parrot.nvim 🦜

This is parrot.nvim, the ultimate stochastic parrot to support your text editing inside Neovim.

Features • Demo • Getting Started • Commands • Configuration • Roadmap • FAQ

[!NOTE] ⚠️ This repository is still a work in progress, as large parts of the code are still being simplified and restructured. It is based on the brilliant work gp.nvim by https://github.com/Robitx.

I started this repository because a perplexity subscription provides $5 of API credits every month for free. Instead of letting them go to waste, I modified my favorite GPT plugin, gp.nvim, to meet my needs - a new Neovim plugin was born! 🔥

Unlike gp.nvim, parrot.nvim prioritizes a seamless out-of-the-box experience by simplifying functionality and focusing solely on text generation, excluding the integration of DALL-E and Whisper.

Features

• Persistent conversations as markdown files stored within the Neovim standard path or a user-defined location
• Custom hooks for inline text editing and to start chats with predefined prompts
• Support for multiple providers:
  • Anthropic API
  • perplexity.ai API
  • OpenAI API
  • Mistral API
  • Gemini API
  • Local and offline serving via ollama
• Custom agent definitions to determine specific prompt and API parameter combinations, similar to GPTs
• Flexible support for providing API credentials from various sources, such as environment variables, bash commands, and your favorite password manager CLI (lazy evaluation)
• Provide repository-specific instructions with a .parrot.md file with the command PrtContext
• No autocompletion and no hidden requests in the background to analyze your files

Demo

Seamlessly switch between providers and agents.

---

Trigger code completions based on comments.

---

Let the parrot fix your bugs.

---

Rewrite a visual selection with `PrtRewrite`.

---

Append code with the visual selection as context with `PrtAppend`.

---

Add comments to a function with `PrtPrepend`.

Getting Started

Dependencies

• neovim
• fzf
• plenary
• ripgrep

lazy.nvim

{
  "frankroeder/parrot.nvim",
  tag = "v0.3.9",
  dependencies = { 'ibhagwan/fzf-lua', 'nvim-lua/plenary.nvim' },
  -- optionally include "rcarriga/nvim-notify" for beautiful notifications
  config = function()
    require("parrot").setup {
      -- Providers must be explicitly added to make them available.
      providers = {
        pplx = {
          api_key = os.getenv "PERPLEXITY_API_KEY",
          -- OPTIONAL
          -- gpg command
          -- api_key = { "gpg", "--decrypt", vim.fn.expand("$HOME") .. "/pplx_api_key.txt.gpg"  },
          -- macOS security tool
          -- api_key = { "/usr/bin/security", "find-generic-password", "-s pplx-api-key", "-w" },
        },
        openai = {
          api_key = os.getenv "OPENAI_API_KEY",
        },
        anthropic = {
          api_key = os.getenv "ANTHROPIC_API_KEY",
        },
        mistral = {
          api_key = os.getenv "MISTRAL_API_KEY",
        },
        gemini = {
          api_key = os.getenv "GEMINI_API_KEY",
        },
        ollama = {} -- provide an empty list to make provider available
      },
    }
  end,
}

Commands

Below are the available commands that can be configured as keybindings. These commands are included in the default setup. Additional useful commands are implemented through hooks (see my example configuration).

General

Command	Description
PrtChatNew <target>	open a new chat
PrtChatToggle <target>	toggle chat (open last chat or new one)
PrtChatPaste <target>	paste visual selection into the latest chat
PrtInfo	print plugin config
PrtContext <target>	edits the local context file
PrtChatFinder	fuzzy search chat files using fzf
PrtChatDelete	delete the current chat file
PrtChatRespond	trigger chat respond (in chat file)
PrtStop	interrupt ongoing respond
PrtProvider <provider>	switch the provider (empty arg triggers fzf)
PrtAgent <agent>	switch the agent (empty arg triggers fzf)
Interactive
PrtRewrite	Rewrites the visual selection based on a provided prompt
PrtAppend	Append text to the visual selection based on a provided prompt
PrtPrepend	Prepend text to the visual selection based on a provided prompt
PrtNew	Prompt the agent to respond in new window
PrtEnew	Prompt the agent to respond in a new buffer
PrtVnew	Prompt the agent to respond in a vsplit
PrtTabnew	Prompt the agent to respond in a new tab
Example Hooks
PrtImplement	implements/translates the visual selection comment into code
PrtAsk	Ask the agent a question

With <target>, we indicate the command to open the chat within one of the following target locations (defaults to toggle_target):

• popup: open a popup window which can be configured via the options provided below
• split: open the chat in a horizontal split
• vsplit: open the chat in a vertical split
• tabnew: open the chat in a new tab

All chat commands (PrtChatNew, PrtChatToggle) and custom hooks support the visual selection to appear in the chat when triggered. Interactive commands require the user to make use of the template placeholders to consider a visual selection within an API request.

Configuration

Options

{
    -- The provider definitions with endpoints, api keys and models used for chat summarization
    providers = ...

    -- the prefix used for all commands
    cmd_prefix = "Prt",

    -- optional parameters for curl
    curl_params = {},

    -- The directory to store persisted state information like the
    -- current provider and the selected agents
    state_dir = vim.fn.stdpath("data"):gsub("/$", "") .. "/parrot/persisted",

    -- Defintion of the agents (similar to GPTs) for the chats and the inline hooks
    agents = {
        chat = ...,
        command = ...,
    },

    -- The directory to store the chats (searched with PrtChatFinder)
    chat_dir = vim.fn.stdpath("data"):gsub("/$", "") .. "/parrot/chats",

    -- Chat user prompt prefix
    chat_user_prefix = "🗨:",

    -- Explicitly confirm deletion of a chat file
    chat_confirm_delete = true,

    -- Local chat buffer shortcuts
    chat_shortcut_respond = { modes = { "n", "i", "v", "x" }, shortcut = "<C-g><C-g>" },
    chat_shortcut_delete = { modes = { "n", "i", "v", "x" }, shortcut = "<C-g>d" },
    chat_shortcut_stop = { modes = { "n", "i", "v", "x" }, shortcut = "<C-g>s" },
    chat_shortcut_new = { modes = { "n", "i", "v", "x" }, shortcut = "<C-g>c" },

    -- Option to move the chat to the end of the file after finished respond
    chat_free_cursor = false,

     -- use prompt buftype for chats (:h prompt-buffer)
    chat_prompt_buf_type = false,

    -- Default target for  PrtChatToggle, PrtChatNew, PrtContext and the chats opened from the ChatFinder
    -- values: popup / split / vsplit / tabnew
    toggle_target = "vsplit",

    -- The interactive user input appearing when can be "native" for
    -- vim.ui.input or "buffer" to query the input within a native nvim buffer
    -- (see video demonstrations below)
    user_input_ui = "native",

    -- Popup window layout
    -- border: "single", "double", "rounded", "solid", "shadow", "none"
    style_popup_border = "single",

    -- margins are number of characters or lines
    style_popup_margin_bottom = 8,
    style_popup_margin_left = 1,
    style_popup_margin_right = 2,
    style_popup_margin_top = 2,
    style_popup_max_width = 160

    -- Prompt used for interactive LLM calls like PrtRewrite where {{agent}} is
    -- a placeholder for the agent name
    command_prompt_prefix_template = "🤖 {{agent}} ~ ",

    -- auto select command response (easier chaining of commands)
    -- if false it also frees up the buffer cursor for further editing elsewhere
    command_auto_select_response = true,

    -- fzf_lua options for PrtAgent and PrtChatFinder when plugin is installed
    fzf_lua_opts = {
        ["--ansi"] = true,
        ["--sort"] = "",
        ["--info"] = "inline",
        ["--layout"] = "reverse",
        ["--preview-window"] = "nohidden:right:75%",
    },

    -- Enables the query spinner animation 
    enable_spinner = true,
    -- Type of spinner animation to display while loading
    -- Available options: "dots", "line", "star", "bouncing_bar", "bouncing_ball"
    spinner_type = "star",
}

Demonstrations

With `user_input_ui = "native"`, use `vim.ui.input` as slim input interface.

With `user_input_ui = "buffer"`, your input is simply a buffer. All of the content is passed to the API when closed.

The spinner is a useful indicator for providers that take longer to respond.

Key Bindings

This plugin provides the following default key mappings:

Keymap	Description
<C-g>c	Opens a new chat via PrtChatNew
<C-g><C-g>	Trigger the API to generate a response via PrtChatRespond
<C-g>s	Stop the current text generation via PrtStop
<C-g>d	Delete the current chat file via PrtChatDelete

Refer to my personal lazy.nvim setup for further hooks and key bindings: https://github.com/frankroeder/dotfiles/blob/master/nvim/lua/plugins/parrot.lua

Adding a new agents

We provide two types of agents that might need different system prompts and API parameters. To make a new chat agent available, one simply adds a new entry to the list chat or to command, respectively.

require("parrot").setup {
    -- ...
    agents = {
        chat = {
          {
              name = "CodeLlama",
              model = { model = "codellama", temperature = 1.5, top_p = 1, num_ctx = 8192, min_p = 0.05 },
              system_prompt = "Help me!",
              provider = "ollama",
          }
        }
    },
    -- ...
}

Adding a new command

Ask a single-turn question and receive the answer in a popup window

require("parrot").setup {
    -- ...
    hooks = {
      Ask = function(parrot, params)
        local template = [[
          In light of your existing knowledge base, please generate a response that
          is succinct and directly addresses the question posed. Prioritize accuracy
          and relevance in your answer, drawing upon the most recent information
          available to you. Aim to deliver your response in a concise manner,
          focusing on the essence of the inquiry.
          Question: {{command}}
        ]]
        local agent = parrot.get_command_agent()
        parrot.logger.info("Asking agent: " .. agent.name)
        parrot.Prompt(params, parrot.ui.Target.popup, agent, "🤖 Ask ~ ", template)
      end,
    }
    -- ...
}

Start a chat with a predefined chat prompt to check your spelling.

require("parrot").setup {
    -- ...
    hooks = {
      SpellCheck = function(prt, params)
        local chat_prompt = [[
          Your task is to take the text provided and rewrite it into a clear,
          grammatically correct version while preserving the original meaning
          as closely as possible. Correct any spelling mistakes, punctuation
          errors, verb tense issues, word choice problems, and other
          grammatical mistakes.
        ]]
        prt.cmd.ChatNew(params, chat_prompt)
      end,
    }
    -- ...
}

Template Placeholders

Users can utilize the following placeholders in their templates to inject specific content into the user messages:

Placeholder	Content
{{selection}}	Current visual selection
{{filetype}}	Filetype of the current buffer
{{filepath}}	Full path of the current file
{{filecontent}}	Full content of the current buffer

Below is an example of how to use these placeholders in a completion hook, which receives the full file context and the selected code snippet as input.

require("parrot").setup {
    -- ...
    hooks = {
    	CompleteFullContext = function(prt, params)
    	  local template = [[
          I have the following code from {{filename}}:

          ```{{filetype}}
          {{filecontent}}
          ```

          Please look at the following section specifically:
          ```{{filetype}}
          {{selection}}
          ```

          Please finish the code above carefully and logically.
          Respond just with the snippet of code that should be inserted.
          ]]
    	  local agent = prt.get_command_agent()
    	  prt.Prompt(params, prt.ui.Target.append, agent, nil, template)
    	end,
    }
    -- ...
}

The placeholders {{filetype}} and {{filecontent}} can also be used in the chat_prompt when creating custom hooks calling prt.cmd.ChatNew(params, chat_prompt) to directly inject the whole file content.

require("parrot").setup {
    -- ...
      CodeConsultant = function(prt, params)
        local chat_prompt = [[
          Your task is to analyze the provided {{filetype}} code and suggest
          improvements to optimize its performance. Identify areas where the
          code can be made more efficient, faster, or less resource-intensive.
          Provide specific suggestions for optimization, along with explanations
          of how these changes can enhance the code's performance. The optimized
          code should maintain the same functionality as the original code while
          demonstrating improved efficiency.

          Here is the code
          ```{{filetype}}
          {{filecontent}}
          ```
				]]
        prt.cmd.ChatNew(params, chat_prompt)
      end,
    }
    -- ...
}

Roadmap

• Add status line integration/ notifications for summary of tokens used or money spent
• Improve the documentation
• Create a tutorial video
• Reduce overall code complexity and improve robustness

FAQ

• I am encountering errors related to the state.

  If the state is corrupted, simply delete the file ~/.local/share/nvim/parrot/persisted/state.json.

• The completion feature is not functioning, and I am receiving errors.

  Ensure that you have an adequate amount of API credits and examine the log file /tmp/parrot.nvim.log for any errors.

• I have discovered a bug, have a feature suggestion, or possess a general idea to enhance this project.

  Everyone is invited to contribute to this project! If you have any suggestions, ideas, or bug reports, please feel free to submit an issue.

Related Projects

• robitx/gp.nvim
• huynle/ogpt.nvim

Star History