一个失控的 AI 代理让我在睡觉时损失了 47 美元（以及我如何解决这个问题）

Waking up to a hefty API bill due to an AI agent stuck in a loop is a frustrating experience. The author recently incurred a cost of $47 while debugging a retry loop in LangGraph, where the agent kept failing, LangGraph kept retrying, and OpenAI continued to charge. To prevent others from facing similar costly lessons, the author created Shekel, a zero-config, open-source Python library designed for budget enforcement and cost tracking for LLMs. It is compatible with LangGraph, CrewAI, AutoGen, and any framework that interacts with OpenAI, Anthropic, or LiteLLM. This tool aims to provide better guardrails for AI agents and help users manage their expenses effectively.