NeMo Guardrails
Toolkit for programmable rails around LLM apps: topic, safety, jailbreak and fact checks.
Made by NVIDIAAI guardrails
About NeMo Guardrails
NeMo Guardrails is an open-source toolkit for adding programmable guardrails to LLM-based conversational systems. NVIDIA makes the NeMo Guardrails library. It is built for developers building LLM-based applications, who add programmable guardrails between the application code and the LLM.
It is designed to protect LLM-powered chat applications against common vulnerabilities such as jailbreaks and prompt injections. One listed use case is question answering over a set of documents, with fact-checking and output moderation. It can be used with multiple LLMs, including OpenAI GPT-3.5, GPT-4, LLaMa-2, Falcon, Vicuna, and Mosaic.
Drafted from github.com · not yet confirmed by NVIDIA
How it works
NeMo Guardrails lets you define rails for a conversational app in a small language called Colang: what topics are allowed, how to respond to jailbreak attempts, when to fact check, which tools can be called. It sits between the user and the model, runs input and output through the rails, and can plug in third party checks like Llama Guard, Presidio or Perspective. It's the most complete open framework for the control layer around an LLM.
From the project
From the project
Reviews
Be the first to review it
AnonymousKnow a team weighing NeMo Guardrails? Send them this page.