Agent tool / platformFree source toolLiveFrom public docs
NeMo Guardrails
NVIDIA
Open-source toolkit for programmable guardrails on LLM inputs, outputs and tool calls.
Overview
NeMo Guardrails adds controllable rules to LLM applications, including execution rails on the input and output of actions (tools) the model calls. Use it as a Python library or run its HTTP guardrails server.
Production criteria
| Approval gatesCan a person or policy hold an action until it is approved? | Not claimed Not documented by the vendor or not yet reviewed. |
|---|---|
| Restart recoveryDoes work resume after a crash or redeploy without repeating finished steps? | Not claimed Not documented by the vendor or not yet reviewed. |
| Evidence trailCan you inspect what ran, with what inputs, and who approved it? | Not claimed Not documented by the vendor or not yet reviewed. |
| Kill switchCan an operator withdraw permission or stop further actions? | Not claimed Not documented by the vendor or not yet reviewed. |
| Policy controlsDo declared rules decide which tools and actions are allowed? | Built in Execution rails check the input and output of tool actions. |
| Sandbox isolationDoes code or tool execution run inside an isolated boundary? | Not claimed Not documented by the vendor or not yet reviewed. |
Marks reflect the documentation linked below as of 2026-10-29. How we mark criteria.
Limits and scope
- Rails are model-adjacent checks; pair with an enforcement point at the executor for irreversible actions.
Free tools, a product pilot, or a focused paid diagnosis.
Choose a tool from the directory, explore Mothership’s technical pilot path, or check fit for a software diagnosis when one issue needs investigation.