Skip to main content
LLM guardrails are configured in Studio under Governance → LLM Governance (not a separate Studio rail). Workspace Governance → Guardrails is analytics.

​
Prerequisites

  • Pro+ (same Governance entitlement)
  • workspace.governance.create / attach
  • For AWS/Azure/GCP: credentials or env secrets ready
  1. Open Graph Studio → Governance → LLM Governance.
  2. New profile.
  3. Provider → select Phinite → Continue.
  4. Configure → set Profile name.
  5. Under Control library:
    • Optionally filter Compliance framework.
    • Enable controls in Before the model, After the model, Session data.
    • Per control: toggle on, choose Block / Log only / Redact, set threshold if shown.
  6. Save profile.
  7. Attach → Entire flow or One agent → select profile → Attach guardrails.
  8. Confirm In Use (N) and Summary LLM Governance card shows attached.
  9. Save the graph → Build → exercise a turn that should block or log.
Detail: Phinite Guardrails.

​
Full path — cloud provider (AWS / Azure / GCP)

  1. New profile → pick AWS Bedrock, Azure Content Safety, or GCP Model Armor → Continue.
  2. Profile name.
  3. Credentials → Enter credentials or Use Environment secret.
  4. Fill credential + Resource fields (Profiles).
  5. Optional Test connection.
  6. Save profile.
  7. Attach as above.
  8. Save + Build.
Cloud consoles still own topic lists, PII policies, and Model Armor templates — Phinite stores the binding and credentials reference.

​
Attach-only (existing profile)

  1. LLM Governance → Library.
  2. Attach on a row (or header Attach).
  3. Modal Attach guardrails → scope + profiles → confirm.

​
Verify

​
Checklist

  • Profile saved
  • Attached to flow or agent
  • Graph saved and built
  • Tested a violating turn in DEV