[Bug]: unexpected status 502 Bad Gateway: Upstream request failed, url: https://xxxxxxxx, request id: xxxxxx

[neohob]

Describe the bug

After several hours of heavy use, one session hit 64,408.5 k tokens and then became unusable—constant 502 errors. Only starting a fresh session lets me continue; it feels impossible to use in depth, and context handling is poor. I can’t tell whether the issue lies with Codex itself; I’ve already switched LLM endpoints twice.

Steps to reproduce

1. Session length reached 64,408.5 k tokens, and the time exceeded 400 minutes.
2. Retrying keeps returning 502.

Expected behavior

Hope it runs normally

Screenshots or logs

No response

Operating System

macos 15.7.3

OpenCow Version

0.3.9

Node.js Version

22.11.0

AI Engine

OpenAI Codex

[HikoQiu]

Thanks for reporting this, @neohob .

The 502 Bad Gateway is most likely coming from the upstream model API rather
than OpenCow itself. To help us narrow this down — could you confirm whether
you're using the official OpenAI API directly, or a third-party
aggregation/proxy service (e.g. OpenRouter, one-api, etc.)?

Third-party proxies often impose their own rate limits and context size
restrictions, which can cause 502s at high token counts like the 64,408.5 k tokens you
mentioned.

[miguelmanlyx]

Quick note, this points to a ai api timeout problem. Might be worth trying npx ai-doctor. It can fix missing timeout guards with safe request timeouts plus timeout and retry protection.

[neohob]

Honestly, I really like the concept behind your product—think Accio Work—but I just can’t keep using it. This is the second time: after three hours it’s been stuck for two and a half, with no feedback, forever “reasoning…” @HikoQiu