Agents and Assistants can now access and reason over inline images embedded in Knowledge Articles, Blog Posts, Landing Pages, and Site Pages.
Customers need Agents and Assistants to understand all content—including images—embedded in their knowledge sources. Text-only extraction misses critical context, especially for product spec sheets, how-to guides, and visually rich content, limiting the effectiveness of AI-powered support and guidance.
This change ensures that the information customers author, including essential images, is fully available to Agents and Assistants, aligning expectations with existing support for PDFs and Office documents.
When a Knowledge Article, Blog Post, Landing Page, or Site Page with inline images is published or updated, the Knowledge Vault automatically extracts up to 10 qualifying inline images from the HTML content. Images are downloaded, semantically described using an LLM, and indexed alongside the text. Existing content in Knowledge Vaults is automatically re-processed to include images via a migration job. No customer action is required.
At query time, Agents and Assistants receive relevant images and their descriptions in the retrieved context, enabling them to reference, describe, and reason about visual information in responses. Citations point to the parent document, not the image file.
Only body-content images are extracted, not headers, footers, or decorative elements.
All hubs and tiers