Nice project! I was hoping for a tool that does something like that. But how will it deal with the context needed to keep in larger projects? Speed and cost will be an issue too? You could rack a millions of tokens in a workday.

Glad you liked it! Currently, context size is an issue. The user must clear the conversation when the context limit is reached. Knowing that I added basic memory access commands so GPT-4 may store some context between sessions. To reduce the context usage I also defined a system prompt that incentivizes selecting code ranges, not entire files. With this, GPT-4 is able to solve medium-complexity tasks, including implementing commands in this extension itself. For the future I expect GPT-4 with 32k tokens will greatly improve this extension capabilities. Also, I am looking into LangChain to reduce conversation length, define goals and employ a better memory solution.