codex - 💡(How to fix) Fix TUI runs out of context with gpt-5.5 despite auto-compact config [1 comments, 2 participants]

TL;DR

Enable auto-compaction by configuring model_auto_compact_token_limit to prevent the model's context window from being exceeded.

Guidance

• Review the commented-out model_auto_compact_token_limit configuration option and consider setting it to a suitable value (e.g., 300000) to enable auto-compaction.
• Verify that the model_context_window is not set too low, as this could also contribute to the issue, and adjust it if necessary.
• Test the chat functionality with the updated configuration to see if auto-compaction occurs as expected.

Example

No code snippet is provided, as the issue is related to configuration.

Notes

The effectiveness of this workaround depends on the specific requirements of the chat application and the model being used. Further experimentation with different configuration values may be necessary.

Recommendation

Apply workaround: Enable auto-compaction by configuring model_auto_compact_token_limit, as this is a straightforward and targeted solution to the problem described.