vim9-ollama

Local AI coding assistant for Vim 9 powered by Ollama.

vim9-ollama is a modern Vim9script plugin that integrates local LLMs into Vim.
It provides AI-powered chat, code rewriting, code review, and inline code completion.

All AI inference runs locally using Ollama, so your code never leaves your machine.

---

Features

• 🧠 Local AI coding assistant
• ⚡ Written in modern Vim9script
• 🔌 Autoload architecture for fast startup
• 💬 Chat with LLM about your code
• ✏️ Rewrite / refactor selected code
• 🔍 Code review and suggestions
• 🤖 Inline AI completion (experimental)
• ⚙️ Fully configurable models and prompts

---

Requirements

• Vim ≥ 9.0
• Ollama

Install Ollama:

curl -fsSL https://ollama.com/install.sh | sh

Start the Ollama server:

ollama serve

Download recommended models:

ollama pull starcoder2:3b
ollama pull qwen2.5-coder:3b

---

Installation

Using vim-plug:

Plug 'greeschenko/vim9-ollama'

Restart Vim.

The plugin will automatically start the Ollama server on Vim startup.

---

Configuration

The plugin can be configured via global Vim variables.

Example configuration:

let g:ollama_api = "http://localhost:11434/api/generate"

let g:ollama_models = {
\ "complete": {
\   "name": "starcoder2:3b",
\   "stream": v:false,
\   "options": {
\     "num_predict": 256,
\     "temperature": 0.2
\   },
\   "prompt_template": [
\     "```{filetype}",
\     "{filecontext}",
\     "{input}<|cursor|>{instruction}"
\   ]
\ },
\
\ "change": {
\   "name": "qwen2.5-coder:3b",
\   "stream": v:true,
\   "options": {
\     "temperature": 0.1,
\     "num_predict": 512
\   }
\ },
\
\ "chat": {
\   "name": "qwen2.5-coder:3b",
\   "stream": v:true,
\   "options": {
\     "temperature": 0.3
\   }
\ }
\ }

You can customize:

• model names
• temperature
• token limits
• prompt templates

---

Commands

Ask the LLM

Ask a question about the current file or code context.

:OllamaAsk Explain this function

The response will appear in a split window.

---

Rewrite selected code

Select code in visual mode and run:

:OllamaChange Refactor this function

Examples:

:OllamaChange Improve readability

:OllamaChange Add logging

:OllamaChange Convert this to Go generics

The selected code will be replaced with the modified version.

---

Review selected code

Select code and run:

:OllamaRead What could be improved here?

This displays AI comments in a separate buffer.

---

Inline Code Completion (Experimental)

Inline completion works similarly to GitHub Copilot.

Trigger completion

Press:

Ctrl + L

This sends the current context to the model and shows a ghost suggestion below the cursor.

Press again to request another suggestion.

---

Accept completion

Press:

Ctrl + F

The suggested code will be inserted into the buffer.

---

How Completion Works

The plugin sends a Fill-in-the-Middle (FIM) prompt to the model.

Example input:

fmt.Printf(<|cursor|>"hello")

The model generates the missing code at the cursor position.

---

Architecture

The plugin uses modern Vim9 autoload architecture.

vim9-ollama
├─ plugin
│  └─ vim9ollama.vim
│
└─ autoload
   └─ vim9ollama.vim

• plugin/ defines commands and key mappings
• autoload/ contains implementation
• functions load only when used

This keeps Vim startup fast.

---

Troubleshooting

Ollama does not respond

Restart the server:

:call ollama.StartServer()

Or manually:

pkill ollama
ollama serve

---

Roadmap

Planned improvements:

• streaming inline completion
• better FIM prompts
• async UI updates
• multi-buffer context
• configurable keymaps
• model selection per filetype

---

License

MIT

---

Author

Olex Hryshchenko https://github.com/greeschenko