flyboy

Flyboy is a plugin for lightweight interaction with ChatGPT.

flyboyrecording.mov

It works by using a simple Markdown format, as follows:

# User

Who was the 1st president of the United States?

# Assistant

George Washington

# User

This makes it easy to:

Start chats
Save/share chats
Edit conversations in line
Have multi-turn conversations

No popups that take over your screen, flyboy operates on any buffer.

Flyboy also supports configuring custom templates, so you can go straight from your buffer to ChatGPT with context:

# User

Write a unit test in Lua for the following code
<Your code from visual selection here>

Installation

Put your OPENAI_API_KEY as an environment variable

export OPENAI_API_KEY=""

Have curl installed on your machine
Install plenary.nvim and flyboy using your package manager:

For example, using plug

Plug 'nvim-lua/plenary.nvim'
Plug 'CamdenClark/flyboy'

Usage

:FlyboyOpen functions open a new chat window. Split opens in a horizontal split, while VSplit opens in a vertical split. They optionally take a template.

:FlyboyOpen
:FlyboyOpenSplit
:FlyboyOpenVSplit

" open a chat buffer with the current text selected in visual mode
:FlyboyOpen visual

:FlyboyStart functions open a new chat window and automatically send the message to the assistant. You need to provide a template or the first message sent will be blank.

" starts a chat session with the current text selected in visual mode
:FlyboyStart visual
:FlyboyStartSplit visual
:FlyboyStartVSplit visual

To send a message:

:FlyboySendMessage

The response from the Assistant will be streamed back to the same buffer.

Configuration

Templates

You can configure custom sources and templates for your ChatGPT prompts.

require('flyboy.config').setup({
  sources = {
    my_source = function () return "world" end
  },
  templates = {
    my_template = {
      template_fn = function(sources) return "# User\nHello, " .. sources.my_source() end
      -- :FlyboyOpen my_template
      -- Output:
      -- # User
      -- Hello, world
    }
  }
})

Sources are intended to be helpers to get common pieces of data that you'd be interested in to build your prompts to ChatGPT. Some sources are pre-created, including visual, which provides the text that's visually selected.

Templates are how you construct prompts that will be sent to ChatGPT.

Visual selection

Flyboy supports adding something you've selected in visual mode to the contents of a prompt:

require('flyboy.config').setup({
  templates = {
    unit_test = {
      template_fn = function(sources)
          return "# User\n"
            .. "Write a unit test for the following code:\n"
            .. sources.visual()
      end
      -- :FlyboyStart unit_test
      -- Output:
      -- # User
      -- Write a unit test for the following
      -- <Your visual selection>
    }
  }
})

Buffer selection

Flyboy supports adding the contents of your current buffer to a prompt:

require('flyboy.config').setup({
  templates = {
    unit_test_buffer = {
      template_fn = function(sources)
          return "# User\n"
            .. "Write unit tests for the code in the following file:\n"
            .. sources.buffer()
      end
      -- :FlyboyStart unit_test_buffer
      -- Output:
      -- # User
      -- Write a unit test for the following
      -- <Your previous buffer's contents>
    }
  }
})

Alternative models: gpt-3.5-turbo-16k / gpt-4 / gpt-4-32k

If you want to use Flyboy with a different model in OpenAI, call setup with the model:

require('flyboy.config').setup({
  -- ...
  model = "gpt-4"
})

To change on the fly, call :FlyboySwitchModel gpt-4

Alternative endpoints (Azure OpenAI)

Flyboy supports configuring the URL and headers with a different endpoint that shares API compatibility (IE: Azure OpenAI) with OpenAI, here's a reference implementation:

require('flyboy.config').setup({
  -- should be like "$AZURE_OPENAI_ENDPOINT/openai/deployments/gpt-35-turbo/chat/completions?api-version=2023-07-01-preview"
  url = vim.env.AZURE_OPENAI_GPT4_URL,
  headers = { 
    Api_Key = vim.env.AZURE_OPENAI_GPT4_KEY,
    Content_Type = "application/json"
  }
})

where you put the values for AZURE_OPENAI_GPT4_URL and AZURE_OPENAI_GPT4_KEY in the environment.

If you want to be able to switch URLs based on model, you should make some lua functions in your init.lua that are bound to re-call setup with the updated URL and API key.

Callback when message finished

Flyboy supports configuring a callback function that is called when a response from the assistant finishes streaming.

require('flyboy.config').setup({
  on_complete = function() print("foo") end
})

Development

Run tests

Running tests requires plenary.nvim to be checked out in the parent directory of this repository. You can then run:

just test

or, more verbose:

nvim --headless --noplugin -u tests/minimal.vim -c "PlenaryBustedDirectory tests/ {minimal_init = 'tests/minimal.vim'}"

Or if you want to run a single test file:

just test chat_spec.lua

nvim --headless --noplugin -u tests/minimal.vim -c "PlenaryBustedDirectory tests/path_to_file.lua {minimal_init = 'tests/minimal.vim'}"

Read the nvim-lua-guide for more information on developing neovim plugins.