As developers, we all know the pain of context-switching between our code editor and separate AI interfaces just to ask a question, get code completion, or analyze an error message. What if you could bring the power of Large Language Models—powered by the incredibly flexible llm CLI by Simon Willison—directly inside Neovim, with streaming replies, model/tool selection, and one-key context management?

Today, I’m excited to announce sllm.nvim: a (roughly 500-line) Lua plugin that brings deep llm CLI integration to Neovim in an asynchronous, focused, and highly hackable way.

Why sllm.nvim?

llm is a CLI Swiss Army knife for working with LLMs, built by Simon Willison (creator of Django, Datasette, and sqlite-utils). It supports hundreds of models (via plugins like llm-openrouter), files, tools, and arbitrary context tricks.

But until now, integrating llm CLI into your real coding environment required a lot of copy-paste—and breaking flow by toggling terminals, browsers, and buffers.

I built sllm.nvim so you can:

  • Chat with models in a streaming popup buffer (no annoying freezes or waiting)
  • Add files, URLs, shell output, code selections, diagnostics—and even tools!—to the LLM context… in one keystroke
  • Switch models and tools on the fly
  • Get token usage/cost feedback and never overrun your limits

…all without leaving Neovim. No more copy-paste error traces. No more browser tab juggling. Just ask, code, and iterate, staying in the flow.

Key Features

  • Interactive Chat: Ask questions and see responses stream directly into a live, markdown-formatted Neovim buffer.
  • Context Management: In one keystroke, you can add the current file, an arbitrary URL, visual selections, diagnostics, shell command outputs, or LLM tools to be considered in your next prompt.
  • Model & Tool Selection: Instantly pick from any installed llm model/plugins/tools and add them to context.
  • Async, Non-blocking: Let LLM jobs run in the background—you keep editing or exploring.
  • Usage Feedback: Optionally show tokens/cost for every request.
  • UI via mini.nvim: Uses proven modules from mini.nvim for notifications and pickers—smooth and minimal.

Example Workflow

  • Open a file, highlight some lines, hit <leader>sv to add that code as context.
  • Press <leader>ss and enter your question.
  • Try <leader>sm to switch models, e.g. between GPT-4 and local models.
  • Add a shell command’s output to context with <leader>sx.
  • Add info on a bug by pressing <leader>sd (diagnostics).
  • Need a third-party API or tool integration for your prompt? <leader>se lets you add LLM tools directly.
  • Review token/cost estimates with every response.

All responses stream into a markdown buffer—ready to reference, copy, or save.

Want to Try It?

Check it out here: https://github.com/mozanunal/sllm.nvim

I’d love to hear feedback, issues, and wishes. PRs and suggestions are very welcome!

Credits

Enjoy chatting and coding—without ever leaving Neovim!