llama.vim

I use llama.vim - https://github.com/ggml-org/llama.vim I run my own models, you can have models for code suggestion/FIM in the middle and you can have another for chat mode. You can point it to your local model if that's your thing or to a cloud model.

There's also https://github.com/ggml-org/llama.vim Which I've been using with Qwen3 Coder. As long as infill is supported, that should work. I'll try later today.

Have you tried llama.vscode [0]? I use the vim equivalent, llama.vim [1] with Qwen3 Coder 30B and personally feel that it's better than Copilot. I have hot keys that allow me to quickly switch between the two and find myself always going back to local. [0] https://github.com/ggml-org/llama.vscode [1] https://github.com/ggml-org/llama.vim

I like VIM. I like LLMs. I like to work offline. I don't like to pay LLM providers for every line of code I write, and I like to have total control of the context. So, I wrote a plugin for vim that lets me use local LLMs (quantized deepseek coder is surprisingly useful and fast), have total control of the context, and integrate with my normal vimming flow. This one doesn't do tab completion or FIM (see https://github.com/ggml-org/llama.vim if you want a good local solution for that). Instead It's centered around building context through conversation and instruction, having the model answer questions and rewrite whole blocks or files of code. Simple keybindings and auto-copying to the numbered registers makes it easy to integrate with macros and (at least my) normal editing flow.

Avante is much more active and polished at this point, but there's alternatives: https://github.com/baketnk/l.nvim (self-shill, docs overhaul needed) which was inspired by the original: https://github.com/yacineMTB/dingllm.nvim Also brand new from ggerganov: https://github.com/ggml-org/llama.vim dingllm is very straightforward, you submit your entire selection/buffer and it streams out to the current position. mine is a bit more complex with configurations, context management and so on. The thing I always tell people is just roll your own. The docs are there, the LLM is there, use them. At the end of the day it's just an http call against text from your buffer to put text in a buffer.