Voice AI for neighbourhood retail
ukan
A voice agent that takes phone orders for neighbourhood supermarkets and kiranas — in Hindi or English, knows the catalog, and routes to a human only when it has to.
Request a callback to try the agent →
Request a callback
What you fill in here, the agent will already know when it dials. Note: during the trial only pre-verified numbers can be called.
Saved as your default address. The agent will confirm rather than ask for it.
Operator cost per outlet today
₹95k/month
4 phone operators at ~₹20k each
Peak-hour calls lost
15–25%
Customers default to Zepto / Blinkit
Repeat callers
60–80%
Same regulars, same baskets, week after week
How a call works
- 1Customer dials the store number — the voice layer routes the call to the Dukan agent.
- 2Agent looks up the caller's phone in the merchant database. Greets returning customers by name in their preferred language.
- 3Customer places the order in natural Hindi / English / mixed. Agent calls
catalog_searchfor each item; suggests substitutes when out of stock. - 4Confirms address (returning customer = on file, new = capture + validate pincode). Reads back the full order with total.
- 5Places the order — written to the merchant's dashboard in realtime. Triggers SMS confirmation. Escalates to a human only for disputes, complaints, or bulk orders.
Dashboard
Live KPIs + active calls
Orders
The phone-order queue
Catalog
What the agent can sell
Escalations
Calls that needed a human
Provider-agnostic voice layer by design.
The voice layer sits behind a VoiceProvider interface with swappable adapters — a hosted voice-AI provider by default, plus a custom ElevenLabs + Deepgram + LLM stack. Swapping providers is one env var.