jay-aye-see-kay / chatbot-buffer.nvim Goto Github PK

Adding commands that send a hunk of text for explanation, refactoring, or test writing would be useful. But they'd need a buffer to go to. Probably in the same directory as the conversation buffers.

Unknowns:

• do we have multiple "working buffers", how are the named and found?
• do we make it possible to scope a working buffer to a project (given there's no vim-level definition of a project, it's something most users implement themselves to some extent)
• given the very async nature of ai apis (~10 second responses are normal) do we also exponse a "done" notification somehow?

curl support the server side events features required for streaming tokens, we just need to handle the format and append to the buffer as they come in. This implementation just worked, but it might be ignore some important data coming in

require("plenary.job")
  :new({
    command = "curl",
    args = { url,
      "-H", "Content-Type: application/json",
      "-H", "Authorization: Bearer " .. key,
      "-d", vim.fn.json_encode({
        model = "gpt-3.5-turbo",
        stream = true,
        messages = {
          { role = "user", content = "Hello" },
        },
      }),
    },
    on_stdout = vim.schedule_wrap(function(_, data)
      local maybeJson = data:sub(7)
      if string.len(maybeJson) > 0 then
        local _, update = pcall(vim.fn.json_decode, maybeJson)
        if update ~= nil then
          print(vim.inspect(update.choices[1]["delta"]["content"]))
        end
      end
    end),
  })
  :start()

This requires creating a more thought out file naming scheme, something like YYYY-MM-DD_<num>_<name>.<suffix> where name starts off as "unnamed". Then the name can be set with a function/command. This means we can let the user name a chat file, or turn on auto naming where we ask an ai endpoint to name the chat for us after the nth question / response.

The reason for doing this is to make a list of chat files more meaningful e.g. "I asked about X and Y on Z day - oh there's that code snippets I was looking for".

I'm not sure if I would use the manual naming feature but it's not much extra work to add it in too, and it might be nice sometimes.

The current telescope picker is great for finding a file, but it would be nice to be able to grep the contents of the files too.

I have no idea how this would work but it would be really cool to allow this to run offline and believe it's possible