Why it matters
Guardrails can help teams apply rules at the point where a model request or agent action occurs. Their scope depends on the policy and runtime path.
How it works
Checks may inspect selected request or response signals, evaluate tool actions, or pause an action for human review. Teams should understand which paths each control covers.
How Halo approaches it
Halo supports selected runtime classifiers for PII, secrets, prompt injection, moderation, and language checks, with masking or redaction in supported gateway paths.