NeMo Guardrails

license

Apache-2.0(OSI)

source

public(full Python implementation)

self-host

yes(library/CLI/Docker/server)

service

none

core features withheld

no

NeMo Guardrails is Apache-2.0, with the complete source public and self-hostable as a Python library, CLI, Docker image or server. Its safety rails can use the application's own LLM, community models or third-party services as the checker, so NVIDIA's safety models and the separately packaged Guardrails microservice are optional rather than required.

• https://docs.nvidia.com/nemo/guardrails/ recorded 2026-09-29

  The NVIDIA docs overview describes the library as "an open-source Python package for adding programmable guardrails to LLM-based applications", distributed as a PyPI package for a "Self-managed Python environment". It lists content-safety approaches beginning with "Use the LLM itself to check inputs and outputs for harmful content" alongside community models and third-party APIs, so NVIDIA's own safety models are one option among several. The separate "NVIDIA NeMo Guardrails microservice provides a production-ready container image built on top of the same guardrails configuration model."

• https://github.com/NVIDIA/NeMo-Guardrails recorded 2026-08-13

  Redirects to NVIDIA-NeMo/Guardrails. README states 'The NeMo Guardrails library is licensed under the Apache License, Version 2.0' and describes it as 'an open-source toolkit for easily adding programmable guardrails to LLM-based conversational applications', installable and runnable yourself. Repo metadata reports spdxId NOASSERTION, which is GitHub failing to classify the file rather than a different license. No enterprise directory, commercial edition or license key on the page.

• https://pypi.org/pypi/nemoguardrails/json recorded 2026-09-29

  The PyPI JSON for the nemoguardrails package (version 0.24.1) sets the license expression to "Apache-2.0" and points its repository URL at "https://github.com/NVIDIA-NeMo/Guardrails". The embedded README says "NVIDIA NeMo Guardrails library is an open-source toolkit for easily adding" guardrails, closes with "The NeMo Guardrails library is licensed under the" Apache License, Version 2.0, and documents self-hosting: "You can use the NeMo Guardrails library CLI to start a guardrails server" and "To start a guardrails server, you can also use a Docker container." Searched the body for enterprise, license key and commercial edition: none.

• https://raw.githubusercontent.com/NVIDIA/NeMo-Guardrails/develop/LICENSE.md recorded 2026-09-17

  The LICENSE.md body, read rather than the classifier label. It opens with two SPDX lines - "SPDX-FileCopyrightText: Copyright (c) 2023 NVIDIA CORPORATION & AFFILIATES" and "SPDX-License-Identifier: Apache-2.0" - above the Apache License 2.0 boilerplate. The SPDX header is what the classifier cannot match.

Adoption is measured by PyPI downloads of the nemoguardrails package, whose project URLs confirm it belongs to NVIDIA-NeMo/Guardrails rather than an unrelated package of the same name. PyPI is the primary distribution channel for a Python library like this one, so downloads are a better signal here than GitHub stars.

• https://pypistats.org/api/packages/nemoguardrails/recent recorded 2026-08-13

  last_month 401,369

Beyond checking inputs and outputs, it applies rails to retrieved content and tool calls, and its Colang language scripts whole dialog flows, which goes further than scanners that only inspect individual messages.

• https://docs.nvidia.com/nemo/guardrails/ recorded 2026-09-29

  The NVIDIA docs overview says "Input, retrieval, dialog, execution, and output rails run at different stages of an LLM interaction" and that "Colang defines conversational flows, guardrail logic, and event-driven behavior." Its use cases cover content safety, jailbreak detection, topic control ("Control what topics the bot can and cannot discuss"), PII masking and agentic security ("Execute rails that validate tool inputs and outputs before and after invocation"), and it lets teams "Combine built-in guardrails, NVIDIA safety models, community models, third-party APIs, and custom Python actions."

• https://github.com/NVIDIA/NeMo-Guardrails recorded 2026-08-13

  README describes programmable guardrails for LLM applications, with dialog rails operating on canonical-form messages via the Colang Guide, fact-checking and output moderation over retrieval pipelines, keeping an assistant on topic, and deciding whether an action should be executed - that is I/O moderation, dialog control, retrieval filtering and tool-use control.