What future improvements can we expect for handling context overflow?

[JerryKwan]

What future improvements can we expect for handling context overflow?
Looking at the current implementation in agent.py:_agent_loop() it just tries to do a tape handoff when context overflow occurs. This feels a bit too abrupt and ends up losing some of the conversation's background info. Do we have any plans to optimize this later? Like using summarization, for instance?

[frostming]

The current implementation can only serve as a fallback approach. However, you can utilize bub's hooks to implement a context processing mechanism, such as automatic summarization and compaction, but as a plugin, not into core. Related hooks are provide_tape_store and build_tape_context

[JerryKwan]

Thanks for the information.
One more question, are there any sample plugins for reference? Just like the ones shared in https://github.com/bubbuild/bub-contrib ? thank you.

[frostming]

No context related plugins for now, while you can look at other plugins in that repo for reference