Most policy tools require you to write code or learn a proprietary DSL. Kovral lets you write policy the way you'd explain it to a colleague.
"Block all prompts containing credit card numbers from reaching external AI models" — that's a complete policy. Kovral's policy compiler translates it into an executable rule and shows you a confidence preview.
Test your policy against a sample of historical traffic before activating it. See exactly what it would have caught and whether it generates false positives on legitimate use.
Every policy change is versioned. Promote from Draft → Staging → Production with separate approval gates. Roll back any policy with one click.
Active policies are enforced at the proxy layer — they apply to every AI interaction in your organisation, regardless of which application, which model, or which user made the call.
Kovral supports four categories of AI policy, covering the full governance lifecycle.
Define which teams can use which AI tools, for which purposes. Example: "The legal team may not use AI to draft external contracts without human review."
Define how sensitive data flows through AI systems. Example: "No customer PII may be included in prompts to models not hosted in the EU."
Define what autonomous agents can and cannot do. Example: "AI agents may not make external API calls to services outside the approved integration list."
Pre-built policy templates for SOC 2, HIPAA, GDPR, EU AI Act, and NIST AI RMF. Import a template and customise for your organisation in minutes.
