Security gateway

Decision models as security guards

One decision model can back prompt-injection and content-safety checks, as an alternative to Prompt Guard and Llama Guard.

What it is

A model with the Decision model role can back prompt-injection checks, content-safety checks, or both. It answers each check with a probability, and Janus suggests a threshold of 0.3 for injection when none is set.

Why you want it

Running Prompt Guard on TEI and Llama Guard on vLLM means two separate servers. A single decision model, hosted or self-hosted, can cover both checks and Usage insights.

How it works

  • Checks read the new turn by default; a per-check option judges the whole conversation
  • The last 12,000 characters are judged by default; a per-check option judges the whole text in windows
  • Content safety on responses is observed and logged, never blocked
  • An observe-mode check never blocks when its classifier is down; fail closed applies to blocking checks only
  • Classifier checks observe or block; they don't redact

Related features

Prompt-injection detection Multiple editions

Run Prompt Guard 2 or a decision model to flag or block jailbreak and injection attempts.

Details →

Content safety checks Multiple editions

A decision model checks conversations against the categories you choose, the built-in Safety set or your own, each with its own minimum confidence.

Details →

Parallel classifier checks

Model-backed checks run at the same time, so a request waits for its slowest guard rather than all of them in turn.

Details →

Decision models

Classifier models that answer typed questions with probabilities, used inside Janus and served to people and apps.

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