Glossary
Guardrail
Guardrail is defined here as follows. Programmable check that can block, alter, or validate user input, model output, or a tool call. NeMo Guardrails runs these checks as rails around an LLM application. The directory source for this term is NVIDIA NeMo Guardrails overview.
Last verified 2026-10-17
Sources
Listings
Directory listings whose sources describe this mechanism.
| Name | Status | Category | Open source | Mechanism |
|---|---|---|---|---|
| Guardrails AI | Live | Governance and mandates | Yes | Python framework that runs input and output guards to detect and mitigate specific risks, and can also extract structured data from model output. guardrails, validation, policy |
| Invariant | Live | Governance and mandates | No | Contextual security layer that checks an agent trace against declarative rules for tool calls, data flow, secrets, and looping. guardrails, policy, traces |
| Lakera Guard | Live | Governance and mandates | No | Guard API that screens an agent conversation in one call, including tool definitions, tool calls, and tool results, against a project policy. guardrails, tool-calls, policy |
| LiteLLM Proxy | Live | Governance and mandates | Yes | Open source proxy that enforces personal, team, and agent spend budgets on model calls when a database is connected, and can attach guardrails. budgets, spend-limits, guardrails |
| Mastra | Live | Frameworks and runtimes | Yes | Open source TypeScript framework for agents and workflows, with memory, tool approvals, and processors that can block or redact content. typescript, orchestration, approvals |
| NeMo Guardrails | Live | Governance and mandates | Yes | Open source toolkit for programmable rails that block, alter, or validate user inputs, model responses, and tool calls. guardrails, policy, tool-calls |
| OpenAI Agents SDK | Live | Frameworks and runtimes | Yes | Library that runs an agent loop in your application, with tools, handoffs, guardrails, sessions, and human approval before a tool runs. orchestration, tools, approvals |
| Pydantic AI | Live | Frameworks and runtimes | Yes | Typed Python agent SDK with structured outputs, guardrails, and companion packages for graph control flow and spending limits. python, orchestration, guardrails |
| Strands Agents | Live | Frameworks and runtimes | Yes | Open source SDK for a model driven agent loop in Python and TypeScript, with hooks, execution limits, and guardrails around each step. orchestration, guardrails, tools |
Questions
- What is Guardrail?
- Programmable check that can block, alter, or validate user input, model output, or a tool call. NeMo Guardrails runs these checks as rails around an LLM application.
- Which listings use Guardrail?
- Guardrails AI, Invariant, Lakera Guard, LiteLLM Proxy, Mastra, NeMo Guardrails, OpenAI Agents SDK, Pydantic AI, Strands Agents.
- When was Guardrail last verified?
- Last verified 2026-10-17.