NVIDIA rule library for applications with models

NVIDIA NeMo Guardrails.

Programmable rules for model inputs, outputs, and actions

4.30 / 5Directory score
Official website

The developer sets checks around the application with the model and chooses which inputs, outputs or actions to handle separately. The rules should be checked with examples.

NVIDIA Guide to Rules for Language Model Applications
NVIDIA Guide to Rules for Language Model Applications. Open the source

First, describe the boundary in words.

The training manual assistant should explain the order of the application and can show an example. A separate permit is required to change the application. In NeMo Guardrails, checks are configured around the model and application. First, identify the permitted actions and where they need to be checked. At the input, you may need to check the topic, at the output, check the response form, and before turning to the tool, a separate rule for its arguments.

Check permission and refusal

Write pairs of requests that differ in one meaningful condition. For one pair, ask for an explanation of the return policy, then ask the assistant to issue a return on the user's behalf. Add harmless questions that use words from prohibited examples. This exposes both missed violations and false positives. Record the response and actual tool calls. A refusal message does little good if the application still passed the data to the tool.

Separate verification from authorization

Control the credentials and permissions of connected systems separately. For the training check, use stubs and fictional entries to ensure that the result of the erroneous classification remains visible in the journal. When changing the model or template of the dialogue, repeat successful and borderline cases. Consider the delay of additional checks and find out what happens when they fail. The decision to block or transfer to the operator must be understandable and reproducible. Start with a short run, where you can name a specific rule for each refusal.

Suitable tasks

  • Check the processing of the request outside the specified topic
  • Track failure before transferring data to tool
Features

Strengths

  • Rules can be described programmatically
  • The official repository contains configuration examples
Things to consider

Limitations

  • A rule can miss a violation or block a permitted request
  • Restrict tool permissions separately

Selection and conditions

Free access
Yes
Trial period
Not confirmed
No sign-up required
Not confirmed
Russian interface
Not confirmed
Local installation
Yes
API
Not confirmed

“Not confirmed” means there is not enough information. The feature may still be available; check the developer’s terms.

Input formats
Text, Code
Result
Text, Data

Terms

The toolkit is open source. Selected models, extra checks, and hosting can incur charges. A rule can miss a violation or block a permitted request.

From task to first result

Practical example

Task
Check the processing of the request outside the specified topic
What to prepare
A test copy of the assistant and a fictional read and write policy.

Task brief or prompt

Run regular and border requests. Check that the valid reading is working, and the action without permission is not in the journal.

Expected result

Report with inputs, observed actions and comments to the rules.

How to check the result

  • Check the tool log separately from the response text.
  • Test only in an agreed environment.

An editorial teaching example of a task brief, not the result of hands-on testing of the service. Check that the required mode is available, that you have rights to the source material and that the intended use of the result is permitted.

Check details

Official sources

  1. NVIDIA NeMo Guardrails
  2. NeMo Guardrails official repository

This description is based on the developer’s material. Features, prices and availability may change; check the primary source before use.