Hello everyone,
I am experiencing an issue with Codex where long conversation compaction consistently fails with a 502 error.
When a Codex session becomes large and triggers remote compaction, the process fails with the following error:
502 Bad Gateway
Codex:
{"message":"response protection is unavailable","type":"internal_error"}
After this happens, the session cannot be successfully compacted and I cannot continue working in the same conversation.
Environment:
- Model: gpt-6.1-sol
- Client: Codex (via ACP integration)
- Account: ChatGPT Plus
- Context size before failure: approximately 237k-256k tokens
- Context window: 272,000 tokens
- OS: macOS ARM64
Steps to reproduce:
1. Start a Codex coding session.
2. Continue the conversation until it reaches a large context size.
3. Wait for automatic remote compaction.
4. The compaction request fails with:
response protection is unavailable
Observations:
- New short conversations can compact normally.
- The issue only happens with long-running sessions.
- Retrying the compaction does not resolve the problem.
- The failure happens before a new compacted history is generated.
Questions:
Has anyone encountered the same issue?
Could this be related to:
- Codex remote compaction backend service;
- long-context handling limits;
- response protection service availability;
- ACP integration compatibility?
Any suggestions or workarounds would be appreciated.
Thank you.