TRIAL aka STATUS "NOT YET DONE" Install OFFLINE Code Assistant by NEOZIM

Building a “Gold Standard” Local AI Environment
Neovim + Ollama + CodeCompanion.nvim

⬢ Zero telemetry · ⬢ Full local control · ⬢ Diff‑based workflow

Since you’re looking to build a professional, extension‑free AI environment, Neovim is the right choice. Unlike VS Code—where the AI is a “black box” extension—in Neovim, you are the architect. This guide walks you through a production‑ready setup using Ollama and CodeCompanion.nvim, with zero telemetry, full local control, and a diff‑based workflow.

1 Prerequisites

Ensure these tools are installed and functional on your system:

Tool	Version	Purpose
Neovim	v0.10.0+	The editor engine
Git	Latest	Plugin and dependency management
Ollama	Latest	Local LLM server
C Compiler	gcc or clang	Builds native UI components

Verify Ollama is running:
ollama serve &> /dev/null &  ·  ollama pull llama3.1 # or your preferred model

2 Bootstrap the Plugin Manager (Lazy.nvim)

Create your Neovim config file if it doesn't exist:

mkdir -p ~/.config/nvim
touch ~/.config/nvim/init.lua

Place this bootstrap code at the top of init.lua — it installs Lazy.nvim automatically on first run:

-- Bootstrap Lazy.nvim
local lazypath = vim.fn.stdpath("data") .. "/lazy/lazy.nvim"
if not vim.loop.fs_stat(lazypath) then
  vim.fn.system({
    "git", "clone", "--filter=blob:none",
    "https://github.com/folke/lazy.nvim.git", "--branch=stable", lazypath
  })
end
vim.opt.rtp:prepend(lazypath)

3 Declare Your Plugins

Pass your plugin specification to require("lazy").setup(). We include CodeCompanion.nvim and its core dependencies:

require("lazy").setup({
  {
    "olimorris/codecompanion.nvim",
    dependencies = {
      "nvim-lua/plenary.nvim", -- Lua functions library
      "nvim-treesitter/nvim-treesitter", -- Syntax awareness
    },
    config = function()
      -- (configuration added in Step 4)
      require("codecompanion").setup({
        -- configuration goes here
      })
    end,
  },
})

4 Configure CodeCompanion for Ollama

Inside the config function from Step 3, tell CodeCompanion to use your local Ollama server — and define a system prompt to enforce a consistent technical persona.

require("codecompanion").setup({
  strategies = {
    chat = { adapter = "ollama" },
    inline = { adapter = "ollama" },
  },
  adapters = {
    ollama = function()
      return require("codecompanion.adapters").extend("ollama", {
        env = {
          url = "http://127.0.0.1:11434", -- Local endpoint
        },
        schema = {
          model = {
            default = "llama3.1", -- Must match `ollama list`
          },
        },
        -- 🔥 PRO TIP: System prompt to enforce your technical persona
        parameters = {
          system_prompt = [[
  You are a senior software architect and automation expert.
  Reply with clarity, precision, and actionable code examples.
  Avoid fluff—focus on maintainability, performance, and correctness.
  ]],
        },
      })
    end,
  },
})

Important: Make sure the model.default name matches exactly what ollama list shows.

5 Restart & Let Lazy Install

• Restart Neovim (nvim +Lazy sync also works).
• Lazy.nvim will download all plugins automatically.
• Once done, you're ready to use the AI.

6 Workflow – How to Use It

Command / Action	Description
:CodeCompanionChat	Opens a dedicated chat buffer where you can converse with the AI about your codebase.
Visual Mode + :CodeCompanion	Select a code block, then run this command. Type e.g., "refactor this" or "add error handling". The plugin shows a diff preview — press Accept or Reject.
Inline Edits	Inside the chat, ask the AI to "apply this to the file" — it stages changes in a reviewable diff format before you save.

All context (current file, selection, etc.) is automatically passed to the AI — no manual curl required.

Why This Setup Satisfies Your Requirements

Requirement	How It's Met
Zero Hardcoding	The plugin handles the JSON‑RPC protocol and context assembly automatically.
No Telemetry	Forcing the adapter to 127.0.0.1:11434 ensures zero bytes leave your machine.
Professional Standard	Diff‑view workflows are identical to those in paid VS Code extensions — but you control the entire stack.
Customizable Persona	The system_prompt (Step 4) lets you steer the AI's tone and depth, perfect for consistent blog‑ready output on ronin.directory.

Bonus: Advanced System Prompt Tuning

To further refine the AI's behavior, you can expand the system_prompt with:

• Style directives: "Use declarative sentences. Prefer pure functions over classes."
• Domain constraints: "Assume a Linux environment. Optimize for ARM64."
• Output structure: "Always include a 'Why this works' section in your answers."

Simply update the parameters.system_prompt string and restart Neovim — no plugin reinstall needed.

Final Check

After setup, run :checkhealth in Neovim to verify all dependencies are ready. Then open a file, select some code, and test :CodeCompanion — you should see the AI respond using your local Ollama model, respecting your system prompt.

⚡ You're now a Neovim AI architect.
Happy coding — and may your diffs always be clean. 🚀

Originally written for ronin.directory · Reproduced with permission