Build a context window manager in 24h: stick to the norm or write your own?

Here’s what I did:

1. Built a proxy that intercepts Codex’s calls to OpenAI and rewrites them on the fly.

2. Replayed 3,807 rounds of SWE-bench Verified traces through it: avg prompt 44k → 6k tokens (-87%).

3. Posted it here to get the next reduction applied to my confidence interval — starting with the inevitable “How about accuracy?”

npx -y pando-proxy