AI powered YouTube summaries - my first workflow

[cwille97]

Hey, I'm sharing my first custom Alfred workflow. This workflow summarizes a YouTube video using GPT-3.5. Right now this workflow is very much in an alpha state, but I would be happy to have this hosted on the gallery once I've had some time to gather user feedback and polish it a bit. If you're interested in giving it a whirl, you'll need to provide your own OpenAI API key. The workflow can be downloaded from the GitHub repo under the releases tab. I'd love to see everyone's thoughts on the workflow and gather feedback here. I'll paste in the README below for the full details.

 

How does it work?
First, the workflow parses the URL with a regex to extract the video ID. Next, it leverages [the youtube-transcript-api Python package](https://pypi.org/project/youtube-transcript-api/) to fetch the transcript for the video. Finally, it calls the ChatGPT API with the transcript and requests a summary of the video. This summary is displayed to the user. The only required parameter aside from the video URL, is a ChatGPT API key.

 

Setup
Copy your OpenAI API key and add it to the [Workflow’s Configuration](https://www.alfredapp.com/help/workflows/user-configuration/).

 

Usage
Trigger the workflow with the `ytsum` keyword, and then paste the URL to a YouTube video. Wait a moment while the model generates a summary

Features I'd like to add
* Caching ChatGPT responses - AI models are expensive, it would be nice to have an option to cache responses as a text file and periodically clean them up after 30 days
* Integrate with other models such as Google's Gemini, or GPT 4o, and let the user pick in the configuration or at runtime
* Use Alfred's script input on the Text View to allow the user to ask followup questions to the model

[vitor]

The workflow includes 35MB of extras, including a full Python installation which is only compiled for ARM64, which won’t work for people on Intel Macs.

The better way to do it is to use /usr/bin/python3. To include a Python package with the workflow, use the --target option and set PYTHONPATH. StackOverflow has other options. Make sure to install with /usr/bin/python3 so packages are built with the system python.

[cwille97]

Thanks for the tip! I didn't realize there was an alternative to bundling the full Python installation so I'll definitely plan this for the next version, I'll write up a GitHub Issue now.