How I built a hard budget-cap proxy to stop runaway LLM API bills

Hey everyone. If you've ever built scripts or side projects using LLM APIs, you probably know the low-key anxiety of accidentally getting stuck in a loop and waking up to an insane cloud bill.
​Big enterprise tools for managing this are usually way too bloated for a solo dev. I just wanted something simple and self-hosted that actually works. So, I built CloudGrip.
​Basically, it’s a lightweight proxy gateway built with Node.js and Express that sits between your code and the LLM provider. It tracks your token usage in real time using Supabase, and it has a hard budget circuit breaker. The moment your spend hits your set limit, it cuts off execution so you don't leak money. I've also wired up a live Paystack subscription flow for it.
​If you want to check it out or test it out, the live project is here: https://cloudgrip-ai.onrender.com
​How do you guys usually handle API budget safety when building AI tools? Would love to hear how other devs manage it