AI Assistant

[FossPrime]

Hot take: it would be easier and better to support a low weight gecko style coding LLM than to deal with general extension support, to then get Intellisense for Web, which is very limited even in github.dev, it's a little better in vscode.dev

This reduces the need for everything else, including the TS Language server, volar, Intellisense light... Etc.

It would run as a worker providing suggestions, I suggest davinci or gecko. We would stand out as the only ones doing that.

https://replit.com/@fossprime/feathers-llm

[TheUltDev]

Yeah this would be a must eventually, at the very least support integration with a self hosted option.
I'm not really well versed in this area but here's some solutions I've found.

Resources

StarCoder

• https://huggingface.co/blog/starcoder
• https://marketplace.visualstudio.com/items?itemName=HuggingFace.huggingface-vscode
• https://github.com/LucienShui/huggingface-vscode-endpoint-server

CodeGen

• https://github.com/fauxpilot/fauxpilot (uses CodeGen, w/ CodeGen2 coming soon)
• https://github.com/salesforce/CodeT5 (newest model, no known implementation)

Other implementations

• https://github.com/TabbyML/tabby