Outcome 1
Recommends products from your catalog
The chat reads your catalog and answers shoppers in their words, in the language they typed. "Wireless gaming mouse, low latency, under €60" returns a specific product with price, stock, and add-to-cart, instead of a search results page. A photo upload triggers visual search; a screenshot of a competitor's product becomes a recommendation from your catalog.
- Text or photo query
- Variant-aware: returns the right size, color, configuration
- Cross-language: French shopper, English catalog, no extra work
For developers
The shopper's question is expanded by a language model, matched against vector embeddings of your catalog in a self-hosted vector database, then reranked by a cross-encoder. Image queries are turned into a text description by a vision model and run through the same retrieval path. Catalog data syncs from the native plugin's webhook, not a script-tag scrape.