One thing I learned over the years is that collecting media is easy.
Finding it again is the hard part.
My collection has grown to include thousands of images, videos, music tracks, documents, and AI reference materials. Traditional folder structures and filename searches simply weren’t enough anymore, especially when I only remembered what something looked like or what it contained rather than its filename.
That’s why I started building Coatl Image Library.
More than an Image Gallery
Despite its name, Coatl Image Library isn’t limited to images.
It organizes multiple types of media into a single searchable library:
- 🖼️ Image collections
- 🎥 Video collections
- 🎵 Music libraries
- 📄 Document archives
Each collection is stored in a local SQLite database while semantic search is powered by Qdrant, allowing searches based on meaning instead of filenames.
Search the Way You Think
One of my goals was to make searching feel natural.
Instead of remembering exact filenames or tags, you can simply describe what you’re looking for.
Want to find:
“A sunset over the ocean with orange skies.”
Or perhaps:
“That PDF talking about multi-tenant architecture.”
Or even drag an image into the search box and ask:
“Find things that look like this.”
Coatl uses different AI models depending on the media type:
- CLIP for visual similarity.
- BLIP for image captioning.
- WD14 for automatic image tagging.
- MiniLM for semantic search across documents and song lyrics.
The result is a library that can be searched by concepts instead of just filenames.
Built for Self-Hosting
Everything runs locally.
Media stays on your own machine, metadata is stored in SQLite, and vector embeddings live in a self-hosted Qdrant instance.
No cloud subscriptions.
No uploading personal media to third-party services.
Just your own searchable media library.
Still a Work in Progress
Like many of my projects, Coatl Image Library began as a personal tool and continues to evolve alongside my own workflow.
There are still plenty of ideas I’d like to explore—better indexing strategies, richer metadata, additional AI models, and support for even more media types.
But even today, it has become one of the tools I use most often.
If you’ve ever thought,
“I know I saved that somewhere…”
Coatl Image Library was built for exactly that problem.
Source Code:
https://github.com/zecchan/coatl-image-lib
