Policy Enforcement

AI Agent Guardrails & Runtime Policies

Enforce pre-dispatch prompt injection detection, client-side PII scrubbing, spend caps, and runtime execution policies for production AI applications.

Early Access · 90 Days Free · No credit card requiredBuilt by the Observyze engineering team for production AI systems.
Architecture Separation

Privacy Controls vs. Runtime Policy Enforcement

Observyze clearly separates client-side data minimization from active gateway policy enforcement:

Layer 01

Privacy & Data Minimization

  • Client-side PII regex scrubbing for emails, keys, and phone numbers.
  • Configurable captureContent: false to omit prompt text completely.
  • Redaction occurs in-memory before payloads leave your application.
Layer 02

Runtime Policy Enforcement

  • Pre-dispatch prompt injection heuristics and jailbreak detection.
  • Hard execution budgets (max dollars and max steps per session).
  • Distributed circuit breakers to halt runaway retry cascades.
Code Configuration

Enforce Telemetry & Policy Rules

Configure privacy redaction flags in your SDK client options or enforce runtime policies through project guardrail settings.

Deterministic client-side PII scrubbing
Configurable max spend threshold per trace
Immediate edge rejection on policy breach
guardrails-config.ts
import { ObservyzeClient } from "@observyze/sdk";

const observyze = new ObservyzeClient({
  apiKey: process.env.OBSERVYZE_API_KEY!,
  projectId: process.env.OBSERVYZE_PROJECT_ID!,
  // Privacy & Data Minimization
  privacy: {
    maskPii: true, // Redact emails, SSNs, credit cards
    captureContent: true, // Set false to omit raw text
  },
  // Runtime Execution Budget
  executionBudget: {
    maxCostUsd: 0.50, // Stop trace if spend exceeds $0.50
    maxSteps: 8,      // Stop loop if turns exceed 8
  },
});
Developer FAQ

Frequently Asked Questions

Technical details, integration patterns, and operational controls.

Client-side PII redaction is a privacy and data-minimization technique that scrubs sensitive values (like emails, phone numbers, and API tokens) from payloads in your application before telemetry is dispatched. Runtime guardrails are active policy checks that evaluate incoming prompts or active sessions for malicious injection patterns, spend limits, or execution thresholds.