Vim AI Assistant (Offline Copilot with Ollama)

A lightweight, offline, privacy-respecting AI coding assistant integrated directly into your terminal Vim.
Powered by Ollama and local models like mistral, llama3, and codellama.
No internet. No telemetry. 100% local.

---

Features

• Trigger the assistant with \a while selecting code in visual mode
• Preview AI-generated code with vimdiff
• Accept or reject changes interactively
• Automatically replaces selected code if approved
• Displays a detailed explanation in a vertical split (formatted as comments)
• Smart comment formatting based on filetype (//, #, etc.)
• Works entirely offline — no OpenAI or cloud APIs required

---

Requirements

• Vim 8+
• Ollama running locally
• Python 3.8+
• Python packages: requests

---

Setup

1. Clone the repository

git clone https://github.com/agace/vim-ai-assistant.git
cd vim-ai-assistant

2. Install dependencies

pip install -r requirements.txt

3. Install and run a local AI model

Make sure Ollama is installed and running. Then install a model (e.g. mistral):

ollama pull mistral
ollama serve

You can replace mistral with any compatible model like llama3, codellama, etc. Just make sure the MODEL variable in assistant.py matches the model you pulled. Ollama currently supports a variety of LLMs, a full list is available here: https://ollama.ai/library

4. Add the Vim integration

cat .vimrc-snippet >> ~/.vimrc

Then reload Vim:

:source ~/.vimrc

How to Use

1. Open a file in Vim
2. Select a block of code in visual mode
3. Press \a
4. Enter a custom instruction (e.g. "simplify this", "optimize performance", "fix and explain")
5. You’ll see:

• A diff preview side-by-side (via vimdiff)
• Press y to accept the changes
• Press n to cancel and leave the code untouched

6. After confirming, an explanation opens in a vertical split with comments

Configuration

• Change the AI model: Edit the MODEL variable in assistant.py to use a different local model (e.g., llama3, codellama).

• Customize the keybinding: The default keybinding is \a in visual mode. You can change it in your .vimrc:

vnoremap <leader>a :<C-U>call AIHelper()<CR>

• Customizing the Prompt: To change how the assistant behaves, edit prompt.txt.
  This is the system message sent to the AI, it controls formatting, tone, and expectations.

• Change the temp file location (optional): You can edit assistant.py or AIHelper() in .vimrc-snippet to change where input/output files are written.

Troubleshooting

• Nothing happens after pressing \a
  Ensure you’re in visual mode and have selected code. Then press \a.

• Error: command not found: ollama
  Ollama must be installed and running. See: https://ollama.com

• Model not found
  Make sure you’ve run ollama pull mistral (or your desired model), and that assistant.py uses the correct name.