About Stoplight

Developers are already running agents with powerful tools and very little safety infrastructure. Security teams typically respond with blanket blockades and policy docs that are hard to enforce and impossible to audit, while developers increasingly rid themselves of approval annoyances and opt toward leaving --dangerously-skip-permissions on.

We believe the guardrails that'll actually mean the most during an era of rapid agentic workflow adoption are the ones that genuinely stay out of your way: quick to install, self-assembling from real usage, and only interrupting work when it has a specific reason.

As it turns out, it's not so easy to understand precisely which guardrails actually increase safety for particular agentic workflows, which ones just interrupt agents needlessly. Stoplight aims to fully automate this by first developing a clear understanding of what it is your agents are actually doing, and generating surgical rules that go well beyond simple glob patterns.