Workout Buddy

Play your favorite workout video on YouTube and mute it. Workout Buddy will attempt to detect the time remaining for the current exercise using an on-device vision language model, and send an audible notification when it's time to stop.

Demo

Source video can be found here. I am not affiliated with nor endorsed by the video author.

Features

• Video-agnostic. It works with any YouTube video, as long as it contains a countdown timer
• Ignores irrelevant scenes (e.g. other tabs, windows)
• Runs entirely on your device. Optimized for LM Studio and the Gemma 3 4B model, but anything with vision and OpenAI API-compatible should work
• Analyzes 3 screenshots at once, allowing the model to find less obvious timers
• Doesn't save captured screenshots for privacy
• Customizable notifications and sounds

Installation

1. Download the pre-built binary from Releases or build the code yourself.
2. In the working directory, create the configuration file - config.json - and fill it out according to the example in config.example.json. Optionally, find a notification sound and point to it.
3. Ensure you have the LM Studio API running and the model is loaded.

Usage

1. Open your workout video on YouTube.
2. Launch the binary.
3. If you disabled send_notification_overtime, wait until the app catches up with the video.
4. Move in silence or to your own music!

Disclaimer

I've created this project both as a Rust exercise and to scratch an itch. While it exceeds my personal expectations, it may not handle all videos as effectively. It should be considered more as a proof-of-concept. Have fun exercising!