# Eval and Safety Partners

Blekline operates **before** model and tool execution. These partners operate **after** or **alongside** for quality and output validation.

## Positioning

| Product | Layer | Role |
|---------|-------|------|
| **Blekline** | L4 | Mask, enforce, audit at MCP ingress |
| [LangSmith](/docs/integrations/langsmith-stack) | L3 | Eval, traces, datasets — consumes audit metadata |
| [Guardrails AI](/docs/integrations/guardrails-stack) | L3/L4 | Post-model output validation — optional second layer |

## Data flow

```text
Agent → Blekline (allow/block) → Model/Tools → Guardrails (optional output check) → User
                ↓
           LangSmith (eval traces, metadata only)
```

LangSmith is **not** a SIEM. For compliance export see [Enterprise telemetry](/docs/enterprise/telemetry).

---

**Next steps:** [Trust boundaries](/docs/security/trust-boundaries) · [Telemetry spec](/docs/reference/telemetry-spec) · [AI Enablement Stack](/docs/introduction/ai-enablement-stack) · [Open workspace](https://app.blekline.com)