Semantic discovery
Find two actual products before a follow-up question.
Restoring your conversation…
How the shopping assistant works in this demo
All 48 products, prices, stock and policies are mock data. Recommendations use actual catalog tools; the assistant does not place orders. Click a scenario to start testing it in this conversation.
Compare products
Compare real descriptions, EUR prices and stock. No invented quality scores.
Gift with a budget
Browse affordable, available examples; refine the recipient next.
Stock-aware follow-up
Keep the last search criteria and exclude unavailable items.
Approximate price
Rank near a target rather than silently treating it as a maximum.
Language and categories
Switch language, ask for categories, or request different suggestions.
Grounded policies
Use the actual demo policy rather than a model-invented promise.
Scope boundary
See a shopping-only refusal and a path back to the store.
Your chat and cart are saved on this device. Try opening a product and going Back, Shift+Enter for a new line, or New conversation to reset only the chat. Conversations close after 20 customer messages or 24,000 text characters; the final reply can finish and the cart stays saved. Retry repeats the failed question without adding a duplicate. A structured request may run without an LLM call; free-form planning and optional wording use the configured model. Read the architecture and tests ↗