Sierra
Protect Sierra customer-service agents via Custom API integrations pointing at Interven.
Sierra
Sierra's agent platform calls external systems through Custom API integrations. Each integration has a URL and an auth config โ point the URL at Interven, leave auth off (Interven supplies the real credential).
Setup
1. Create the inbound route in Interven
Console โ Inbound Routes โ Add route
- Path prefix:
sierra-orders - Upstream URL:
https://your-order-api.example.com/v1 - Upstream headers:
Authorization: Bearer your-real-order-api-token - Tool name:
custom_proxy(or a specific tool name if it matches a built-in normalizer)
2. Configure Sierra integration
In Sierra's agent admin โ Integrations โ Custom API โ New integration:
- Method:
POST - URL:
https://api.intervensecurity.com/inbound/sierra-orders/lookup - Authentication: None โ Interven adds the real auth on egress
- Headers: keep what your agent flow needs (Sierra context headers pass through)
- Body schema: match what your real upstream expects (Interven only inspects, doesn't reshape)
3. Use in agent flows
Reference the integration in Sierra's flow builder. The agent invokes it during conversations; Interven scans every call before forwarding to the real upstream.
Behavior
| Interven decision | Sierra agent sees |
|---|---|
ALLOW | Real upstream response, transparently |
DENY | HTTP 403 โ flow surfaces an error message to the customer |
SANITIZE | Forwards redacted body to real upstream; customer's PII never reaches it |
REQUIRE_APPROVAL | HTTP 202 + approval_id; flow can wait + retry, or escalate to human |
Why this works
Sierra exposes the Custom API URL as a first-class config field. You can repoint it to Interven any time without redeploying the agent flow. The agent never holds the real upstream credential โ that lives in Interven's encrypted credential vault.
Caveats
- Sierra's response handling expects upstream-style HTTP status codes. The
REQUIRE_APPROVALpath returns 202; check your flow's retry logic handles non-200 responses gracefully (or maps 202 to a "wait + retry" branch). - For complex multi-step actions, define one inbound route per upstream service rather than a single catch-all route โ it gives you per-route policy + tool-name attribution in Activity.