← Back to changelog

AI Agent Recognises Products from Photos

The AI agent can now identify a product in your catalogue from a photo sent by a customer. It matches the photo against your product cards and responds strictly from the card — pulling price and specs from it, not from the image. If the photo isn't clear enough to identify the model, the agent asks a follow-up question — for the article number, a different angle, or a label shot — and keeps the conversation without escalating to a manager.

Training is based on the product cards in your Knowledge Base. One model equals one card, with a name, article number, price, and several photos: full view, details, and label. A different article number or strap colour gets its own card. Once photos are uploaded, training starts automatically and takes a few minutes.

The Knowledge Base now includes a setup guide block, a preparation status indicator, and a "Test in chat" button — upload a customer photo and ask "is this model available?" to verify the agent responds correctly.