Solutions  ·  2026-07-02

NVIDIA NeMo Guardrails v0.23.0 — IORails Tool Calling for Streaming + Non-Streaming, Local Rail Validation of Tool Results

SolutionsMedium impactGlobal
Released July 1, 2026, NeMo Guardrails v0.23.0 expands IORails tool-calling to cover both streaming and non-streaming requests. New capabilities include local rails that validate model-emitted tool calls and application-returned tool results — meaning guardrails can now inspect the full tool invocation cycle, not just prompt/response pairs. Observability improvements also shipped in this release.
Tool-call validation closes the most significant gap in LLM guardrail coverage: prior versions could filter prompts and outputs but could not inspect whether a model's tool-use decision or the tool's return value was safe or policy-compliant. This is directly relevant to MCP-connected agents where tool results can carry injected payloads.
Teams using NeMo Guardrails in agentic pipelines with tool use should upgrade to v0.23.0 immediately to gain validated tool-call inspection; particularly important for streaming agent deployments where tool results flow continuously.
Sources
GitHub — NVIDIA/NeMo-Guardrails v0.23.0 Release
See this in the live feed Explore related AI security and governance findings — updated every morning.
Open the feed →