Own your pipeline.
Don't rent it.

Podcast clips, generated on your laptop. Open source. MIT.

Paste this prompt into Claude Code (or any AI coding assistant). It walks you through every step, explains what's happening, and handles errors. Recommended if you've never opened a terminal.

Learn more

How It Works

Three steps. Zero manual editing.

Feed it your episode

Point Azelia Clips at any video or audio file. It transcribes with speaker diarization automatically.

AI finds viral moments

Three AI agents (Finder, Critic, Ranker) analyze your transcript on 10 virality dimensions. Only the best survive.

Get clips ready to post

Vertical 9:16 clips with smart face tracking, animated subtitles, and platform-ready captions.

What's in the pipeline

Four moves between your episode and your clip.

Multi-Agent Curation

Finder, Critic and Ranker work in sequence. The Critic rejects ~40% of candidates with reasoning you can read.

Smart Reframing

Face tracking converts 16:9 to 9:16 without cutting heads. Speaker switches pivot the camera.

Word-Level Subtitles

Animated ASS captions with keyword highlighting, burned into the final MP4 via FFmpeg.

10-Dimension Scoring

Hook strength, quotability, storytelling, pacing, emotional arc — each clip scored across ten axes.

Terminal

$ azelia start

╭──────────────────────────────────────────────────────────────────────────────────╮
│   █████╗ ███████╗███████╗██╗     ██╗ █████╗      ██████╗██╗     ██╗██████╗ ███████╗  │
│  ██╔══██╗╚══███╔╝██╔════╝██║     ██║██╔══██╗    ██╔════╝██║     ██║██╔══██╗██╔════╝  │
│  ███████║  ███╔╝ █████╗  ██║     ██║███████║    ██║     ██║     ██║██████╔╝███████╗  │
│  ██╔══██║ ███╔╝  ██╔══╝  ██║     ██║██╔══██║    ██║     ██║     ██║██╔═══╝ ╚════██║  │
│  ██║  ██║███████╗███████╗███████╗██║██║  ██║    ╚██████╗███████╗██║██║     ███████║  │
│  ╚═╝  ╚═╝╚══════╝╚══════╝╚══════╝╚═╝╚═╝  ╚═╝     ╚═════╝╚══════╝╚═╝╚═╝     ╚══════╝  │
╰──────────────────────────────────────────────────────────────────────────────────╯

● Azelia Clips | AI Video Toolkit

🎬 Building frontend...

✓ Frontend built successfully

🎬 Server running at http://localhost:8000

Press Ctrl+C to stop.

$ _

Open the dashboard. Drop your episode. Get clips.

Built in the open. Yours to fork.

MIT licensed · Self-hosted · Your data stays on your machine.

Star on GitHub

Frequently Asked Questions

Is it really free?

Yes. MIT licensed. 100% free, no paid tier, no subscription. You pay your LLM provider directly (or $0 if you use Claude Code with your existing Claude subscription).

What do I need to run it?

macOS or Linux, Python 3.11+, FFmpeg, and either a Claude Code subscription (recommended — no API key needed) or an Anthropic API key. Apple Silicon with 16GB+ RAM recommended for speed.

Does my audio ever leave my machine?

No. Transcription runs locally via Whisper (MLX on Apple Silicon). Only the transcript text (not the audio) is sent to the LLM provider you choose. If you use Claude Code, the whole pipeline runs through your local CLI — no cloud round-trip from Azelia itself.

Does it work in other languages?

Yes. Whisper supports 50+ languages out of the box. English and Spanish are most tested.

Can I use it commercially?

Yes. MIT allows full commercial use — client work, agencies, any commercial purpose, without restrictions.

What is Azelia?

Azelia is a growing set of open-source tools for podcasters. Azelia Clips is the first one.

Your audience is scrolling right now.

The clips are already in your podcast. Let Azelia Clips find them.