# Egress governance

Egress governance controls what agents **expose** after model calls and before third-party APIs — complementary to ingress controls on prompts and tool args.

## Shipped today

| Capability | Surface |
|------------|---------|
| **Response mask** | `/api/ingress/v1/chat/completions`, `/api/ingress/v1/messages` — assistant content redacted on return path |
| **SDK outbound mask** | Python/TS `mask()` before API posts |
| **Audit metadata** | Activity log records `entitiesMasked`; no raw content by default |

## Enterprise program (not full DLP)

- Browser paste/download block enforcement (extension telemetry only today)
- Cross-SaaS Workspace DLP
- Endpoint CASB sensors

Do not expect Blekline to replace network DLP. It governs **agent runtime surfaces** at the execution boundary.

## Code pattern (Python agent)

Mask before outbound HTTP from your agent runtime:

```python
from blekline import BleklineClient

client = BleklineClient(token="blw_...")
masked = client.mask(text=payload)
# post masked.masked_text to third-party API
```

## Ingress proxy response path

When using the OpenAI-compatible ingress proxy, Blekline masks user/system messages on the request and assistant content on the response before your agent receives the payload.

---

**Next steps:** [Runtime enforcement](/docs/introduction/interaction-governance) · [NHIM overview](/docs/introduction/nhim) · [Trust boundaries](/docs/security/trust-boundaries)