Endpoint Backend Profiles

Endpoint backend profiles split production and development traffic into separate backend lanes. This lets operators keep production traffic on the lowest-latency path while sending human, agent, JIT, review, or masking-heavy traffic through a governed development route.

Profiles are useful when the same logical endpoint needs different operational posture for application traffic and human or agent workflows. The endpoint name can stay stable while Eve chooses the correct backend lane from policy and request context.

Profiles

Profile	Use it for
`production`	Live application traffic with predictable latency and pool capacity.
`development`	Developer access, JIT access, SQL review, request analysis, full endpoint masking, per-user checks, and richer audit attribution.

Endpoint Shape

json

{
  "backend_profile": "development",
  "backend_pool_limits": {
    "total_max_connections": 80,
    "production_max_connections": 64,
    "development_max_connections": 32
  }
}

Profiled routing uses explicit production and development branches:

json

{
  "routing": {
    "type": "Profiled",
    "production": {
      "type": "Direct",
      "endpoint": "11111111-1111-1111-1111-111111111111"
    },
    "development": {
      "type": "Direct",
      "endpoint": "22222222-2222-2222-2222-222222222222"
    }
  }
}

Safety Guarantees

A development-profile endpoint without an explicit development route fails closed.
SQL, database, cache, and service/API masking require the development profile; production permits only LLM and agent request-content PII masking.
Eden keys upstream pools by endpoint UUID, keeping production and development backend connections isolated.
Development endpoints should use lower-priority or developer credentials.
Production capacity is allocated first when a shared total_max_connections limit is configured.

Operational Patterns

Pattern	Why it helps
Production application lane	Keeps latency-sensitive application traffic on a direct backend route.
Development analysis lane	Sends human and agent work through lower-privilege credentials and richer audit controls.
Review or JIT lane	Allows temporary access without changing the production application's backend path.
Masking-heavy lane	Applies inspection or redaction controls away from the production pool budget.
Migration validation lane	Separates validation traffic from normal client traffic while the target is being evaluated.

Capacity Planning

Use backend_pool_limits to keep lower-priority lanes from exhausting total endpoint capacity. When a total pool limit and per-profile limits are both configured, reserve production capacity first and size development lanes for bursty operator or agent workflows.

Profile decisions should be visible in telemetry so operators can distinguish production, development, review, and validation traffic when debugging latency or pool pressure.