Skip to main content
FrostWolf sits between your application and the model. It inspects what goes in and what the model tries to do, and it acts before anything reaches a provider.

Guard

Detects prompt injection and jailbreaks, validates tool calls before they run, and blocks or redacts a request before it reaches the model.

Cloak

Finds personal data in a request, replaces it before the model sees it, and restores it in the response for the identity that owns it.

How you use it

You keep the model client you already use. FrostWolf wraps the call.
Working in Claude Code? The Claude Code plugin brings Guard and Cloak into every coding session, with no code to change.

Where detection happens

Detection runs on the FrostWolf control plane, not in your process. The SDK sends text and receives a verdict, and it holds no rules, patterns, or recognisers of its own. That is deliberate: a rule set that ships to a client is a rule set an attacker can read and step around, and one implementation on the server means there is no stale copy to reconcile. The cost is that inspection needs the network. Every SDK has an explicit policy for when it is not there, and the default is to refuse the request rather than guess. See Failure handling.

What is in these docs

Quickstart

Install, scan a prompt, and wrap a model call in five minutes.

Cloak

Protect personal data, payment cards, and credentials.

API reference

The control plane endpoints, if you are not using an SDK.