Configurable auto compact threshold?

[stevoland]

Hi, I use micode with github-copilot provider that has low caps for models, around ~128k tokens.

Would you consider a PR to make this configurable?

thanks

[vtemian]

sure thing

[vtemian]

Fixed in v0.9.0! 🎉

You can now configure the compaction threshold in ~/.config/opencode/micode.json:

{
  "compactionThreshold": 0.3
}

The threshold is a ratio (0-1) of the context window. Default is 0.5 (50%).

Additionally, model context limits are now automatically read from your opencode.json provider configs (the limit.context field), so if you define your github-copilot provider with the correct context limit, it will be used automatically.

Release: https://github.com/vtemian/micode/releases/tag/v0.9.0