For AI agents, and the people building them
Software can’t buy what it can’t find.
Someone asks an assistant for a phone charger in the next hour. It knows every warehouse on earth and nothing about the store down the street. Canter is how it finds out.
Tell us what you’re buildingfind “usb-c phone charger”
near 94607, within the hour
- store
- Corner market, 0.3 mi
- answer
- likely on the shelf
- confidence
- 0.82
- last sold
- 41 minutes ago, $12.99
- how to buy
- the store’s delivery listing
- not known
- how many are left
Built to be asked, by something that won’t guess.
Agents act on what they’re told. So what they’re told has to say where it came from and how far to trust it.
Ask for a product, near a place.
Standard products, not whatever a receipt called them. A scanner drops digits and a register abbreviates. That mess is ours to clean up.
Get back how sure it is.
Confidence and recency are part of every answer. “Probably, sold 41 minutes ago” is something an agent can reason about. A bare “yes” isn’t.
Get told what isn’t known.
A scan shows a sale. It doesn’t show what’s left on the shelf. The answer says so, so your agent doesn’t promise what nobody knows.
Act through the store’s own channels.
Send the customer to the listing, the delivery app or the door. The sale stays the store’s.
Money, maps and shipping each got one way for software to ask. What’s on the shelf around the corner is next.
The same signal answers other questions too
Tell us what your agent can’t find.
Name the kind of product, store and place your users keep asking about. That’s where we’ll go next.