Built a framework-zero proxy to stop Cursor from burning tokens on boilerplate
Hey guys,
I just finished auditing a 17k line legacy codebase for a client and got absolutely crushed on LLM token costs. Cursor and Claude Code were wasting thousands of tokens reading useless framework bloat and redundant dependencies before even touching our actual logic.
I got fed up, so I spent the last few days building BlankLogic.io.
It’s a lightweight local proxy engine that strips out the repository noise and framework fluff before feeding the payload to your AI agents. It keeps your context hyper-focused and slashes your token bill.
Practicing what I preach, the stack is completely framework-zero: just vanilla HTML/CSS on the front, and PocketBase/Docker on the back.
If your AI limits are getting killed by heavy codebase context, grab the config JSON off the homepage and give it a spin.
Let me know what you think or if you hit any setup snags.