Four things shipped.
Two dated, not sold.
Four capabilities are running for merchants today. Two are dated and not sold. We would rather you knew the difference before the contract than after it.
What you can use now.
Buyer Agent
A full-screen guided conversation that helps a shopper work out which product is right, then puts it in the cart. Not a bubble in the corner.
- Answers grounded in your own catalog
- Product cards with images, price and stock
- Several matching products as cards, with the trade-off explained in the answer
- Add to your store's own cart on Shopify and WooCommerce, then checkout handoff
- Points shoppers to your team when retrieval is thin, if you've set a contact page
Support Concierge
The questions that arrive after the order — can I return it, does this part fit, what does the warranty cover — answered from your own policy and support pages.
- Returns, warranty and policy questions from your published pages
- Compatibility and fitment questions from the catalog
- Hands off to your team when the answer isn't published
- Every unanswered one logged as a gap in your console
Merchant Intelligence
Catalog gap detection, driven by the questions shoppers actually asked that your catalog couldn't answer, plus listing quality scoring on Pro and above.
- Catalog gap detection from real questions
- Listing quality scoring on Pro and above
- Where your product data is thin
Analytics console
Your own view of what happened: who talked to the agent, where they dropped, and which questions your catalog couldn't answer.
- Sessions → engaged → cart → checkout funnel
- Full transcripts on Pro and above, PII masked by default
- Catalog gap detection
- Scoped to your data only
Where the answers actually come from.
The agent retrieves from your indexed catalog before it writes anything, and only what it retrieved is put in front of the model. Nothing reads the finished answer back to re-check it — we don’t run a fact-checker on the way out, and we don’t claim one.
Your catalog, indexed
Products, specifications, FAQs, policies and delivery rules are ingested at onboarding and refreshed each time we re-sync your catalog and re-crawl your site, so answers reflect your last sync. Embeddings run on NVIDIA NeMo Retriever.
Only your records reach the model
Prices, stock and specifications are read from your data and put in front of the model as retrieved records, not composed by it. Nothing re-reads the finished answer.
Fast when it can be, careful when it matters
Simple questions come back quickly. On Pro and above, comparisons and long threads get a slower, more careful pass; Starter stays on the fast path. The shopper never waits longer than the question warrants.
Built to say so when it can't
If retrieval is thin, the agent is instructed to say it doesn't know rather than guess, and if you've set a contact page it can offer the shopper a link to your team. If every model provider is down, it says the assistant is temporarily unavailable and shows the best-matching catalog products — rather than filling the gap with something plausible.
This is a catalog-grounded system: the answer is written from your data, and when your data doesn’t cover the question the agent is built to say so. There is no fact-checker on the way out. It is not a guarantee of perfection, and we won’t describe it as one.
Not shipped. Not sold. Dated anyway.
Both are targeted for Q4 2026 and are not included in any plan today. If you need one of them, tell us — it changes where it sits in the queue, but it doesn’t change what we’ll promise you now.
Voice commerce
Phone and SIP-ready product discovery for categories where a meaningful share of orders still arrive by phone.
AI visibility (AEO/GEO)
Tracking and improving how your products surface inside ChatGPT, Gemini and Perplexity, with attribution back to revenue.
See it answer questions from your own catalog.
Send us your store URL. We’ll index a slice of your catalog and show you the agent answering real questions about your actual products — before you commit to anything.
No catalog credentials needed for the first preview