chat.vim

An asynchronous AI chat interface

Requirements

• vim version 8 or above
  • compatible with neovim version 0.8 and above
• curl
• ollama
  • If using the default settings, llama3.2:latest is required. See the Installation section on how to use models of your choice

Integrations with other AI Chat services can be requested in the Github issues (or feel free to open a pull request).

Usage

Launch a Chat buffer with :Chat

>>> user

Enter your query and press <CR>, and this plugin will asynchronously update the chat buffer with a response.

>>> user
In two sentences, explain what quaternions are

<<< user
>>> assistant
Quaternions are a mathematical system that extends complex numbers to four dimensions, consisting of
one real part and three imaginary parts. They are particularly useful in representing rotations in
three-dimensional space and have applications in computer graphics, robotics, and physics for
efficiently handling orientation calculations without the gimbal lock issue associated with Euler
angles.

<<< assistant
>>> user

All chats are saved locally to g:vim_chat_path (default: ~/.chat.vim) as *.chat.vim.json files. Past chats can be re-opened and continued by using :edit on any of these files.

Installation

Use vim's built-in package support or your favourite package manager.

My suggested method is vim-plug

Plug 'https://github.com/BlakeJC94/vim-chat'

This plugin can be configured using a dictionary in your .vimrc

let g:vim_chat_config = {
\ "default": {
\   "model": "llama3.2:latest",
\   "endpoint_url": "http://localhost:11434/api/chat"
\   },
\ }

Multiple configs can be configured, load a config with :Chat [config-name] (defaults to 'default' if arg is omitted).

If using an endpoint that requires a token for authentication (such as OpenAI), export the authentication token to your environment ($ export TOKEN=<token-value>) and let vim-chat know where to access this key

let g:vim_chat_config = {
\ "default": {
\   ...
\   "token_var": "TOKEN"
\   }
\ }

A system prompt can be specified with the "system_prompt" key:

let g:vim_chat_config = {
\ "default": {
\   ...
\   "system_prompt": "You are a pirate"
\   }
\ }

Multi-line system prompts can be given as a list

let g:vim_chat_config = {
\ "default": {
\    ...
\    "system_prompt": ["You are a pirate", "You give answers in limerick form whenever you can"]
\   }
\ }

Issues

If any errors are encountered (or you would like to make a feature request), raise an issue on the repository so we can discuss. Pull requests are also welcomed

Development

The main branch is reserved for releases and should be considered stable. Changes should occur in the dev branch, which will periodically be merged into main.

Credits

A lot of credit goes to madox2/vim-ai, I highly recommend you check it out if you're interested in AI integrations in vim. I'm open to collaborating to add async support to this plugin as well if there's appetite to do so

Licence

Distributed under the same terms as Vim itself. See :help license.