codex-lb is a Python proxy for people who use several Codex and ChatGPT accounts: it balances load across them. It is MIT licensed, has 3.3K+ stars and belongs to another maintainer. I contributed to it, and four of my pull requests were merged between April and May 2026.
Three of the fixes stop context-window blow-up. The fourth adds SSE keepalive frames so client streams stop hanging.
What I contributed
- PR 397, merged 2026-04-14: preserves previous_response_id on bridge recovery to stop context blow-up
- PR 448, merged 2026-04-21: trims input when the client supplies previous_response_id
- PR 456, merged 2026-04-22: injects a session-level previous_response_id so input trimming works for every client
- PR 524, merged 2026-05-11: injects SSE keepalive comment frames to stop client stream hangs
