# NVIDIA NeMo Guardrails

> Source: https://agentsindex.ai/nvidia-nemo-guardrails

Open-source Python library from NVIDIA that adds programmable input, dialog and output rails to LLM apps.

NVIDIA NeMo Guardrails is an open-source Python library for adding programmable guardrails to LLM applications, blocking or validating unsafe, off-topic or malicious inputs and responses. Rails are defined in YAML and Colang and run through a Python SDK or an OpenAI-style guardrails server.

- Best for: NeMo Guardrails is best for developers who need to add policy checks around LLM inputs and outputs.

Starting price

Free

From [github.com/NVIDIA-NeMo/Guardrails](https://github.com/NVIDIA-NeMo/Guardrails), checked 11 November 2026

- MCP server Partly From docs.nvidia.com/nemo/guardrails/get-started/use-with-ai-agent, 11 November 2026 [docs.nvidia.com](https://docs.nvidia.com/nemo/guardrails/get-started/use-with-ai-agent)
- API Yes From docs.nvidia.com/nemo/guardrails/about-nemo-guardrails-library/overview, 11 November 2026 [docs.nvidia.com](https://docs.nvidia.com/nemo/guardrails/about-nemo-guardrails-library/overview)
- Open source Yes Checked by us github.com/NVIDIA-NeMo/Guardrails/blob/develop/LICENSE.md, 15 November 2026 [Checked by us](https://github.com/NVIDIA-NeMo/Guardrails/blob/develop/LICENSE.md)
- Self-host Yes

- GitHub stars: 7.3k

[Docs](https://docs.nvidia.com/nemo/guardrails) [GitHub](https://github.com/NVIDIA-NeMo/Guardrails)

What we checked

- Pages last read: 15 November 2026
- Facts checked: 5 facts · From 6 sources
- AI answers: Too few read · 58 buyer-question answers read, November 2026

[How we check](https://agentsindex.ai/methodology) [Editorial policy](https://agentsindex.ai/editorial-policy#not-for-sale)

[Visit the site](https://agentsindex.ai/go/nvidia-nemo-guardrails) [Compare](https://agentsindex.ai/compare?cat=agent-tools-and-integrations&products=nvidia-nemo-guardrails)

## At a glance

- Categories: [Agent tools and integrations](https://agentsindex.ai/categories/agent-tools-and-integrations) [Agent frameworks](https://agentsindex.ai/categories/agent-frameworks) [Securing AI use](https://agentsindex.ai/categories/security-ai-agents)
- Tags: Open source, Self-hosted

### Code

- Last push: 15 October 2026

### For developers

- Works with: GitHub, PyPI, ActiveFence, AlignScore, AutoAlign, Cisco AI Defense and 24 more

## NVIDIA NeMo Guardrails in our research

[All research](https://agentsindex.ai/blog)

Study 19 October 2026 – 16 November 2026

### [AI Tool Pricing Index, November 2026](https://agentsindex.ai/data/ai-tool-pricing-index)

Pricing

Free only

29 of the 361 tools we checked are free with no paid plan.

Edition price, checked 11 November 2026.

Pricing pages of the AI tools, November 2026 edition as of 16 November 2026 n = 361 tools checked

## Dialog rails steer whole conversations, beyond message filters

Most guardrail products check a prompt and a response, then block or pass. NeMo Guardrails runs rails at five stages (input, retrieval, dialog, execution and output), and the dialog layer is the unusual one. NVIDIA's own comparison claims it is "the only guardrails toolkit that also offers a solution for modeling the dialog between the user and the LLM". That claim is only NVIDIA's.

The product is a Python package configured with YAML and Colang files. Applications call it through the Python SDK, through a [self-hosted server with OpenAI-style chat completion endpoints](https://docs.nvidia.com/nemo/guardrails/about-nemo-guardrails-library/overview), or through LangChain and LangGraph. The LangChain integration is opt-in and switched on with an environment variable.

## Models and moderation checks come from many vendors

The NVIDIA name does not tie the library to NVIDIA models.

The [third-party API page](https://docs.nvidia.com/nemo/guardrails/configure-guardrails/guardrail-catalog/third-party) documents ready-made rails for ActiveFence, PolicyAI, AutoAlign, Clavata, Cleanlab, GCP Text Moderation, GuardrailsAI validators, Fiddler, Prompt Security and CrowdStrike AIDR, among others. Each needs that vendor's own API key, so those checks carry their own bills and terms. The most recent feature release notes add an [F5 AI Guardrails integration](https://docs.nvidia.com/nemo/guardrails/about-nemo-guardrails-library/release-notes) with configurable fail-open or fail-closed handling.

## The library is free; production support is sold as AI Enterprise

The library is licensed under [Apache 2.0](https://github.com/NVIDIA-NeMo/Guardrails), installed with `pip install nemoguardrails`, and has no plan, seat count or usage meter. Support for it is listed as "Open source community".

The paid path is a separate product. The [microservice page](https://docs.nvidia.com/nemo/guardrails/more-deployment-options/using-microservice) describes the NeMo Guardrails microservice as the production option, "available for the NVIDIA AI Enterprise plan", with support through that plan. It is not a drop-in upgrade: "It is not a standalone solution" and it "requires other microservices to function" inside the NeMo microservices platform. The same page rates the library's own server as "Limited (self-managed and no high availability support out of the box)" and recommends it for proofs of concept, development and testing.

NVIDIA publishes no price for AI Enterprise. Its [product page](https://www.nvidia.com/en-us/data-center/products/ai-enterprise/) offers a "Request a Free 90-Day License" button and a contact link, not a rate card. Configurations move between the two: the overview says "Configurations are portable between the library and microservice".

## The guardrails server has no authentication of its own

The optional Actions Server is not isolated from the Guardrails server by default, and any caller past the deployer's gateway can invoke any endpoint or registered action, because the library does no per-endpoint or per-action authorization.

The guardrails in this product protect the model's inputs and outputs; securing the guardrails service is left to the team running it. NVIDIA invites proposals for "an optional in-process authentication layer" as feature requests, which places it as a gap rather than a shipped option.

## Anonymous telemetry to NVIDIA is on by default

NVIDIA states that prompts, completions, API keys, endpoints, token counts, file paths, usernames and IP addresses are not included.

Each event is also written to a local audit file at `~/.config/nemoguardrails/usage_stats.json`. The opt-out covers the library only, not whichever model endpoint it calls.

## Upgrades regularly break integration code

The project is pre-1.0, and its [release notes](https://docs.nvidia.com/nemo/guardrails/about-nemo-guardrails-library/release-notes) carry a Breaking Changes section. The latest feature release removed the `output_mapping` parameter from `@action` (passing it "now raises TypeError"), changed what `IORails.generate()` returns and its parameter order, stopped the `/v1/checks` endpoint accepting inline configurations, renamed the Cleanlab, Fiddler and GCP action names used in custom Colang 1.0 flows, and dropped the `hf-classifier` install extra.

The same notes show steady work on the server: liveness endpoints, OpenAI-compatible error envelopes, and fixes so that output rails fail closed when an action fails instead of passing silently. Teams adopting it should budget for code changes on most upgrades and pin a version, which the [versioned documentation](https://docs.nvidia.com/nemo/guardrails/about-nemo-guardrails-library/overview) supports with snapshots per release.

For AI coding agents, NVIDIA hosts a [documentation MCP server](https://docs.nvidia.com/nemo/guardrails/get-started/use-with-ai-agent) that serves the docs only, plus Markdown versions of every page. It does not expose the guardrails runtime as tools.

## NVIDIA NeMo Guardrails price and terms

### Terms

- Free plan: Yes The Apache 2.0 license grants a perpetual no-charge, royalty-free license

The terms in full

NVIDIA-hosted models used with the library require an NVIDIA API key.

### For AI agents

- MCP server: Partly [docs.nvidia.com, 11 November 2026](https://docs.nvidia.com/nemo/guardrails/get-started/use-with-ai-agent)
- API: Yes [docs.nvidia.com, 11 November 2026](https://docs.nvidia.com/nemo/guardrails/about-nemo-guardrails-library/overview)

[Developer docs](https://docs.nvidia.com/nemo/guardrails)

Where these prices come from

[github.com/NVIDIA-NeMo/Guardrails](https://github.com/NVIDIA-NeMo/Guardrails), checked 11 November 2026

### All plans and limits

| Plan | Price | Main limits |
| --- | --- | --- |
| NVIDIA NeMo Guardrails | Free | None stated |

Read 11 November 2026 from [github.com/NVIDIA-NeMo/Guardrails](https://github.com/NVIDIA-NeMo/Guardrails).

## What we checked on NVIDIA NeMo Guardrails

Yes Partly No

What it costs to start No

### What the price counts

The library license is no-charge and royalty-free, so NVIDIA does not document a paid pricing meter.

> perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license

Section 2, Grant of Copyright License [raw.githubusercontent.com/NVIDIA-NeMo/G…](https://raw.githubusercontent.com/NVIDIA-NeMo/Guardrails/develop/LICENSE-Apache-2.0.txt) Checked 30 October 2026

What it costs to start Yes

### Free to use

The library is an open-source Python package installed from PyPI and licensed under Apache 2.0, so it can be used without paying; third-party model endpoints carry their own terms.

> The NVIDIA NeMo Guardrails library (PyPI | GitHub) is an open-source Python package for adding programmable guardrails to LLM-based applications.

Overview page, opening paragraph [docs.nvidia.com/nemo/guardrails/about-n…](https://docs.nvidia.com/nemo/guardrails/about-nemo-guardrails-library/overview) Checked 11 November 2026

What it works with Yes

### Public API

Its self-hosted API server exposes OpenAI-compatible REST endpoints for guarded chat completions and model listing.

> You can integrate the NVIDIA NeMo Guardrails library directly into a Python application or run a FastAPI service and call endpoints over HTTP/REST.

Overview page, Integration Paths section [docs.nvidia.com/nemo/guardrails/about-n…](https://docs.nvidia.com/nemo/guardrails/about-nemo-guardrails-library/overview) Checked 11 November 2026

What it works with Partly

### MCP server

NVIDIA hosts a documentation MCP server that lets coding agents retrieve the Guardrails docs; it serves documentation only and does not expose the guardrails runtime itself as MCP tools.

> The MCP server lets your agent retrieve the canonical documentation without cloning the GitHub repository.

Use the Documentation MCP Server section [docs.nvidia.com/nemo/guardrails/get-sta…](https://docs.nvidia.com/nemo/guardrails/get-started/use-with-ai-agent) Checked 11 November 2026

What happens to your data Yes

### Open source

Apache-2.0 licence in its public repository

> Apache License, Version 2.0

LICENSE.md in github.com/NVIDIA-NeMo/Guardrails [github.com/NVIDIA-NeMo/Guardrails/blob/…](https://github.com/NVIDIA-NeMo/Guardrails/blob/develop/LICENSE.md) Checked by us, 15 November 2026

## Questions about NVIDIA NeMo Guardrails

- How much does NVIDIA NeMo Guardrails cost?: NVIDIA NeMo Guardrails is free to use. From github.com/NVIDIA-NeMo/Guardrails, checked 11 November 2026.
- Does NVIDIA NeMo Guardrails have an API or an MCP server?: NVIDIA NeMo Guardrails has an API. From docs.nvidia.com/nemo/guardrails/about-nemo-guardrails-library/overview, checked 11 November 2026. NVIDIA NeMo Guardrails has a limited MCP server. From docs.nvidia.com/nemo/guardrails/get-started/use-with-ai-agent, checked 11 November 2026.
- Is NVIDIA NeMo Guardrails open source, and can you host it yourself?: NVIDIA NeMo Guardrails is open source. We checked github.com/NVIDIA-NeMo/Guardrails/blob/develop/LICENSE.md ourselves on 15 November 2026. You can host NVIDIA NeMo Guardrails yourself.
- What have we checked on NVIDIA NeMo Guardrails?: 5 facts, under these headings: What it costs to start, what it works with, what happens to your data.
- How do we check NVIDIA NeMo Guardrails?: We hold 5 facts on NVIDIA NeMo Guardrails, quoted from 6 pages. Facts last checked 15 November 2026. Price last checked 11 November 2026. Last change found 15 November 2026.

[How we check every listing](https://agentsindex.ai/methodology)

## Sources and changes

Last full check 11 November 2026

Sources 5

- Read 11 November 2026 · [docs.nvidia.com](https://docs.nvidia.com/nemo/guardrails/about-nemo-guardrails-library/overview) /nemo/guardrails/about-nemo-guardrails-library/overview
- Read 30 October 2026 · [raw.githubusercontent.com](https://raw.githubusercontent.com/NVIDIA-NeMo/Guardrails/develop/LICENSE-Apache-2.0.txt) /NVIDIA-NeMo/Guardrails/develop/LICENSE-Apache-2.0.txt
- Read 11 November 2026 · [github.com](https://github.com/NVIDIA-NeMo/Guardrails) /NVIDIA-NeMo/Guardrails
- Read 11 November 2026 · [docs.nvidia.com](https://docs.nvidia.com/nemo/guardrails/get-started/use-with-ai-agent) /nemo/guardrails/get-started/use-with-ai-agent
- Read 30 October 2026 · [docs.nvidia.com](https://docs.nvidia.com/nemo/guardrails/run-guardrailed-inference/using-fastapi-server/overview) /nemo/guardrails/run-guardrailed-inference/using-fastapi-server/overview

What changed 2

- 15 November 2026 What it does
- 11 November 2026 What it does

Repository

- Stars: 7,305
- Last push: 15 October 2026

## NVIDIA NeMo Guardrails price history

We have read the price once so far, on 11 November 2026: free.

| Price | Plan | First seen | Last seen | Readings | Source |
| --- | --- | --- | --- | --- | --- |
| Free |   | 11 November 2026 | 11 November 2026 | 1 | [Pricing page, read 11 November 2026](https://github.com/NVIDIA-NeMo/Guardrails) |

From our research on 11 November 2026. A date is when we read the page, not when the price changed.

## Similar in Agent tools and integrations

[See all in Agent tools and integrations](https://agentsindex.ai/categories/agent-tools-and-integrations)

- [Base2026](https://agentsindex.ai/base2026) Free SEO research MCP server, public API and browser SEO checks with original sources attached. Free plan
- [CurlShot](https://agentsindex.ai/curlshot) Screenshot and PDF rendering API with an MCP server, so an agent can see any web page. Free plan, then $19 a month
- [Bot Street](https://agentsindex.ai/botstreet) Marketplace where AI agents post offers, take paid tasks and run shops, reached over MCP, a CLI or REST. Free plan

## Is NVIDIA NeMo Guardrails yours?

Free, and the page then carries the verified owner mark. You confirm it with an email at docs.nvidia.com. [Claim this listing](https://agentsindex.ai/claim/nvidia-nemo-guardrails?from=owner-line)

[Report an error](mailto:hello@agentsindex.ai?subject=Report%20an%20error%3A%20NVIDIA%20NeMo%20Guardrails%20(nvidia-nemo-guardrails)&body=What%20is%20wrong%20on%20the%20page%2C%20and%20where%20on%20the%20product) if a fact is wrong.
