Your AI agent needs a brake pedal: stopping loops and runaway cost at runtime

Описание и применение

Слой безопасности SteerPlane ограничивает бесконечные циклы, расходы и запрещённые действия агентов, предотвращая дорогостоящие ошибки.

Задача

AI agents can repeat steps endlessly, incur high API costs, or perform unintended destructive actions.

Результат

Agents are prevented from looping indefinitely, exceeding cost ceilings, or executing forbidden actions, reducing costly failures.

Рабочий процесс

• 1. Import SteerPlane guard decorator. 2. Decorate agent function with max_cost_usd and denied_actions. 3. Run the agent; SteerPlane monitors cost and actions. 4. If cost exceeds limit or a denied action is attempted, the run stops.
• 5. Optionally set max_steps to limit iterations.

Доказательства

It stops an agent that keeps repeating the same steps.

Источник

It sets a hard cost ceiling per run.

Источник

It allows or denies tool calls by rule.

Источник