Skip to main content

Setting up the AI Rewrite Plugin

Matcha includes an ai_rewrite.lua plugin that allows you to rewrite email drafts using an AI model. By default, it works with any OpenAI-compatible API.

To get started, install the plugin:

matcha install https://raw.githubusercontent.com/floatpane/matcha/master/plugins/ai_rewrite.lua

You can then edit ~/.config/matcha/plugins/ai_rewrite.lua to configure it for your preferred AI provider (matcha config ai_rewrite). Since the plugin relies on the OpenAI chat completions format, it seamlessly integrates with OpenAI, local providers like Ollama, and other services that offer an OpenAI-compatible endpoint (like Gemini). For providers without native OpenAI compatibility (like Claude), you can use a local proxy like LiteLLM.

Here are the configuration snippets for various popular AI providers. Update the variables at the top of your ai_rewrite.lua file.

Ollama (Local)​

Ollama provides a local, private OpenAI-compatible endpoint out of the box. Make sure you have pulled a model first (e.g., ollama run llama3).

-- Configuration for Ollama
local API_URL       = "http://localhost:11434/v1/chat/completions"
local API_KEY       = "" -- Not needed for Ollama
local MODEL         = "llama3" -- Replace with your downloaded model name

OpenAI​

To use OpenAI's models, you will need an API key from your OpenAI platform dashboard.

-- Configuration for OpenAI
local API_URL       = "https://api.openai.com/v1/chat/completions"
local API_KEY       = "sk-proj-YOUR_OPENAI_API_KEY"
local MODEL         = "gpt-4o-mini" -- Or "gpt-4o", "gpt-3.5-turbo", etc.

Google Gemini​

Google Gemini recently introduced an OpenAI-compatible endpoint. You can use your Gemini API Key directly.

-- Configuration for Google Gemini
local API_URL       = "https://generativelanguage.googleapis.com/v1beta/openai/chat/completions"
local API_KEY       = "YOUR_GEMINI_API_KEY"
local MODEL         = "gemini-1.5-flash" -- Or "gemini-1.5-pro"

Anthropic Claude​

Anthropic's API does not natively use the OpenAI format. However, you can easily use Claude by running a lightweight local proxy like LiteLLM, which translates OpenAI-formatted requests into Anthropic's format.

  1. Install and start LiteLLM with your Anthropic key: 
    pip install litellm
    litellm --model claude-3-5-sonnet-20241022 --api_key YOUR_ANTHROPIC_API_KEY
  2. LiteLLM will start a local server, usually on port 4000.

Configure the plugin to point to your LiteLLM instance:

-- Configuration for Claude (via LiteLLM proxy)
local API_URL       = "http://localhost:4000/v1/chat/completions"
local API_KEY       = "" -- Handled by LiteLLM
local MODEL         = "claude-3-5-sonnet-20241022" -- Must match the LiteLLM model

Usage​

Once configured:

  1. Restart Matcha to load the plugin.
  2. Open the composer.
  3. Draft your email.
  4. Press ctrl+r to trigger the AI Rewrite prompt.
  5. Provide an instruction (e.g., "Make it more formal", "Fix typos", "Shorten it").
  6. The AI will rewrite your draft and replace the body content automatically.