Auto compress does not trigger when using custom model

[pkradioman]

Is that by-design behavior or any configuration needed?

[factory-ben]

some custom providers don’t return usage stats, so the token-threshold auto-compress may not fire. Manual /compress remains available in the CLI while we address that gap.

[pkradioman]

Would it help if we can set max context window size of each model in the configuration file?

[theoi]

hi @pkradioman - this setting is in the works right now. meanwhile, could you send a session id where auto compress didn't trigger?

[theoi]

(run /status within the session to grab the session id)

[pkradioman]

I have to configure it or work out of the box for custom model?