Guardrails / content filter
Checks inputs and outputs for policy, PII and prompt injection.
Problem it solves: LLMs can produce unsafe output or follow injected instructions.
Where it sits: Before and/or after the LLM.
When to use
- User-facing generative features.
Failure modes
- Over-blocking hurts usefulness; under-blocking creates risk.
Metrics to watch
- Block rate
- False-positive rate
- Added latency
Configuration in Labs (illustrative defaults)
- Instances (count) = 2 — Identical copies behind this logical component. Capacity scales linearly in this model.
- Capacity per instance (req/s) = 1000 — Maximum requests one instance completes per second before it saturates.
- Base service time (ms) = 15 — Processing time per request at zero load. Queueing adds to this as utilisation rises.