Image indexing
Kapa can index images from your knowledge sources so they can be found during retrieval and used in answers. See Images in answers for how images show up across the Prebuilt Agents.
How it works
During ingestion, Kapa detects the images in your sources, filters out decorative ones such as icons and logos, and indexes the ones that carry useful information, such as screenshots and diagrams. Indexed images are matched during retrieval like any other content.
Image indexing requires no configuration. Images are picked up automatically during ingestion and kept in sync with the regular refreshes of your sources.
For a deep dive into how image indexing works under the hood, read How we index images for RAG on our blog.
Supported sources
The following connectors support image indexing:
- Web Crawling: images in the HTML of your web pages, as well as inside PDFs included in the crawl
- File Upload: images in uploaded PDF documents
- S3 Storage: images in PDF documents in your bucket
- Google Drive: images in PDF documents
Images in other sources are not yet indexed.
Supported image formats
Kapa indexes images in PNG, JPEG, and WebP format. On web pages, SVG images are supported as well.
How images are served
How an image is served when it appears in an answer depends on where it came from. Images from crawled web pages are embedded from their original URL on your website. Images extracted from PDF documents have no URL of their own, so Kapa stores and serves them itself; for security, their URLs are signed and expire after 15 minutes.