AI features
AI features
Two shopping assistants, image generation, growth insights, and how to customize them.
NShop ships AI as a first-class platform feature, not an add-on. There are two shopping assistants, an image generator for admins, and a growth advisor for store owners. All of them run in the EU on Vertex Gemini, with tenant-aware access controls.
At a glance
| Surface | Audience | What it does |
|---|---|---|
| Storefront chat | Shoppers | Answers product and order questions; suggests items; works in EN/NL. |
| Shopping agent (ADK + AG-UI) | Shoppers (opt-in) | Searches, adds to cart, looks up orders, remembers preferences — with authenticated tools. |
| Image generation | Admins | Generates product photography with a preview-and-approve workflow. |
| Growth insights | Admins | Reads 30 days of stats and proposes next moves. |
What makes this different
- Two surfaces, one platform. The casual chat is great for browsing; the agent is great for acting. They share the same persona and the same product database, so the experience is consistent.
- Per-tool authentication. The agent doesn't have a single all-or-
nothing permission. Every tool —
get_orders,add_to_cart,remember_preference— checks the caller's identity individually. Public tools work for anonymous shoppers; private tools require a signed-in session. - Tenant-aware by construction. The agent only sees products and
orders for the tenant whose store the shopper is on. There's no way
to leak across tenants — the JWT minted for each tool call encodes
tenantId. - EU-pinned. Vertex Gemini runs in
europe-west4. Prompts, responses, and embeddings never leave the EU. - Generative UI, not just text. The shopping agent uses AG-UI, so it can render product cards, comparison tables, and an order summary inline — not just paragraphs.
What's next
- How it helps — concrete shopper and admin outcomes, with examples.
- Customize the assistant — persona, tone, and industry settings.