Introduction
The safety layer for the AGI era. Runtime guardrails for autonomous AI agents.
The safety layer for the AGI era. Runtime guardrails for autonomous AI agents. Observability tells you your agent failed — SteerPlane stops it before it does. It sits directly in the execution path and enforces cost ceilings, deterministic loop termination, and hierarchical policies — with one line of code and zero LLM-judge latency.
SteerPlane is Patent Pending — Indian Patent Application IN 202641071111 A1 (System and Method for Runtime Monitoring and Controlled Execution of Autonomous AI Agents).
Why SteerPlane
Autonomous agents call APIs, run code, and make real-world decisions. Without guardrails, a single misconfigured agent can:
- 🔄 spin into an infinite loop,
- 💸 burn thousands of dollars in API credits overnight, or
- 💀 take a destructive action — with zero ability to intervene.
Observability platforms can report these after the fact. SteerPlane is built to intervene in real time.
Core capabilities
Loop Termination
Deterministic O(W²) sliding-window detection of repeating action patterns — sub-millisecond, no LLM judge.
Cost Circuit-Breaker
Per-session USD ceilings across 25+ models, with mid-stream termination bounded to a single step of overshoot.
Policy Engine
Hierarchical deny → allow → rate-limit → approval evaluation with glob-pattern matching.
Deterministic Kill Mode
Terminate instantly on breach. Human-approval alert mode is available on the hosted/enterprise plan.
Zero-Code Gateway
An OpenAI-compatible proxy — change only the base URL. Provider keys never touch the agent.
Framework Integrations
LangChain, CrewAI, AutoGen, and the OpenAI Agents SDK — with your agent code unchanged.
Two ways to integrate
- SDK decorator — wrap a function with
@guard(...)for deep, in-process control. - Gateway proxy — point your OpenAI client at the SteerPlane gateway for zero-code enforcement.
Both routes feed the same guardrail pipeline: loop detection, cost tracking, and policy evaluation.