Autocompact/compaction occurs despite autocompact.mode=off and remote_compaction=false

Resolved 💬 2 comments Opened Feb 2, 2026 by rodar22 Closed Feb 2, 2026

💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

No description provided.

View original on GitHub ↗

2 Comments

github-actions[bot] contributor · 6 months ago

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

#9419
#9743
#8116
#9601
#10346

Powered by Codex Action

etraut-openai contributor · 6 months ago

There is no supported autocompact config setting, so that explains why that's having no effect.

The remote_compaction feature flag is supported, but that controls whether you are using remote or local compaction. We recommend leaving that feature on.

For a full list of supported config keys, refer to this documentation.

We do support a model_auto_compact_token_limit config key to control the token threshold for compaction, but we don't recommend changing this value because it varies by model.

By default, auto compaction occurs when you use 90% of the available context window. If you're seeing it occur when the UI meter indicates a smaller number (like 40%), that's probably explained by a large tool output that consumed much of the remaining context window. If you suspect that's not the case, please use the /feedback slash command to upload your logs and session and file a new bug report with the thread ID, and I can take a look at it.