Add programmable guardrails to LLM-based conversational applications.
An open-source toolkit for adding programmable guardrails to LLM conversational systems.
NeMo Guardrails is a Python toolkit that helps you control LLM outputs and steer conversational flows. It suits developers building secure, domain-specific chatbots and RAG applications.
It lets you define input, dialog, retrieval, execution, and output rails to filter sensitive data, enforce fact-checking, and prevent jailbreaks. You can integrate it into your code using either the Python API or a standalone guardrails server.
It suits developers and engineers integrating LLMs into production environments who need strict conversational control and safety checks. It is not designed for non-technical users or those looking for a ready-to-use end-user chat interface.
Apply input, dialog, retrieval, execution, and output rails to thoroughly filter data.
Build performant applications using native synchronous and asynchronous methods.
Enforce pre-defined conversational paths and standard operating procedures using Colang.
Defend your chat application against common LLM vulnerabilities like jailbreaks and prompt injections.
The toolkit is free under its open-source license, and the only financial cost is running your own infrastructure, models, and API calls.
879 open remote AI-training roles, updated November 15, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
Python 3.11, 3.12, and 3.13 are supported. Python 3.10 support was dropped in version 0.25.0.
Yes, you can add a guardrails layer around your LangChain chains by setting an environment variable.
You can steer dialog paths using Colang, a domain-specific language designed for conversational interactions.
See Zekai first in Google