Is it possible to use NeMo Guardrails only for input/output moderation (without using LLM for generation)?

[mu-fazil-vk]

Hi team,

I'm building a custom chat orchestration flow and would like to use NeMo Guardrails only for content moderation, not for response generation.

Specifically:

• I want to run input guardrails (e.g., jailbreak detection, PII masking) on user input,
• Then pass the cleaned input to my custom LLM pipeline (outside NeMo),
• And finally run output guardrails (e.g., hallucination or tone checks) on the generated response,
• Without NeMo performing any llm_rails.generate() call for generation.

In short, I’d like to use NeMo as a guardrail middleware/orchestrator that wraps input/output moderation only.

I went through the docs and didn’t find a direct way to do this with just input/output rails.
Thanks!

[Pouyanpi]

Hi @mu-fazil-vk yes it is possible, you can use input or output rails only option. Please have a look at https://docs.nvidia.com/nemo/guardrails/latest/user-guides/input-output-rails-only/README.html#using-only-input-and-output-rails

[mu-fazil-vk]

Hi @Pouyanpi , thank you for your response — it was really helpful to me.