I have an idea to improve context management. For now if we call ellama-chat with universal argument C-u it will start new session and user can select which model to use in new session. I think we should add user ability to continue current session with current model or create new one with interactively selected model and add some stuff to question context interactively. I think user should be able to add files, buffers and regions into context. Its content should be included in prompt for llm, but should not be showed to user in chat buffer. Instead user should see links to this things only. It will make interactions clean.
I have an idea to improve context management. For now if we call
ellama-chat
with universal argumentC-u
it will start new session and user can select which model to use in new session. I think we should add user ability to continue current session with current model or create new one with interactively selected model and add some stuff to question context interactively. I think user should be able to add files, buffers and regions into context. Its content should be included in prompt for llm, but should not be showed to user in chat buffer. Instead user should see links to this things only. It will make interactions clean.