Support configuring remote_compaction per model

Open 💬 0 comments Opened Aug 5, 2026 by yiteng-guo

What variant of Codex are you using?

CLI

What feature would you like to see?

#36981 added v1 remote compaction support to amazon-bedrock providers. I think "remote compaction capability" is more a model capability instead of provider.

My use case is that we use a LLM gateway bifrost that supports both azure (slug gpt-5.6-sol) and bedrock (slug openai.gpt-5.6-sol). Currently we configure codex with a custom provider bifrost so codex won't use remote compaction even if azure supports it.

Ideally, I can have a custom model catalog to set remote compaction v2 for gpt-5.6-sol (which will also be the value in bundled catalog json) while v1 for openai.gpt-5.6-sol (a custom entry I added).

Additional information

_No response_

View original on GitHub ↗