Open-source 10-agent system automates YouTube Shorts from podcasts

Jazzlike_Ad_3604 · reddit · 2026-07-22

What it does

The author built and open-sourced a 10-agent system that turns a YouTube podcast into Shorts with almost no manual work: it transcribes locally, finds clip-worthy moments, cuts vertical 9:16 clips with face tracking, burns in captions, adds music/effects, checks for broken renders, and schedules posts.

Agent roles

Finder: transcribes and scores clip-worthy moments
Editor: cuts, reframes, captions, adds music, zooms
Uploader: writes titles/hashtags and posts to YouTube Shorts
Manager: feeds real performance metrics back into the system
Trend Scout / Planner / Community / Finishing Editor / Trainer / Compiler: handle trend detection, creative direction, comment drafting, quality control, weekly playbook updates, and long-form compilation

Notable details

The Finishing Editor is the key guardrail: it rejects broken clips like bad captions, dead air, frozen frames, or poor audio.
It runs on free providers such as OpenRouter, Groq, and Gemini; transcription is local via faster-whisper.
The author says runtime cost is basically zero because the music is synthesized and safe to monetize.
The project is MIT licensed, includes a PDF setup guide, and is designed to be easy to install on Windows/Mac/Linux.

Original post →

Open-source 10-agent system automates YouTube Shorts from podcasts

More from coding & agent