Integrate Operations

Credentials

Credentials resolve from Vault at the integration boundary. They are never held by calling application code, and never appear in the integration definition as literals.

This is the main security argument for the block. An embedded vendor SDK puts the credential in your process, in your config, and often in your logs; a proxied integration keeps it in one place with one rotation path.

Tenancy

An integration is scoped to the tenant that declared it. A shared connector cannot leak across tenants, because resolution is per-tenant rather than per-deployment.

Failure modes

The vendor is down. Calls to that integration fail; others are unaffected. Per-integration failure rate tells you which vendor before their status page does.

A credential expired or was rotated. Resolution fails at the boundary. This is a distinct failure from a vendor outage and worth alerting separately — the fix is different and faster.

You hit the vendor’s rate limit. Call volume per integration is the number to watch, since vendors rarely warn before throttling and the resulting errors look like an outage.

What it does not do

Integrate is a routing layer, not an integration platform. It has no transformation engine, no scheduler and no retry semantics of its own. When you need those:

• Bulk data movement and transformation → Data Pipes
• Scheduled or triggered calls → Jobs
• Multi-step processes with state → Workflows

Reach for Integrate for the connection itself, and for those blocks for what you do around it.

What to watch

Signal	Why it matters
Per-integration failure rate	Which vendor is down, before support tickets arrive
Credential resolution failures	An expired or rotated secret — distinct from an outage
Call volume by integration	Rate-limit exposure on the vendor’s side
Latency by integration	A slow vendor dragging whatever calls it synchronously