AI Guardrails You Control: Secrets, Personal Data & Safety Scans on the Request Path

Guardrails shouldn't be a one-size-fits-all switch

Most gateway guardrails are all-or-nothing and tuned by the vendor. Teams get surprise blocks on legitimate traffic, or miss the coverage their region requires, and heavy safety scans add cost to every call.

• Vendor-tuned defaults block traffic you never asked to block
• Personal-data formats differ by country, so one global detector rarely fits
• Heavy semantic scans add cost and latency on every call
• Guardrail changes need an owner, a version history, and an audit trail

Your profiles, merged by strictest rule

A guardrail profile protects the data in every request before the model sees it. Switch on Cloptima templates, add your own rules, and redact or block each match. Content safety adds checks for prompt injection and harmful content.

• Cloptima templates for credentials, regional personal data, finance, and health
• Your own rules for business terms, each set to redact or block
• Redacted values return in the response, so your app keeps working
• An org baseline that teams can tighten but never loosen

Watch first, then redact, then block

Attach a profile to a policy in Monitor only and see what it would redact or block, with no change to your traffic. Then enforce the policy, move the cases that must never reach a model to block, and make the profile your organization baseline.

Cost and outages stay in your hands

Local rules run inside the gateway with no extra network call, so a request they block never pays for a scan. If a provider scan cannot run, each profile chooses: keep traffic flowing on your own rules, or block. Unchecked requests show as Not checked, and every profile change is versioned and audited.