Total visibility and control over AI agents and users. AI Control closes the governance gap — before your next audit, or your next incident.
apiVersion: aicontrol.nirmata.io/v1
kind: AIPolicy
metadata:
name: prod-agent-policy
spec:
rules:
# Require verified identity on all calls
- name: require-oidc-identity
validate:
identity:
require: oidc-jwt
# Block PII and secrets from leaving
- name: block-pii-egress
validate:
dlp:
block: [PII, SECRETS, CREDIT_CARD]
# Enforce per-session spend limit
- name: session-budget
validate:
budget:
perSession: $5.00
action: terminateAs AI agents proliferate, three critical risks emerge that no existing tool addresses.
Users, agents, and tool calls lack proof of authorization. No cryptographic identity at the boundary. API keys are not identities.
PII data and secrets can flow to external tools unchecked. No session or tool call audit records. Okta verified the human; everything after is invisible.
High costs, destructive tool use, executed without approvals. No session budgets. No circuit breakers. You find out when the bill arrives.
The same battle-tested approach behind Kubernetes admission controls — now governing every AI agent call.
Everything you need to govern AI agents in production — from identity to audit to cost control.
Validates OIDC JWTs for every call — not just at login. API keys are not identities. Cryptographic proof at the AI boundary.
Every session and tool call recorded in a tamper-proof, exportable log. Know exactly who did what — and who approved it.
Per-identity spend tracking with session budgets and circuit breakers. Stop runaway 5K sessions before they hit your invoice.
Scans for PII, secrets, and jailbreak attempts using Presidio and LlamaGuard before data reaches any external tool.
Define approval workflows for high-risk operations. No destructive or sensitive action executes without explicit authorization.
Blast radius scoring and egress controls limit what data and actions can leave your environment. Know your exposure.
Kyverno CEL policies enforce allow/deny rules and rewrite model arguments. Auditable, versionable, GitOps-friendly.
AI Control enforces identity, policy, and audit wherever your AI stack already runs. No rip-and-replace required.
Deploy where it makes sense for your architecture — shared proxy, sidecar, or service mesh authorization provider.
A shared governance proxy routes all agent traffic through centralized policy enforcement. Simplest path to full coverage.
Kyverno sidecar proxy co-deployed with each agent pod in regulated Kubernetes clusters. Isolated policy per workload.
Integrates with Envoy/Istio as an external authorization provider via gRPC/HTTP. Fits your existing service mesh.
Join engineering teams securing their AI agents with AI Control.
Drop-in governance, no infrastructure changes required.
Built by the team behind Kyverno — the Kubernetes policy engine trusted by thousands of production clusters.