# FrostWolf

> Runtime security for AI agents: prompt injection defense, tool-call validation, and personal data protection.

- [Introduction](https://docs.frostwolf.app/introduction.md): Runtime security for AI agents. Stop prompt injection, validate tool calls, and keep personal data out of your models.
- [Quickstart](https://docs.frostwolf.app/quickstart.md): Install the SDK, scan a prompt, and protect a model call.
- [How it works](https://docs.frostwolf.app/concepts/how-it-works.md): The control plane, the two detection stages, what leaves your process, and how failures are handled.
- [Scan a payload](https://docs.frostwolf.app/guard/scan.md): Inspect text or a chat payload and read the verdict.
- [Wrap and sanitise](https://docs.frostwolf.app/guard/wrap-and-sanitise.md): Block a model call, or redact the matched spans and forward it.
- [Decorate a client](https://docs.frostwolf.app/guard/decorate.md): Wrap the completions function you already call, including streams and Anthropic.
- [Tool calls](https://docs.frostwolf.app/guard/tool-calls.md): Validate what the model tries to do, before it does it.
- [Detection](https://docs.frostwolf.app/guard/detection.md): The attack categories, the two stages, the semantic pass, and how to read a verdict.
- [Failure handling](https://docs.frostwolf.app/guard/failure-handling.md): What the guard does when it cannot get a verdict.
- [Cloak](https://docs.frostwolf.app/cloak/overview.md): Replace personal data before a model sees it, and restore it in the response, only for the identity that owns it.
- [Protect and restore](https://docs.frostwolf.app/cloak/protect-and-restore.md): Protect a request, call the model, and restore the response.
- [Strategies](https://docs.frostwolf.app/cloak/strategies.md): Tokenize, hash, mask, or redact, chosen per entity type or per category.
- [Identity and the vault](https://docs.frostwolf.app/cloak/identity-and-vault.md): How a token is bound to an identity, where values are kept, and how to erase them.
- [With the attack check](https://docs.frostwolf.app/cloak/with-attack-check.md): Attacks and personal data in one call, failing closed, and the semantic pass.
- [Entity types](https://docs.frostwolf.app/cloak/entity-types.md): The 24 types Cloak detects and the four categories they belong to.
- [Limits and roadmap](https://docs.frostwolf.app/cloak/limits.md): What Cloak does not do yet, and the limits that apply today.
- [Claude Code](https://docs.frostwolf.app/integrations/claude-code.md): Cloak personal data and secrets before Claude sees them, and block prompt injection, in every Claude Code session.
- [Telemetry and capture](https://docs.frostwolf.app/operations/telemetry-and-capture.md): What the SDK reports to the console, what it never does, and how to turn bodies on.
- [Security and privacy](https://docs.frostwolf.app/operations/security-and-privacy.md): What is sent, what is stored, what is never sent, and who can see what.
- [The console](https://docs.frostwolf.app/console/overview.md): The dashboard at frostwolf.app/dashboard: overview, telemetry, Cloak, captures, keys, and team.
- [TypeScript SDK](https://docs.frostwolf.app/sdk/typescript.md): @frostwolfai/sdk: install, client options, methods, and defaults.
- [Python SDK](https://docs.frostwolf.app/sdk/python.md): frostwolf: install, client options, methods, and defaults.
- [Language notes](https://docs.frostwolf.app/sdk/language-notes.md): Where the TypeScript and Python SDKs differ, and why.
- [API overview](https://docs.frostwolf.app/api/overview.md): Authentication, errors, and the endpoints the SDKs call. Use it directly when there is no SDK for your language.
- [POST /v1/scan](https://docs.frostwolf.app/api/scan.md): Inspect a payload and return a verdict, optionally with the personal data in it.
- [POST /v1/sanitise](https://docs.frostwolf.app/api/sanitise.md): Inspect a payload and return it with every matched span redacted.
- [POST /v1/cloak/detect](https://docs.frostwolf.app/api/cloak-detect.md): Find personal data in a text and return where it is.
- [POST /v1/telemetry](https://docs.frostwolf.app/api/telemetry.md): Ship batched decision records to the console.
- [POST /v1/captures](https://docs.frostwolf.app/api/captures.md): Ship batched request and response bodies. Off unless capture is enabled for the key.
- [/v1/me](https://docs.frostwolf.app/api/me.md): Key metadata, and the capture setting.
- [Changelog](https://docs.frostwolf.app/changelog.md): Release notes for the SDKs and the control plane.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.