SteerPlane

SteerPlane is a runtime control plane for autonomous AI agents. Observability tells you your agent failed, SteerPlane stops it before it does. Cost circuit-breakers, loop termination, and deterministic kill-mode enforcement, patent published. 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.