llama.vim

Why we included this project

llama.vim wires Vim and Neovim directly to a local llama.cpp server, so code completion runs entirely on your own machine. In Insert mode it offers fill-in-the-middle suggestions you can accept with Tab, and you can select text and ask the model to rewrite or continue it. The context handling is what makes it practical: a ring buffer collects chunks from open files, edited files, and yanked text, then reuses that context so large codebases stay responsive even on low-end hardware. Configuration comes down to a g:llama_config variable and remappable keymaps, and the plugin is opinionated about pairing with a FIM-compatible model from its recommended collection. That model requirement is the main thing to plan for, but for teams that want Copilot-style completion without the cloud, it's a lightweight way to get it inside the editor they already use.