mozanunal/sllm.nvim

github

ai

102

11

3

7

CREATED

2025-03-15

UPDATED

2026-01-25

---

sllm.nvim

A lightweight Neovim wrapper for Simon Willison's llm CLI. Chat with LLMs, run agentic workflows, and complete code without leaving your editor.

Features

• Streaming chat with markdown rendering and syntax highlighting
• Context management for files, selections, URLs, diagnostics, and shell output
• Agentic mode with Python function tools (bash, read, write, edit, grep)
• Template/mode system for different workflows (chat, review, agent, complete)
• History browsing to continue past conversations
• Slash commands for quick actions (/new, /model, /add-file, etc.)
• Inline completion at cursor position
• Token usage stats and cost tracking in the winbar

Quick start

1. Install the llm CLI:

brew install llm        # macOS
pipx install llm        # or pip install llm

2. Install a provider and set your API key:

llm install llm-openrouter && llm keys set openrouter
# or: llm install llm-anthropic && llm keys set anthropic

3. Install the plugin (lazy.nvim):

{
  'mozanunal/sllm.nvim',
  dependencies = {
    'echasnovski/mini.notify',  -- optional, nicer notifications
    'echasnovski/mini.pick',    -- optional, better picker UI
  },
  config = function()
    require('sllm').setup()
  end,
}

Press <leader>ss to start chatting.

Keymaps

Defaults use the <leader>s prefix (<leader>ss to ask). See doc/configure.md for the full list and how to override.

Modes (templates)

Templates configure the LLM's behavior. Defaults are described in doc/modes.md; switch with <leader>sM or /template.

Slash commands

Press <leader>sx to open the slash command picker, or see doc/slash_commands.md for the full list (e.g., /new, /model, /add-file).

Why sllm.nvim

• Super easy customizable agentic modes — Modes are just YAML files, tools are just Python functions
• Explicit context control — You decide what the LLM sees
• Native llm integration — Uses llm's templates, tools, and history
• Async streaming — Responses stream in real-time with loading indicator
• Token tracking — See usage and cost in the winbar
• Minimal footprint — Follows mini.nvim patterns, thin wrapper over llm

Configuration

require('sllm').setup({
  default_model = 'claude-3-5-sonnet',  -- or 'default' for llm's default
  default_mode = 'sllm_chat',           -- template to use on startup
  window_type = 'float',                -- 'vertical', 'horizontal', or 'float'
  reset_ctx_each_prompt = false,        -- keep context across turns
  keymaps = {
    ask = '<leader>a',                  -- customize keymaps
  },
})

See doc/configure.md for all options.

Documentation

Full guides in doc/:

• README.md — Overview and quick start
• configure.md — Configuration reference
• slash_commands.md — All slash commands
• modes.md — Templates and custom modes
• hooks.md — Pre/post hooks
• backend_llm.md — LLM CLI setup
• api.md — Public Lua API

Or in Neovim: :help sllm

License

Apache 2.0 — see LICENSE.