Sllm.nvim v0.4.0: Modes as YAML, Radical Simplicity

I’m excited to announce sllm.nvim v0.4.0 – a significant release that doubles down on the plugin’s core philosophy: modes are just YAML, tools are just Python functions, and you stay in control.

The Philosophy Recap

For those new to sllm.nvim, it’s a thin Neovim wrapper around Simon Willison’s llm CLI. The design philosophy is simple:

• You control context – Add files, selections, URLs, diagnostics, or shell output explicitly. The LLM sees exactly what you give it.
• Radical simplicity – The plugin delegates all heavy lifting to the battle-tested llm CLI (~500 lines of Lua).
• CLI ecosystem access – 100+ models via OpenRouter, local models via Ollama, plus plugins for PDFs, web search, and more.

What’s New in v0.4.0

Modes Are Just YAML

The template system has been refined and expanded. Creating a custom mode is now as simple as writing a 5-line YAML file:

# ~/.config/io.datasette.llm/templates/my_reviewer.yaml
system: |
  You are a code reviewer. Be concise and actionable.
  Focus on bugs, not style.

The plugin ships with four default modes:

• sllm_chat – General-purpose conversations
• sllm_read – Explore codebases with read-only tools (list, read, grep, glob)
• sllm_agent – Full agentic mode with bash/read/write/edit/grep tools
• sllm_complete – Inline completion at cursor position

Switch modes with <leader>sM or /template. Edit them with llm templates edit sllm_agent.

On-the-fly Tools Refined

The on-the-fly function registration (/add-function or <leader>sx → “add-function”) continues to be a standout feature. Select any Python function in your buffer and the LLM can use it immediately:

def fetch_user_data(user_id: str) -> str:
    """Fetch user data from the database.

    Args:
        user_id: The user's ID
    """
    # Your implementation here
    return json.dumps(db.get_user(user_id))

Select it, run /add-function, and the LLM can now query your database mid-conversation. No configuration, no restart needed.

Slash Commands System

A new unified command system replaces scattered keymaps with discoverable actions:

• <leader>sx opens a fuzzy command picker
• Type /command directly in the prompt (e.g., /new, /model, /add-file)

Commands are organized by purpose:

• Chat: /new, /history, /cancel
• Context: /add-file, /add-url, /add-selection, /add-diagnostics, /add-output, /add-tool, /add-function, /clear-context
• Model: /model, /template, /online, /system
• Options: /options, /set-option, /reset-options
• Copy: /copy-code, /copy-code-first, /copy-response

Inline Completion

<leader><Tab> triggers inline completion at your cursor position. The sllm_complete template sends buffer context and returns raw code suitable for direct insertion – no markdown wrapping, just completions that fit.

Token Tracking in Winbar

The LLM buffer’s winbar now displays token usage stats and estimated costs, giving you visibility into your API consumption at a glance.

Pre/Post Hooks

Automate context gathering with hooks:

require('sllm').setup({
  pre_hooks = {
    { command = 'git diff --cached', add_to_context = true },
  },
  post_hooks = {
    { command = 'notify-send "sllm.nvim" "Response ready"' },
  },
})

Now every prompt automatically includes your staged changes.

Quick Start

# Install llm and a provider
brew install llm && llm install llm-openrouter && llm keys set openrouter

-- lazy.nvim
{
  'mozanunal/sllm.nvim',
  config = function()
    require('sllm').setup()
  end,
}

Press <leader>ss to start chatting.

Documentation

• Modes & Templates
• Slash Commands
• Configuration
• Hooks

Try It Out

If this CLI-first, co-pilot-centric approach resonates with you, give sllm.nvim a try: mozanunal/sllm.nvim

Feedback, issues, and contributions are always welcome!