Logging and monitoring
Can logs go to a SIEM?
Yes, two ways. Export audit logs as a file or query them through the audit log API, or stream them to an external system with audit log Destinations, on Enterprise. See Audit logs.Can CI/CD authenticate without a static key?
Yes, through workload federation, on the Enterprise plan. A workflow presents an OIDC token from GitHub Actions or another issuer and exchanges it for a short-lived key. See Set up GitHub Actions step by step.Are single sign-on and SCIM provisioning supported?
Yes, on the Enterprise plan. Configure SSO covers SAML and OIDC connections, SSO enforcement, and SCIM provisioning for users and groups.Certifications and data handling
Which certifications and reports are available?
SOC 2 Type II, ISO 27001 status, penetration test reports, a HIPAA BAA, a DPA, and the sub-processor list. Request them from the Trust Center; Trust Center and compliance lists what each one covers.How long is data retained?
14 days on Free, 30 days on Pay as you go, and up to 365 days on Enterprise. Audit logs follow their own window, up to 3,650 days. See Data retention.Can inference be restricted to providers that retain nothing?
Not through the filter alone. The Zero Data Retention filter narrows the model list in the console to providers that guarantee no retention after the call, and it is off by default (Sovereign AI). Restricting inference itself means choosing only those models in the request path.Can PII masking be enforced for the whole workspace?
Yes. Enable the PII Redaction plugin under Settings > Plugins: it then applies to every request on the routes it supports, and a request cannot switch it off. Trace scrubbing can also be set for the whole workspace, a routing rule, or an MCP gateway. In AI Studio, masking is set per variant, on each input flagged as PII (Data compliance).What is the difference between masking and redaction?
Masking, throughsecurity.mask, hides content from stored traces only and the provider still receives the full payload. PII Redaction keeps the values away from the provider. See Request and response security.
Access control
Can masking or access rules differ by identity or agent?
Not as a setting on the identity or the agent. Masking is set per request, per workspace, per routing rule, per MCP gateway, or per variant; a routing rule can match on an identity, so the effect can differ per identity, but no identity or agent carries a masking setting of its own. Access is scoped by API key permissions and project, not by identity.Which access controls exist?
- API keys: permissions of All, Read only, or Restricted, an optional expiration, and a project scope (API keys).
- Project isolation: a key reaches its project, or every project when created without one.
- IP allowlist: restrict requests to the Orq.ai backend APIs to known ranges; the AI Studio frontend is not covered (Workspace security).
- Guardrails: rule-based blocking of requests that fail a check; System Guardrails fail closed when the check itself errors (Guardrail rules).
- Identity provider controls: SSO enforcement, SCIM deprovisioning, and short-lived workload federation tokens.