tool results sent back to a model through the gateway, for indirect prompt injection.
Child attributes
- enabledboolean
  Whether the scan runs on this deployment.
- default_actionstring
  What a finding does on a virtual key without its own setting: MONITOR (recorded, the request is unchanged), REDACT (the matched text is replaced) or BLOCK (the request is refused).
- keys_by_actionmap<string, integer<int32>>
  Virtual keys with their own scan action, counted per action (MONITOR, REDACT, BLOCK).
- validatorsstring[]
  The validator ids the scan runs.
statementsstring[]
The main figures as plain sentences, ready for a questionnaire or a trust page.
notesstring[]
How to read the report: what is and is not counted.

VectorAxis is built by GK Labs, a member of NVIDIA Inception.