Verified project record

NVIDIA-NeMo/Guardrails

NeMo Guardrails adds programmable guardrails to LLM-based conversational applications, offering input, output, and dialog-level controls to help mitigate risks like prompt injections and jailbreaks. It is a Python library and server that developers configure to define how an LLM should behave and interact with tools.

Project overview

Project type: AI Agent · Infrastructure
Use cases: Knowledge Q&A
Deployment: Refer to project documentation
License: License pending

Best for

Developers and AI engineers who have Python coding knowledge and understand LLM concepts, and who need to program specific controls for conversational agents.

Key capabilities

Applies checks to user input to reject or alter it, such as masking sensitive data or preventing jailbreaks.
Influences how the LLM is prompted and determines if an action should be executed or if a predefined response should be used.
Applies checks to the output generated by the LLM to reject or alter it, such as removing sensitive data.
Applies checks to retrieved chunks in retrieval-augmented generation scenarios to reject or alter them before prompting the LLM.
Applies checks to the input and output of custom actions, or tools, called by the LLM.
A server that loads guardrails configurations and exposes an HTTP API for chat completions.
A built-in CLI to start a server, start an interactive chat session, and run evaluations.
A modeling language created for designing flexible and controllable dialogue flows.

Limitations and risks

The built-in guardrails may or may not be suitable for a specific production use case without additional customization.
Third-party endpoints used by the library have separate terms and privacy practices independent of the library's telemetry opt-out.

Getting started

Setup difficulty is medium because it requires Python coding knowledge and an understanding of LLM guardrails concepts. First success involves installing the package with pip, loading a configuration using RailsConfig.from_path, and calling rails.generate with messages.

Evidence and sources

GitHub project description: NeMo Guardrails is an open-source toolkit for easily adding programmable guardrails to LLM-based conversational systems.
README: NVIDIA NeMo Guardrails library is an open-source toolkit for easily adding *programmable guardrails* to LLM-based conversational applications.
README: The NeMo Guardrails library provides several mechanisms for protecting an LLM-powered chat application against common LLM vulnerabilities, such as jailbreaks and prompt injections.
README: The NeMo Guardrails library enables developers building LLM-based applications to add **programmable guardrails** between the application code and the LLM.
README: If you use the NeMo Guardrails library, cite the [EMNLP 2023 paper](https://aclanthology.org/2023.emnlp-demo.40) that introduces it.

Reviewed at: 2026-07-15T15:54:29Z