Library
The Library finds your videos, photos, audio and documents by what they show, say or contain. Search in your own words, in English or Spanish. Everything is indexed on your computer (<data>/index.db) and nothing leaves it.
#How to use it
- Open Library in the sidebar and click Add folder. Subfolders are included. What you create with Genesis (AI Studio, Assistant…) is always there, with its prompt as text.
- Click Index now. Only what's new or changed since last time is processed, and deleted files are removed from the index.
- Type what you're looking for. Genesis separates the filters from the words and shows you what it understood.
Search examples:
| Search | What it understands |
|---|---|
vertical videos of the beach |
videos · vertical · «beach» |
where I talk about the budget |
«budget» in what is said or written |
red photos generated with AI |
images · red color · generated with AI |
clips under 2 min in English |
videos · duration < 2 min · English language |
invoice pdfs from this month |
documents · last 31 days · «invoice» |
fotos cuadradas de mi perro de los últimos 3 días |
images · square · last 3 days · «perro» |
For videos and audio, each result lists the moments where what you searched for is said. Clicking one opens the video in the editor right at that point.
#What gets indexed
| Type | Data |
|---|---|
| Videos | Duration, resolution, orientation, cover, dominant colors and what is said (transcription with local Whisper, with timestamps). |
| Audio | Duration and what is said. |
| Images | Resolution, orientation and dominant colors (red, blue, green…). |
| Documents | Text from Word (DOCX), PDF, Markdown and TXT. |
| Spreadsheets | Cell contents (XLSX, XLS, ODS, CSV). |
| Presentations | Titles, text, bullets and notes (PPTX). |
Automatic tags:
- type;
- vertical, horizontal or square;
- short (≤ 1 min) or long (≥ 10 min);
- spoken language;
- colors;
- «generated with AI».
With Describe photos and videos with AI, a vision model looks at each photo (and three frames of each video). It saves a sentence, keywords and the text that appears in the image. That way «photos of dogs on the beach» or a name written on a sign finds files nobody tagged.
- With Qwen2.5-VL downloaded in Local AI, everything happens on your computer.
- If you don't have it, your connected account (Claude, GPT or Gemini) is used, and the images are sent to that service. That's why the option is off by default.
- Only what hasn't been described yet is described.
With Summarize and tag with AI, a model (local or your connected account) adds a summary and topics to text-heavy files.
#Search by meaning
Download Qwen3 Embedding 0.6B (640 MB, over 100 languages) in Settings → Local AI → Search. With it, the Library finds «sunset on the coast» even if the file says «evening at the beach». It's combined with keyword search: exact matches weigh a little more. Results found only by meaning are marked Similar.
Transcription uses whichever Whisper model you've downloaded. Indexing never downloads models or uses paid services.
#From the CLI and via MCP
Add a folder:
genesis run library.add_folder path=~/VideosIndex:
genesis run library.indexIndex and also describe photos and videos with AI:
genesis run library.index describe=trueDescribe a single image:
genesis run ai.describe_image input=photo.jpgSearch:
genesis run library.search query="where I talk about the budget"library.search and library.index are also available to any AI connected via MCP. For example, you can ask Claude Desktop «find the clip where I talk about the budget and cut it»: it finds the moment with library.search and cuts it with video.cut.