Online learning to improve code completion experience

[QuanluZhang]

First of all, thanks for your great work! I am using Continue and working on it to try interesting stuff. Recently, interacting with real environment to further improve the intelligence of LLMs becomes more and more feasible, especially using reinforcement learning. The online learning workflow is shown in the following figure. Users interact with the backend LLM to send completion request and receive completion result. After users decide accept or reject. The data is sent to the learner to train the LLM for improving the completion performance.

I studied several popular coding tools and found that Continue is most suitable for this task. It supports integration with any backend model, it has the Autocomplete feature. However, there is no interface/API to expose the request/reponse accept/reject data from the IDE, i.e., the dotted arrow in the following figure. I am wondering does Continue have any plan to support this feature. It is a little like MCP, LLMs can use MCP to interact with tools and environments, while IDE is also a kind of environment that can provide valuable information:)