Home / Glossary / What are AI agent guardrails?

What are AI agent guardrails?

AI agent guardrails are constraints that keep an autonomous agent within safe bounds. They range from output-quality checks (filtering, validation) to runtime control guardrails — hard limits on cost, loops, steps and duration that stop a run before it goes off the rails.

Guardrails fall into two broad families: content guardrails that judge what an agent says or does, and runtime guardrails that bound how long and how expensively it runs.

AgentBrake is a runtime control guardrail — a circuit breaker. It doesn't score output; it stops a run that loops, blows a cost ceiling, or never terminates. It works on LangChain and CrewAI in one line.

Related

Stop your next runaway agent

One line of code. Open source. Works on LangChain 1.x, CrewAI 1.x, and the classic AgentExecutor.

$pip install agentbrake-sdk
Star on GitHub

Frequently asked questions

Are guardrails the same as a kill-switch?
A kill-switch is one kind of runtime guardrail. AgentBrake is a runtime guardrail that acts as a per-run kill-switch for cost and loops.
Do guardrails slow my agent down?
Runtime control guardrails like AgentBrake run in-process with negligible overhead and no proxy in the request path.