Back to Product

Unified Policy Engine

One policy language for all your AI agents. Consistent enforcement across every provider and framework—OpenAI, Anthropic, LangChain, and more.

Universal Policy

Write once, enforce everywhere

Define policies once in YAML and apply them to any AI provider or framework. No more duplicating security logic across integrations.

  • Match by agent, resource, or custom attributes
  • Chain multiple actions together
  • Built-in audit logging and notifications
  • Test in dry-run mode before deployment
policies/pii-protection.yaml
# PII Protection Policy
name: pii-protection
version: 1.0.0

rules:
  - name: mask-customer-pii
    match:
      agents: ["support-bot", "sales-assistant"]
      resources: ["customer_profiles", "user_data"]

    actions:
      - type: mask
        fields:
          - ssn
          - credit_card
          - date_of_birth
        pattern: "***-**-{last4}"

    audit:
      level: detailed
      retention: 90d

  - name: block-secrets-access
    match:
      agents: ["*"]
      resources: ["secrets_vault", "api_keys"]

    actions:
      - type: block
        message: "Access to secrets requires elevated permissions"

    notify:
      - channel: slack
        severity: high
Actions

Six ways to handle data access

Policies define what happens when an agent tries to access data. Choose from built-in actions or create your own.

allow

Permit the data access request

block

Deny access and return an error

mask

Allow access with sensitive fields masked

redact

Remove sensitive content entirely

transform

Apply custom transformations to data

log

Allow and create detailed audit entry

Capabilities

One engine, every AI

Provider Agnostic

One policy language for OpenAI, Anthropic, Google, Azure, AWS, and any other AI provider. No vendor lock-in.

Policy Inheritance

Create base policies and extend them for specific use cases. Build composable rule sets that scale with your organization.

Framework Integration

Works with LangChain, LlamaIndex, CrewAI, AutoGPT, and custom frameworks. One policy, every agent.

Hot Reload

Update policies without restarting your agents. Changes propagate instantly across all connected clients.

Edge Enforcement

Policies execute locally with sub-millisecond latency. No round-trips to a central server.

Dry Run Mode

Test new policies in shadow mode before enforcement. See what would be blocked without affecting production traffic.

One policy for all your AI

Stop duplicating security logic. Define policies once and enforce them everywhere.

Start free trial