So… Why exactly?⌗

LLM-powered coding suggestions have been a hot topic since they were first introduced to the mainstream development industry with the launch of GitHub’s Copilot program. Some developers swear by their productivity gains (and some stats show it is helpful), while droves of other developers have found themselves turning off LLM-powered suggestions for a plethora of reasons: too distracting, too much reliability on the suggestions, less engaging work, and honestly much more.

I don’t think anyone is wrong here. I think AI pair programming is likely going to become the future of an ideal development environment, but… The way we currently deliver AI-powered suggestions just won’t cut it in the long-term IMO.

So, here is my rather basic premise: use a completion engine, wire up LSPs & your LLM of choice to that completion engine, and then manage suggestions through a singular interface.

Note:

This flow should work just fine for Copilot or Supermaven. Just ensure you include copilot-cmp and copilot plugins.

Neovim Plugins Required⌗

  • nvim-cmp
    • A LUA-based neovim completion engine, which will be used to manage our LLM suggestions (supermaven or copilot) and the builtin Neovim LSP
  • nvim-lspconfig
    • An official neovim community plugin which handles defaults of MANY lsp configs
  • cmp-nvim-lsp
    • Provides the "nvim_lsp" source, enabling support for use with nvim-cmp
  • supermaven-inc/supermaven-nvim
    • Objectively better, and faster AI-powered coding suggestions, with builtin support for nvim-cmp

Neovim LUA Configuration⌗

require'supermaven-nvim'.setup({
  keymaps = {
    accept_suggestion = "",
    clear_suggestion = "",
    accept_word = ""
  },
  disable_inline_completion = true
})

local cmp = require'cmp'
cmp.setup({
  sources = {
    {name = "supermaven"},
    {name = "nvim_lsp"}
  },
  map = cmp.mapping.preset.insert({
    # control + space to accept the top suggestion
    ['<C-Space>'] = cmp.mapping.confirm({select = true}),
  })
})