# Guardrails AI Stack

| Layer | Product | Role |
|-------|---------|------|
| **L4** | Blekline | Pre-model/tool: mask, enforce, audit |
| **L3** | Guardrails AI | Post-model: output structure and validation |

## Dual-deploy flow

```text
Agent → Blekline (ingress) → Model → Guardrails (output) → User
```

Blekline and Guardrails are **complementary**, not overlapping. Blekline does not store duplicate PII for Guardrails validation.

See [Guardrails documentation](https://docs.guardrailsai.com/).

---

**Next steps:** [Eval hub](/docs/integrations/eval-and-safety) · [Trust boundaries](/docs/security/trust-boundaries) · [MCP proxy](/docs/mcp/proxy) · [Open workspace](https://app.blekline.com)