This is a submission for the AI Agents Challenge powered by n8n and Bright Data
What I Built
I have subscribed to some amazing youtube channels from my country Ghana which i always look forward to watching their videos daily. Most times, i just want to forward the video links somewhere to get a summary and have a before-taste of since most of their videos are often more than 20 mins long, and i sometimes don't have the luxury to fully watch them at that instant. That's where the idea of this workflow came from. A way for me to get a summary and insights of a video that i can read in less than 5 minutes to build up anticipation, just by forwarding the link of the video to my system.
Demo
Youtube demo:
Screenshots:
n8n Workflow
Workflow gist:
Technical Implementation
System Instructions:
The system is instructed to follow these key components in its entire journey: Role, Task, Tools and Outputs. A specific role as summarizer for Youtube & TikTok videos for a daily digest newsletter. A comprehensive task on how to process the data, styles to use for summarisation and which tools to use based on its description. It's being given tools as well. An input and output format were shown to it as examples of what it would be receiving and what its output should look like.
Model Choice:
The workflow is powered by Google Gemini 2.5 Flash. It is designed to be model-agnostic, but the two-agent structure allows for optimization. For instance, a faster, cheaper model can be used for the technical parsing task, while a more powerful, creative model like Gemini 2.5 Pro could be used for the final content generation if desired.
Memory:
It is using the Simple Memory to keep track of the process.
Tools:
Only the "sendNewsletter" tools is used to send the cacth-up email to the user through the send gmail node.
Bright Data Verified Node
I used the Bright Data Verified Node to get data of Youtube and TikTok videos. It was also able to get me transcript of Youtube videos but couldn't for TikTok videos so i used the HTTP REQUEST node for that.
Journey
As a first time n8n workflow builder, it was tough. Thanks to some amazing resources from n8n the community support form, youtube videos like https://youtu.be/SoDip-tvZ0Y?si=5Y1Y1nfk_9_Y1aH9; i was able to make it through.
Top comments (2)
Practical idea and a clean implementation. The model-agnostic, two-agent setup makes sense, and the Bright Data + HTTP fallback for TikTok transcripts is a nice touch. Would be great to hear how you handle rate limits or non-English transcripts. Nice work, and good luck with the challenge.
Great work! Love seeing you build and share more of these projects 👏