Hey Product Hunt! I'm Vijay, a second-year CS/AI student at PES University in Bengaluru. I kept reading about AI agents going wrong in production, stuck in a loop, burning through API budget overnight, taking an action nobody approved, and every tool I found could only tell you about it after it already happened. That gap is what got me building SteerPlane. It sits directly inside an agent's execution path, either through a Python decorator or a gateway proxy you point your OpenAI client at, and enforces cost ceilings, loop detection, and policy rules before an action goes through, not after. No LLM judging another LLM, just deterministic checks that behave the same way every time. The core detection mechanism has a published Indian patent behind it, and the whole thing is open source and free to self-host, forever. There's a hosted version with a couple of extra pieces for teams that don't want to run it themselves, but the core guardrail engine is never behind a paywall. I built the first version in about two months, in my first year of college, on top of a full course load. Would love feedback, especially from anyone who's had an agent actually go sideways in production.

SteerPlane

Stop AI agents before they burn your budget, not after

Stop AI agents before they burn your budget, not after