macLLM is a utility that makes it easy to work with LLMs from the macOS desktop. It is launched via a hotkey and can do things like:
- Send a prompt to an LLM, e.g. "What state is Kansas City in?"
- Translate or summarize text, e.g. "Translate the @clipboard to Spanish"
- Look at desktop windows, e.g. "Write a short biography based on this LinkedIn @window"
- Find good emojis for a specific topic, e.g. "@emojis: diving in hawaii"
- Work with URLs, e.g. "summarize @https://github.com/appenz/macLLM/edit/main/README.md"
- Work with files, e.g. "find the ship date in @~/Documents/Notes/team-meeting.md"
macLLM is open source (Apache 2.0) and designed to be easily extensible with your own workflows. Here is a short demo video (with sound) of some of the capabilities.
FFMpeg-5fps.mp4
macLLM uses the uv package manager, make sure you have it installed first (https://github.com/astral-sh/uv).
Your OpenAI API key should be configured in the environment variable OPENAI_API_KEY.
OPENAI_API_KEY="your_api_key"
Now you can run macllm it with:
uv run macllm/macllm.py
uv should take care of all the dependencies.
Press the hotkey. By default, it is option-space (⌥-space) but it can be easily remapped. A small window will appear in the center of the screen and you can start typing a query to the LLM. For example:
Capital of france?
After a second or so you should ge the answer "Paris". You can now do a few things:
- Hit escape to close the window. Pressing the hotkey again will close it as well.
- You can copy the text to the clipboard by pressing Command-C (⌘-C). This will also close the macLLM window.
- Type a new query into the text box.
macLLM understands a number of external data sources:
- @clipboard is the current clipboard content
- @window is any desktop window. You can select it after entering the query with your mouse.
- @selection allows you to select any area on the screen.
- @ is any file in macOS. The path has to start with "/" or "~"
- @ for an http url. It has to start with "http" or "https"
The data can be referenced in the query, e.g. "translate @clipboard into French" or "summarize the slide @window".
Shortcuts are shorthand for specifc prompts. They always start with the @ symbol. So for example "@fix" is a prompt that corrects spelling or "#tr-es" is short for translate to Spanish.
@fix My Canaidian Mooose is Braun.
This gets internally expanded to:
#Correct any spelling or grammar mistakes in the following text: My Canaidian Mooose is Braun.
Which GPT will correct to:
My Canadian Moose is Brown.
You can add your own shortcuts in the shortcuts.py file.
#exp: Expand the text using sophisticated and concise language.#fix: Correct spelling and grammar mistakes in the following text.#fix-de: Correct spelling and grammar mistakes in the following German text.#rewrite: Rewrite the text to be extremely concise while keeping the same meaning.#tr-de: Translate the text from English to German.#emoji: Pick a relevant emoji for the text and reply with only that emoji.#emoji: Pick a few relevant emojis for the text and reply with only those emojis.
If the clipboard contains a text that starts with the trigger (default is "@@"), it will be sent to the LLM. This can be useful if you want to use macLLM from withint an editor. Just type "@@" followed by instructions (e.g. "@@shorten this paragraph:"), hit copy (i.e. Apple-C), wait 1-2 seconds, and paste (Apple-P).
Apache 2.0