The knowledge base lets you teach your agents from your own documents: product manuals, sizing guides, policies, FAQs, brand books. Answers drawn from the knowledge base stay specific to your products without those details needing to live in your catalog.
Supported sources
Upload documents as PDF, DOCX, or PPTX. Indexing extracts text per page and also extracts figures and charts using vision AI: images get captions, stay attached to their page, and can be rendered inline in chat answers when they help answer the question.
Indexing hints
Each source accepts optional indexing hints — simple key–value pairs such as product name, SKU, or brand. The hints are attached to every indexed chunk of that document, so a deep page of a manual stays searchable by the product it belongs to, and answers can cite "from the XL-500 manual" even when the page itself never repeats the product name. Labels are yours to choose.
Managing sources
From the knowledge base section of your dashboard you can:
- Upload and edit sources and their hints.
- Re-index a document after you replace the file.
- Browse the extracted images of a document in its edit dialog.
Public versus internal sources
You control whether a source is public or internal. Public sources can have their figures shown inline to shoppers; internal-only sources keep their images behind authenticated, signed links that never reach the public web.
How answers use it
Agents combine knowledge base content with your synced catalog and storefront context, and cite the source when they answer from it. See AI agents overview for how grounding works.