Skip to content
NVIDIA NeMo Guardrails Releases· Pouyanpi·Indexed by this site · Originally publishedDate not given· Tools →· Its place in the scan record →SelectedAttention38

NVIDIA NeMo Guardrails v0.25.0 released: adds agent tool call checks and a gateway

v0.25.0

AI summary

NVIDIA NeMo Guardrails released version 0.25.0, adding agent tool call and result checks, a hooks server for Claude Code, Codex and NeMo Relay, and NeMo Guardrails Gateway, which applies input and output checks around existing model APIs. The gateway supports buffered and streaming OpenAI Chat Completions and Responses and an explicitly selected LiteLLM dialect; the scope of checks is limited to a single user text input and a single assistant text output, and does not check conversation history, tools, audio or images. New guardrails include LLM-judged tool safety checks, CEL expressions and language filtering, and restore Zscaler AI Guard input and output moderation.

Why it was selected

This version extends guardrail checks to agent tool calls and results and provides a gateway, so teams deploying agent guardrails can use it to assess breaking changes in the upgrade.

Read originalgithub.com