NerdVIM

My Neovim AI Enabled Configuration

UI shots (Without AI Autocomplete enabled.):

NerdVIM with CodeCompanion running Anthropic Claude 3.7 paid model. Picks up Env var for API key automatically. NerdPrompt Perplexity.AI "\np" leader hotkey.

This configuration is focused on productivity, code quality, and a modern Neovim experience, with special attention on development and AI-powered tooling.

Python, Lua, Javascript, Typescript, Bash support. Add your own!

This README describes the main plugins and features included in your Neovim setup, based on your init.lua. It highlights the primary plugins, their purpose, and how they enhance your workflow.

Key Features & Custom Mappings

Leader Key

The leader key is \ (backslash) - Neovim's default. All <leader> keybindings use backslash. Usage: Press \ followed by the key combo (e.g., \t for terminal, \cc for CodeCompanion)

• t: Open a terminal at the bottom split.
• m: Maximize current window.
• np: Open a "nerdprompt" terminal with user input.
• gcp: Add, commit, and push current file with a prompted git commit message while inside a git repo.
• ct: Toggle Claude Code terminal.
• cs: Send visual selection to Claude Code (visual mode).
• ca: Accept Claude Code diff suggestions.
• cr: Reject Claude Code diff suggestions.
• sc: vim-slime sends current cell to tmux top right ipython instance (use '# %%' to mark cell). Makes NVIM more like Jupyter Notebook
• sv: vim-slime sends code in visual mode
• ? : Cheatsheet Holding extensive list of leader commands like there (Update cheatsheet.txt to add more)
• do: Expands gutter message error and warning message.

Main Plugins

AI Web Search

• NerdPrompt
  <leader>np is mapped to trigger Perplexity.AI via NerdPrompt, enabling you to interact with Perplexity.AI from within Neovim. Designed to match your themes and the terminal with ascii art instead of markdown in terminal Note: NerdPrompt must be installed separately.

AI Code Tools

I started with CodeCompanion , but using ClaudeCode now. Leaving CodeCompanion for my local LLM experiments.

• codecompanion.nvim
  Integrates with external AI models (e.g., Anthropic Claude) for code chat and inline assistance directly in Neovim.

• claudecode.vim ClaudeCode integrated with NeoVIM. Agent type work or chat

• minuet-ai.nvim RAG-powered AI code completion using local Ollama models. Provides semantic code completions with full codebase context awareness through VectorCode integration. Uses VectorCode

• vectorcode Semantic code indexing and RAG (Retrieval-Augmented Generation) system using nomic-embed-text embeddings via Ollama. Automatically indexes your codebase and retrieves relevant snippets for AI completions. Uses fill-in-the-middle LLM model

Productivity

• Mini and Rest Break Timers Get reminders to take a microbreak and a restbreak. Currently disabled

• Persistent Undo
  Undo history is saved across sessions for better workflow continuity.

• Clipboard Integration
  Seamless copy-paste between Neovim and your system clipboard.

• Python-centric Indentation
  Tabs and indentation are set to 4 spaces, tailored for Python development.

• Vim-Slime Send code in cell (# %% delimter to mark each cell) to tmux top right pane, use ipython to create jupyter notebook like experience

• Auto-linting and Formatting
  Linting and formatting are triggered automatically on file save and buffer enter.

• Command Cheatsheet
  Press ? to open your custom cheatsheet for quick reference.

Theme

• Dracula is the default color scheme, providing a dark, vibrant look.

UI & Usability

• cheatsheet.nvim
  Provides an in-editor "?" custom command cheatsheet, allowing you to quickly reference and edit custom commands. Just the ones you add to cheatsheet.txt.

• noice.nvim
  Enhances command-line and message UI with popups and a centered command palette for better visibility.

• lualine.nvim
  A fast and customizable statusline, displaying useful information at the bottom of the editor.

• barbar.nvim
  Offers a modern tabline for buffer management, making it easy to switch between open files.

• neo-tree.nvim
  File explorer sidebar for navigating and managing the project tree visually.

• diagflow.nvim
  Displays diagnostics (like LSP, Lint errors and warnings when you hover over squiggles) at the top of the window, right-aligned, for immediate feedback while coding.

LSP, Completion, and Code Intelligence

• nvim-lspconfig
  Simplifies configuring built-in LSP (Language Server Protocol) support for multiple languages, including Python, Lua, Bash, and JavaScript/TypeScript.

• mason.nvim
  Manages external tools (LSP servers, linters, formatters, DAPs) with automatic installation support.

• nvim-treesitter
  Advanced syntax highlighting and code parsing for better code understanding and manipulation.

• none-ls.nvim
  Allows integrating external tools (like linters and formatters) as sources for LSP features.

• nvim-lint
  Asynchronous linting engine, automatically runs linters (like pylint, luacheck, eslint) on save and buffer enter.

• conform.nvim
  Handles code formatting, with support for format-on-save and fallback to LSP formatting.

• blink.cmp
  Modern and fast autocompletion engine, supporting multiple sources (LSP, snippets, buffer, path) and customizable key mappings.

Utilities

• telescope.nvim
  (As a dependency) Powerful fuzzy finder for files, buffers, and more.

• friendly-snippets
  (As a dependency) Collection of community-driven code snippets for use with completion engines.

---

How Intelligent AI Autocomplete using Minuet & VectorCode(ChormaDB RAG) Works

Semantic RAG Flow:

1. VectorCode continuously indexes your codebase using nomic-embed-text embeddings (Ollama runs at startup)
2. When you type, VectorCode retrieves the 3 most semantically relevant code snippets
3. Minuet sends these snippets + your current context to Ollama's qwen2.5-coder model
4. The FIM model generates completions aware of your entire codebase
5. Completions appear in blink.cmp alongside LSP suggestions

Usage

Normal Workflow:

• Install VectorCode via system wide pip, not .venv for the project.
• Just start typing - minuet completions appear automatically in the completion menu
• Press <C-y> to accept (standard blink.cmp)
• Press <A-y> to manually trigger minuet completions

VectorCode Commands:

• :VectorCode register - Manually register current buffer
• :VectorCode deregister - Stop RAG for current buffer
• :VectorCode vectorise - Index entire project
• :VectorCode query <text> - Search semantic context text

You can also run at CLI commands see VectorCode DOCS

• :VectorCode register - Manually register current buffer (auto-enabled)
• :VectorCode deregister - Stop RAG for current buffer
• :VectorCode vectorise - Index entire project
• :VectorCode query <text> - Search semantic context

Per-Project Setup

When you cd into a new codebase:

1. VectorCode will create a .vectorcode/ directory in that project
2. Buffers auto-register on file open
3. Each project maintains its own semantic index
4. RAG context is project-specific

Testing

To test the setup:

1. Reload Neovim (restart nvim or :Lazy sync)
2. Open a Python/Lua/JS file in any project
3. Start typing a function - you should see completions with RAG context
4. Check :messages for VectorCode registration notifications

Troubleshooting

If completions aren't working:

• Ensure Llama is running
• Check VectorCode: vectorcode version (should be 0.7.20)
• Verify registration: Look for VectorCode notifications when opening files
• Check logs: :messages for errors

AI Completion - Alt-y

Alt-y (Option-y on Mac): Manually trigger Minuet AI completion with RAG context

• Queries your codebase semantically via VectorCode
• Sends relevant code snippets to Llama qwen2.5-coder:7B-instruct-Q5_K_M.GGUF model. My research shows this is on of the better model currently.
• Generates context-aware completions using FIM (Fill-In-Middle)
• Ensure Ollama is running: ollama list
• Check VectorCode: vectorcode version (should be 0.7.20)
• Verify registration: Look for VectorCode notifications when opening files
• Check logs: :messages for errors