Skip to Content
EnterpriseCompliance Pack

Compliance Pack

Enterprise (compliance_pack entitlement). The open-source platform ships the guardrail plugins — pii-filter and security-scanner — but each agent opts in individually. The compliance pack turns them into org policy: a central declaration of which guardrails are mandatory and how strict they are, plus a posture report you can hand to an auditor.

This is the operational half of the story told on the Compliance page: Art. 9 risk management wants demonstrable, organization-wide controls, not per-agent good intentions.

The org policy

curl -X PUT /compliance/policy \ -H "Content-Type: application/json" \ -d '{ "enforcedPlugins": ["pii-filter", "security-scanner"], "securityThreshold": "medium", "blockOnViolation": true, "piiCategories": [] }'
FieldValuesMeaning
enforcedPluginspii-filter, security-scannerwhich built-in guardrails are mandatory org-wide
securityThresholdlow | medium | high | criticalseverity at which the security-scanner acts
blockOnViolationbooleanblock the offending call vs. log-and-continue
piiCategoriesstring[]PII categories to filter; empty = all

Secure by default: until an admin configures a policy, the effective policy enforces both plugins with securityThreshold: medium and blockOnViolation: true. Configuring can only relax from there, deliberately.

GET /compliance/policy returns the effective policy (including whether it’s the default or configured). Admin/owner only.

The posture report

curl "/compliance/report?from=2026-06-01&to=2026-07-01" # defaults: last 30 days

The report is designed to be attached to a conformity file as-is:

  • Effective policy for the window.
  • Enforced controls — per plugin: enforced or not, and the effective settings (threshold, blocking, categories).
  • Activity — audit-log-derived counts: total events, breakdown by action, how many were scannable executions (tool/agent runs), and credential-access events.
  • Posture score (0–100) — a blunt summary: 40 points per enforced guardrail plugin, 20 for blocking mode. Full marks means both guardrails enforced and violations blocked, not merely logged.