Support for AI Language Servers (e.g., Copilots) in LSP Server

[bajrangCoder]

This project is fantastic! It is compatible with a ton of coding language servers like rust-analyzer, gopls, and more.

I'm curious, though. Does it also support AI through LSP? For example, GitHub Copilot and Codium work on editors like Vim and Neovim using language servers. I gave it a shot with GitHub Copilot, but it didn't seem to work. Are there any plans to integrate this feature?

In this era of AI, it's really important. It could be incredibly helpful for developers to build an LSP server that works universally across major editors.

Some references:

• https://github.com/Exafunction/codeium.nvim
• https://github.com/Exafunction/codeium-chrome
• https://github.com/github/copilot.vim
• https://github.com/TerminalFi/LSP-copilot?tab=readme-ov-file

[mkslanc]

Hmm, I'll investigate, it shouldn't be anyhow different from other language servers

[bajrangCoder]

Yes , they are similar

[mkslanc]

I couldn't find any standalone language server, everything is just hard-coded plugin/extension for different IDE (which are simply wrapper on API). Do you know any?

[bajrangCoder]

Everything are just compiled binaries, there is no any actual code available for the ai LSP 😔

Btw check out this : https://stackoverflow.com/questions/76741410/how-to-invoke-github-copilot-programmatically

And clients of other editors are compatible such as code mirror, monaco, etc, Why not this one?

Recently Zed code editor also added support for these GitHub copilot through binary LSP server , here is source code

There is one more project similar to this one which is completely* open source, here

[bajrangCoder]

...

[mkslanc]

There were few small bugs in client implementation, new version should support this.

• Installation page
• Full example

I played and it seems a bit bumpy at the moment - server is not supporting abort request (it's on roadmap though), so you need carefully configure liveAutocompletionDelay not to eat all your credits, if you are using OpenAI or Atrophic.
I also tried local llm, but seems my laptop couldn't handle load normally or I haven't configured model normally. It took ~1 minute for response for simple request.

But other than that - it's working, so try to enjoy :)

[bajrangCoder]

I have tried it with groq endpoint and the lsp-ai is providing completion (it can be seen in websocket message tab of inspect tool) but that completion isn't rendered on ace .

I am missing something? 🤔
I have used LSP ai demo from this repo

[mkslanc]

There are annoying bug with showing inner completion first time, try ctrl +space first to manually trigger completion

[bajrangCoder]

Hmm, it's working 👍, (is there no any way to fix that issue)

[mkslanc]

Better to solve it on ace side, the inline autocompletion code is quite confusing