Acknowledgements

I would like first to thank gera2ld for his work on ai.nvim, because this plugin is a fork of his work. Without his plugin, there wouldnt be this one. Thank you Gerald for your work.

code-ai.nvim

A Neovim plugin powered by Google Gemini and ChatGPT.

Here is a demo without using the agents:

Here is a demo using the agents:

How it works

• The plugin resides in Neovim.
• If the agents URLs are defined, it will send the prompt to the agent, which will send it to the corresponding LLM API.
  • It scans the files corresponding to the patterns defined in .ai-scanned-files and builds a multi-turn chat from them.
  • It sends the multi-turn chat and the prompt to the agents
  • It receives the response from the agents and displays it in a popup
• If the agent URL are not defined, it will send the prompt directly to the corresponding LLM API.
  • It does not scan any files
  • It does not build a multi-turn chat but directly sends the prompt to the LLM API
  • It receives the response from the agents and displays it in a popup

The agent

You can find the agent in the repository code-ai-agent.

Installation

First get API keys from Google Cloud and ChatGPT and set them in your environment:

For usage WITHOUT the agents, don't set the gemini_agent_host and chatgpt_agent_host. For usage WITH the agents, set the gemini_agent_host and chatgpt_agent_host.

This is the configuration for the plugin:

{
  'rakotomandimby/code-ai.nvim',
  dependencies = 'nvim-lua/plenary.nvim',
  opts = {
    gemini_model='gemini-2.0-flash-exp',
    chatgpt_model='gpt-4o-mini',

    gemini_api_key = 'YOUR_GEMINI_API_KEY',         -- or read from env: `os.getenv('GEMINI_API_KEY')`
    chatgpt_api_key = 'YOUR_CHATGPT_API_KEY',       -- or read from env: `os.getenv('CHATGPT_API_KEY')`

    gemini_agent_host='http://172.16.76.1:5000',    -- dont set if you dont want to use the agent
                                                    -- if you set, make sure the agents are running
    chatgpt_agent_host='http://172.16.76.1:4000',   -- dont set if you dont want to use the agent
                                                    -- if you set, make sure the agents are running
    result_popup_gets_focus = true,
    -- Define custom prompts here, see below for more details
    locale = 'en',
    prompts = {
        javascript_vanilla = {
            command = 'AIJavascriptVanilla',
            instruction_tpl = 'Act as a Vanilla Javascript developer. Format you answer with Markdown.',
            prompt_tpl = '${input}',
            result_tpl = '${output}',
            loading_tpl = 'Loading...',
            require_input = true,
        },
        php_bare = {
            command = 'AIPhpBare',
            instruction_tpl = 'Act as a PHP developer. Format you answer with Markdown.',
            prompt_tpl = '${input}',
            result_tpl = '${output}',
            loading_tpl = 'Loading...',
            require_input = true,
        },
    },
  },
  event = 'VimEnter',
},

Usage

If you ave configured the plugin following the instructions above, you can use the plugin by:

• Selecting a text in normal model
• Pressing : to enter the command mode
• Typing AIJavascriptVanilla or AIPhpBare and pressing Enter

Several documentation

The prompts will be merged into built-in prompts. Here are the available fields for each prompt:

Fields	Required	Description
gemini_model	Yes	The model to use for the Gemini API.
chatgpt_model	Yes	The model to use for the ChatGPT API.
gemini_api_key	Yes	The API key for the Gemini API.
chatgpt_api_key	Yes	The API key for the ChatGPT API.
gemini_agent_host	No	The host of the Gemini agent.
chatgpt_agent_host	No	The host of the ChatGPT agent.
command	No	A user command will be created for this prompt.
instruction_tpl	Yes	Template for the instruction given to the model
loading_tpl	No	Template for content shown when communicating with Gemini. See below for available placeholders.
prompt_tpl	Yes	Template for the prompt string passed to Gemini. See below for available placeholders.
result_tpl	No	Template for the result shown in the popup. See below for available placeholders.
require_input	No	If set to true, the prompt will only be sent if text is selected or passed to the command.

Placeholders can be used in templates. If not available, it will be left as is.

Placeholders	Description	Availability
${locale}	opts.locale	Always
${input}	The text selected or passed to the command.	Always
${output}	The result returned by Gemini.	After the request