All-Hands-AI
Memory Management & Context Condense for CodeAct Agent
What problem or use case are you trying to solve?
Currently, CodeAct is bounded by the context window of a particular LLM. While this provides simplicity, actual long-range tasks require longer interaction (e.g., even more than 500 turns). So, how to effectively condense the context window is the next major point of improvement.
Describe the UX of the solution you'd like
Maintains a fixed context around 32k (which is good enough for most open-source LLMs), and if context go beyond 32k, condense it back to 32k. We probably need to do it a bit conservative: don't condense unless absolutely needed.
The agent should have a way to access their original memory (e.g., by open some files that contains its past interaction in the sandbox).
Do you have thoughts on the technical implementation?
Describe alternatives you've considered
Additional context
@xingyaoww we moved the memory mgmt stuff into a common folder, so you should be able to get this pretty easily!
This issue is stale because it has been open for 30 days with no activity. Remove stale label or comment or this will be closed in 7 days.
![github-actions[bot] avatar](https://avatars.githubusercontent.com/in/15368?v=4 "Published by a pub.dev verified publisher"){kind=small}
This issue is stale because it has been open for 30 days with no activity. Remove stale label or comment or this will be closed in 7 days.
This issue is stale because it has been open for 30 days with no activity. Remove stale label or comment or this will be closed in 7 days.
This issue is stale because it has been open for 30 days with no activity. Remove stale label or comment or this will be closed in 7 days.