Policy as Code

SecuriX allows you to decouple your application logic from your security logic. By using Policy as Code, you can define granular restrictions that are enforced in real-time by the SecuriX Gateway.

Developer Policy Console

The Developer Policy Console (dev.securix.app) is where you manage your security posture without redeploying code.

• Separation of Concerns: Your application layer handles the "How", and SecuriX handles the "Who" and "What".
• Real-time Updates: Change a policy in the console, and it is instantly enforced across all active agent sessions.

Context-Aware Restrictions

Go beyond simple "allow/deny" with context-aware boundaries:

• Draft-only Mode: Allow an AI agent to read emails but only create drafts, never send.
• Domain Blocking: Block data access if the recipient domain matches @bank.com or @competitor.com.
• PII Redaction: Automatically redact Social Security Numbers or credit card details from agent inputs/outputs.

Example: Real-World Guardrails

Here are common policies you can implement using SecuriX PDL (Policy Definition Language):

1. Draft-only Mode (Gmail)

Ensure an AI agent can never send an email autonomously, but can still draft them for user review.

policy "gmail_draft_only" {
  description = "Block actual sending of emails"
  target      = "google.gmail.messages.send"
  effect      = "deny"
  
  llm_error = "I am restricted to creating drafts only. I have created a draft for you to review in Gmail."
}

2. PII Redaction

Automatically redact sensitive information before it reaches the agent's context.

policy "redact_pii" {
  description = "Redact SSNs and Credit Cards"
  target      = "google.*"
  action      = "redact"
  
  patterns = ["ssn", "credit_card"]
}

3. Domain Whitelisting

Restrict the agent to only communicating with specific domains.

policy "whitelist_domains" {
  target = "google.gmail.send"
  
  condition {
    recipient_domain not_in ["yourcompany.com", "trusted-partner.org"]
  }
  
  effect = "deny"
  llm_error = "Security policy restricts me to internal domains only. Please use a company email address."
}

By moving security to the policy layer, you reduce the surface area for bugs and ensure that your AI agents operate within safe, predefined boundaries.