> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://www.comet.com/docs/opik/guardrails/llms.txt.

# Guardrails

> Debug, evaluate, and monitor your LLM applications, RAG systems, and agentic workflows with comprehensive tracing, automated evaluations, and production-ready dashboards.

## Docs

- [Guardrails overview](https://www.comet.com/docs/opik/guardrails/overview.md): Learn what guardrails are, why they matter, and when to use guardrails versus online evaluation to protect your LLM application.
- [Policies](https://www.comet.com/docs/opik/guardrails/policies.md): Create and manage named guardrail policies in the Opik UI, then build a guardrail from them with Guardrail.from_stored_policies.
- [Guards](https://www.comet.com/docs/opik/guardrails/guardrails.md): The five guard types available in Opik guardrails — PII, topic, prompt injection, LLM judge, and custom classifier — what each one checks and what you configure.
- [Prompt injection model](https://www.comet.com/docs/opik/guardrails/prompt-injection-model.md): Opik Guard is Opik's custom-trained model behind the prompt injection guardrail, built on real attacks and evaluated out-of-distribution for honest generalization.
- [Custom models](https://www.comet.com/docs/opik/guardrails/custom-guardrails.md): Fine-tune a language model on your own labeled examples to create a custom guardrail in Opik, then run it from a policy. Opik hosts the training and serves the model for you.
- [Guardrails server](https://www.comet.com/docs/opik/guardrails/server.md): How to install, configure, start, and call the Opik guardrails backend that runs the PII, Topic, prompt injection, and custom classifier models.