The consumer OSfor the 600Mwho don't shopon apps.
Two-thirds of India buys from the shop around the corner — in 12 languages, on cash, credit and WhatsApp. That market had no operating system. We built it: 17,042 stores, 4.25M households on one graph, and an AI orchestration layer that connects farmers, brands and sellers to Bharat through the kirana next door. The rail — not the store — is the product.
One graph. Three surfaces. Producer to household — in one operating system.
Every other player picks a lane — D2C, marketplace, quick-commerce, distributor tech. We built the connecting layer. The retailer's OS became the shelf. The shelf became the AI. The AI became the consumer's shortcut to Bharat. Read left to right.
Marketplace orchestration that starts at the farm and ends at the household.
Producers write into one graph. AI orchestrates supply, price, credit and route. 17,042 retailers hold the shelf. CLUB is the consumer surface. Loyalty and credit ride sidecar — closing the flywheel. Nobody else in India has all five columns in one graph.
Every producer — farm to brand — writes into one shared demand graph. The graph is the coordination.
Eight agents fan out on each request. A verifier gates every output before it hits a shelf or a household.
The retailer never sees the complexity. One SaaS OS — POS, catalog, credit, replenish, route.
Orders arrive through CLUB on the phone Bharat already has. KingCoins locks the household; PayLater unlocks the wallet.
Bharat is already online. The shop wasn't.
The smartphone reached Bharat before the shelf did. The consumer is ready to order. The kirana around the corner is what needs the AI.
The consumer already has the phone. The retailer already has the shelf. The only thing missing was the five-second connection between them — and that's the layer AI just made cheap to build.
— The reason this works now, not in 2018.Any store. Live in 5 minutes. The killer we built.
Ask any distributor-tech company what onboarding takes — the answer is weeks. We onboard a store in the time it takes to make tea. Photograph the shop, point the phone at the shelf; AI reads the SKUs. Price, catalog, delivery pin, WhatsApp rail — provisioned. Not a claim — the median.
Kirana is the beachhead. The OS runs any shop.
Bakery, chemist, electronics, gifts, fruits, restaurant, organic, supermarket — the ROS is format-agnostic. The graph knows what a bakery sells at 7 am and what an electronics store sells on Diwali weekend. Each bar chart is that format's hourly basket curve.
Seven planes. We rent the commodity — only the owned ones compound.
Frontier LLMs, GPU compute, WhatsApp rails, Bhashini ASR — all rented. The orchestrator, the graph, the agents and the primitives — built, owned, versioned. Rent depreciates. Built compounds. Hover a plane to see its share of one request.
Eight agents. One graph. Zero touch on the retailer or the household.
Copilot is a LangGraph state machine — not a chain, not a ReAct loop. Every node has typed I/O, a retry budget, a fallback branch and a verifier before state can advance. Every transition is durable — replayable to the same output six months later.
Twelve languages, one intent graph. The keyboard doesn't reach Bharat.
Bhashini gives us the acoustic model. 4.2M live conversations give us what nobody else can build: grocery vocab, code-mixing, household names — and the confidence to route "godhi hittu" to atta at 8:14 am in Kumbalagudu. Pick a language.
Not a demo. Live software — reaching 600M consumers today.
Bright Store is what the retailer touches every morning. CLUB is what Bharat opens every evening. Distributor is what a brand plugs into to reach the shelf without opening one. Same stack — voice, credit, replenish, route, coin — that turns on the next 13M kiranas. And on every one of them, Swa is the one who talks.
Eight agents need one face. Bharat calls her Swa.
Nobody in Salem wants to talk to an "orchestration layer". They talk to Swa — the assistant inside every ROS screen and every CLUB storefront. She answers in the user's language, cites where every number came from, pulls live data through tools, and only acts after a tap. When she isn't sure, she says so and hands off to a human on WhatsApp. One backend, three audiences — retailer, household, ops.
The kitchens of Bharat, mapped.
Five years × 17,042 stores × pincode × basket × household — the only data-set in India that connects what a manufacturer makes to what a household actually eats. Metro platforms don't have it. Distributors don't. Marketplaces can't rebuild it.
Loyalty locks the household. Credit lifts the basket.
Two sidecar rails ride the graph — KingCoins for repeat, PayLater for reach. More baskets → deeper graph → better credit decisions → bigger baskets. That's what turns a consumption graph into a financial rail.
We didn't build a consumer app. We built the operating system of consumption for a billion people.
The retailer's OS was the foundation. The AI orchestration is the intelligence. CLUB is one surface — but the same graph now underwrites credit, maps farms to kitchens, and quietly rewrites how a billion Indians buy, sell and get financed. One graph. One AI. One India.
We didn't build a consumer app. We built the operating system of consumption for a billion people.
If you build for Bharat — a brand, a lender, a model, a marketplace — this is the rail you plug into.