CodeCompanionChat Close

[bdogabriel]

I don't know if this is a problem only with me, but if I quit nvim and codecompanion is open on a split, when I open the editor again, the buffer and the window are still there, but the chat itself is not functional anymore. So I need to always remember to close codecompanion before leaving nvim, or I'll have to "clean the mess" when I reopen it.

This is a minor annoyance, but could be easily solved with a Close command. But since Close does not exist, I just use a floating window because it is on my face and there is no way I will forget to close it before leaving the editor.

When I used CopilotChat, I had an autocmd that closes it on "QuitPre", and it solved this exact same problem. But I cannot achieve this with codecompanion using Toggle with an autocmd: if the chat is not open when I close nvim, it will open one, and then close, producing the same result.

I couldn't find a workaround for this on the discussions/docs/issues, and I cannot figure it out myself due to skill issues. I know just enough of lua and nvim to get a decent config working, I am still learning. But even if there is a workaround, I think it wouldn't be a bad Idea to have a separate Close command (maybe a Open too), because it would enable more customization.

I know there is no bug, and this is normal behavior. But it's just my motivation and example to show that adding a Close command could be beneficial to cover use cases which Toggle can't.

I hope this can be interesting for someone else :)

[olimorris]

function()
  -- Ensure no CodeCompanion buffers are saved into the session
  for _, buf in ipairs(vim.api.nvim_list_bufs()) do
    if vim.bo[buf].filetype == "codecompanion" then
      vim.api.nvim_buf_delete(buf, { force = true })
    end
  end
end,

FYI, see chat.new and create_buf for why this exists

[bdogabriel]

Thank you for responding! That works like a charm. I feel kind of dumb for not being able to figure it out myself, lol

[olimorris]

The internal Neovim APIs are huge. One workflow I like to employ is to use the /help slash commands to send certain pages of the help docs to my LLM and let that either figure it out or teach me.