Security gateway

Prompt-injection detection

Run Prompt Guard 2 on your own hardware to flag or block jailbreak and injection attempts.

What it is

Run Llama Prompt Guard 2, or any text classifier served by Hugging Face TEI, on incoming requests to flag jailbreak and injection attempts.

Why you want it

Agents that read untrusted content can be hijacked. A classifier on your own hardware flags attempts without sending prompts to a third party.

How it works

  • Classifier is a model on your own TEI upstream, marked with a classifier role
  • Circuit breaker after 3 failures, with a critical admin alert
  • Classifier calls are metered like other traffic
  • Jailbreak captures retained longer as a tuning corpus

Related features

Security policies and bindings Multiple editions

Bundle checks into policies and bind them to everyone, a group, a model, a provider or a service token.

Details →

Secret detection and redaction Multiple editions

Twenty-seven detectors catch API keys, cloud credentials and passwords, then redact them or block the request.

Details →

PII detection Multiple editions

Card numbers, IBANs, Social Security numbers, emails and phone numbers are recognized in the gateway itself.

Details →

See it on your own network.

The Community edition is free for up to 25 people. The 30-day Business trial unlocks every Business feature.

Download free Start a Business trial