Inferrail：自托管 LLM 网关，追踪 Cursor/AI 智能体的成本（不存提示词与响应）
原文标题：Inferrail: Self-hosted LLM gateway to track your Cursor/AI agent costs (Payload-free)
开发者发布自托管开源 LLM 网关 Inferrail，可把模型调用归集为单次任务的成本，例如 20 次模型调用对应 1 个任务、花费 $0.43，且不保存提示词和响应。
当前语言的正文正在等待翻译，暂时显示原文。
2026 年9 月 26 日 16:22 1
I kept running into a surprisingly simple question while building with AI:
What did that piece of work actually cost?
Not the monthly bill. Not total token usage. The actual job.
So I built Inferrail.
It turns model calls into something closer to:
20 model calls → 1 job → $0.43
It tracks token usage and cost across the work, while keeping prompts and responses out of the receipts.
It’s self-hosted, open source, and still early. I’m not looking to sell anyone here anything. I’d actually rather have people break it and tell me what’s missing.
Curious whether other people building heavily with Cursor/agents have the same problem: do you actually know what one completed piece of AI work costs you?
来源：Cursor Forum · Showcase · forum.cursor.com