Woxly

Zero-commission commerce for Indian sellers: storefront, seller dashboard, AI assistant, and a public API.

Next.js, TypeScript, NestJS, PostgreSQL, Razorpay, Multi-tenant, AI assistant, Public API
Role
Founding engineer · 2025 – present
Owned
Architecture, multi-tenant data model, UI system, the seller dashboard, the in-app AI assistant, and the landing, storefront, and docs frontends.
Result
Four surfaces live on one commerce core: a seller dashboard covering orders, catalogue, customers, growth, and reports; a streaming AI assistant with 96 built-in guides; Razorpay and COD checkout; and a public /api/v1 that AI agents can build on.
Next.jsTypeScriptNestJSPostgreSQLRazorpayMulti-tenantAI assistantPublic API
Woxly cover

Overview

Woxly is a commerce platform for Indian sellers who run their business out of WhatsApp and Instagram chats. A seller claims a name.woxly.store address, adds products, and shares one link. Orders, payments, delivery, and customer chats then run from one dashboard. Woxly charges a flat monthly price and takes nothing per order.

I joined as the founding engineer. I own the architecture and the UI system, and I build every surface sellers and their customers touch:

Screenshots come from the demo store. Customer names, phone numbers, emails, and addresses are blurred.

Problem

Small sellers in India already sell well over chat. What breaks is everything after the message: stock in a spreadsheet, payment screenshots in a gallery, addresses copied by hand, and no idea which orders still need attention. Marketplaces fix the operations but take a cut of every order. Generic store builders assume a designer, a developer, and an evening spent on themes and plugins.

So the bar for Woxly was:

  • A seller with no technical help has a store they're proud to share in under ten minutes.
  • Every day starts with one answer: what needs me right now?
  • The seller keeps selling in the chats where their customers already are.
  • When they're stuck, help is one question away, inside the product.

Seller dashboard

The dashboard is the product. Most of my time goes here, because it decides whether a seller comes back tomorrow.

Home opens with Today's work: the count of orders waiting for confirmation, flagged by priority, with one link to review them. Sales, orders, and store visitors sit below with sparklines and a live-visitor count. The two most common actions, Open online store and Share shop, are always one click away.

Orders move through clear stages (New, Preparing, Out for delivery, Delivered) with a Needs action filter that collects everything waiting on the seller. Each order page answers three questions at the top: what stage is it in, what's been paid, and what's the next step. The primary button changes with the state: Start preparing, or Send payment link when an online payment is still due.

The rest of the page covers fulfilment:

  • Payment links the seller can copy, open, or send over WhatsApp, with a live payment check
  • Shiprocket shipping with live courier rates, plus manual tracking for any other courier
  • Invoices with a shareable link, normal print, and Bluetooth thermal-printer printing
  • Customer card with one-tap Call and WhatsApp Chat, a delivery address, and tags
  • Activity timeline and team-only notes, so staff can hand an order over without a call

The product list shows stock, variants, sold-out toggles, and live status, with bulk upload for sellers whose catalogue already lives in a sheet. Adding a product is a three-step flow (details, pricing and stock, review and publish) instead of one long form. Generate with AI writes a title and description from what the seller has, and a Woxly Quick Help card offers a guided walkthrough the first time.

  • Customers: every buyer from the online store, manual entry, or import, with tags, loyalty coins, and export.
  • Offers and discounts: code-based or automatic offers, scoped to products, new customers, or online payments, with usage counts and a pause switch.
  • Abandoned carts: shows how much is sitting in carts, sends automatic email reminders, and tracks how much those reminders recover.
  • Loyalty, wishlists, reviews, and email marketing sit in the same Grow your store section.

Reports split into sales, products, customers, traffic, and loyalty, each with the same date controls. Ads & tracking connects Google Analytics, Meta Pixel, Microsoft Clarity, and Google Tag Manager behind one store-wide switch, and tells sellers which ones they actually need, so they don't connect all four out of fear. Settings covers the store profile, plans and billing, security, staff with roles, order alerts, and developer tools.

Many sellers run the whole business from their phone, so mobile is not a squeezed desktop. The sidebar becomes a bottom tab bar (Home, Orders, Products, Customers, More). Order tables become cards with the next action on each, and the order page moves the stage, payment status, and primary button to the top, where a thumb can reach them.

Woxly Assistant: help inside the product

The dashboard has a lot of surface area, and new sellers get stuck in the same places: how to add variants, where the reports are, how to connect a pixel. Instead of sending them to a help centre, the Woxly Assistant answers inside the dashboard and takes them to the right screen.

An answer isn't a wall of text. It comes back as a structured card: Go to the page that matters, numbered Steps with deep links, the Result the seller should expect, a primary action button that opens the right screen, and two follow-up questions. The seller can copy the answer, retry it, edit their question, or rate it.

How a question becomes an answer
  1. Open from anywhere

    The launcher sits on every dashboard page. Contextual Quick Help cards (like Add a product → Show me how) and onboarding nudges open it with the question already filled in, through one woxly:open-assistant browser event any screen can send. A per-store flag turns the assistant on or off.

  2. Send the question with context

    The client POSTs to /api/assistant with the query, the page the seller is on, the last six turns of the conversation (each capped at 400 characters), and an optional coach step ID. The session is checked on the server; an expired session asks the seller to sign in again instead of failing silently.

  3. Ground it in the product

    Answers are tied to a catalogue of 96 guides across 24 dashboard areas, from adding variants to connecting Shiprocket. Each guide names the page and the feature it needs, so an answer can point to the exact screen, and staff without access get a notice about it instead of steps they can't follow.

  4. Stream the answer

    The response streams as NDJSON: partial events carry the answer text as it's written, then one final event carries the structured response (answer, steps, go-to targets, links, result, tip, warnings, intent, and a feedback token). If streaming isn't available, the same response comes back as plain JSON.

  5. Stay calm when things fail

    A 20-second stall timer resets on every chunk. Stop cancels the request with an AbortController and keeps the partial text. Every response is checked against the expected shape before it renders. When it fails, the seller gets Try again, Ask us on WhatsApp, and Help Center, never a blank bubble.

  6. Remember the conversation

    Conversations are saved per store in the browser: the last 10, up to 80 messages each, titled from the first question, and checked when they're loaded back. History and New chat sit in the panel header.

  7. Learn from every answer

    Thumbs up or down posts the signed feedback token to /api/assistant/feedback and logs the intent and feature, so weak answers show up by topic, not by anecdote.

txt
{"type":"partial","answer":"Let's look at your store performance…"}{"type":"partial","answer":"…Start with Sales, then check Traffic"}{"type":"final","response":{"answer":"…","goTo":["Reports"],"steps":[…],"result":"…","intent":"…","feedbackToken":"…"}}

What I cared about most:

  • Answers lead somewhere. Every answer ends on a button that opens the right screen, because the seller asked in order to do something.
  • Context without effort. The current page and recent turns go with every question, so "how do I change this?" works without the seller explaining where they are.
  • Honest failure. Streaming has a watchdog, Stop keeps what was written, and every error ends in a human route (WhatsApp or the Help Center).
  • Respect roles. Guides know which page they need, so a staff member without access is told why, instead of being sent to a screen they can't open.

Landing: sell the evening, not the feature list

The landing page has one job: get a seller to type their store name. The hero leads with the outcome ("Build Online Store in minutes"), removes the three fears (no designer, no developer, no technical knowledge), and puts the .woxly.store name field right there. Below it:

  • The all-in-one platform: storefront, orders and delivery, payments, and social selling, each with a real product preview instead of an icon grid.
  • 4 steps, one evening: build, add products, sell everywhere, run everything.
  • Flat pricing: Foundation, Professional, and Enterprise, all with 0% commission, a monthly/yearly toggle, and a 1-week trial with no card.
  • Questions sellers ask first: commission, COD, keeping WhatsApp and Instagram, Google visibility, and outgrowing a plan.

Build with AI: docs as a product surface

Many sellers already use ChatGPT or Claude, so the landing page says it plainly: paste one link into your agent and let it build the store. That only works if the docs are written for machines as well as people. docs.woxly.store documents the public /api/v1 for Professional and Enterprise plans:

  • Storefront API: catalog, cart, checkout, shopper OTP, wishlist, reviews
  • Admin API: products, orders, review moderation, abandoned carts, wishlist reminders
  • Webhooks and OAuth: event delivery and third-party installs

Every page has Copy for AI and a Markdown version (append .md). An llms.txt index, an AI context pack with the non-negotiable rules, an llms-full.txt dump, and an interactive OpenAPI reference give an agent everything it needs in one fetch. A headless Next.js starter and copy-paste prompts cover the rest.

Storefront

demo.woxly.store is a live store on the same engine every seller gets. It has category browsing, product variants, discount codes, reviews, cart, and checkout with Razorpay or cash on delivery, and every order lands in the dashboard pipeline above. Themes control colours, fonts, and banners, so two stores on Woxly don't look like the same template.

Architecture

  • Multi-tenant core. One model for products, customers, orders, and payment state, with tenant-aware routing and isolation. Every store gets its own subdomain or custom domain.
  • Two API layers. A first-party API for the dashboard and hosted storefront, and a separate public /api/v1 with storefront and admin tokens for custom frontends and agents.
  • Payments and shipping. Razorpay and COD on every plan, payment links for orders still due, and Shiprocket for rates and AWBs. Payment events drive the order lifecycle.
  • Messaging. WhatsApp Business (Meta sign-in or Interakt) for order updates, and a shared inbox for WhatsApp, Instagram, and Messenger chats, with auto-reply for shared Reels and posts.
  • Marketing stack. Google Analytics, Meta Pixel, Clarity, and GTM behind one tracking switch.
  • One UI system. The landing page, dashboard, storefront, and docs share tokens and components, so the product feels like one company built it.

Working with the team

As a founding engineer I sat between the founders, a product manager, and QA.

  • Founders: they set the direction. I turned it into scope we could ship, and flagged early when a feature would slow down a seller's first ten minutes.
  • Product manager: we shaped specs together and broke features into slices that could ship on their own.
  • QA: I worked through test passes on checkout, payments, the dashboard, and the assistant, and fixed what they found before it reached sellers.

Outcome

  • Four surfaces live: seller dashboard, landing, storefront, and developer docs
  • A seller dashboard covering orders, catalogue, customers, growth, reports, tracking, and settings, designed for desktop and phone
  • A streaming in-app assistant grounded in 96 guides across 24 dashboard areas, with structured answers, deep links, and a feedback loop
  • Setup in under ten minutes, 0% commission on every plan, Razorpay and COD from day one
  • A public API documented for humans and AI agents, with llms.txt, Markdown pages, and OpenAPI
  • Woxly is a Meta Verified Tech Provider

Stack

Next.js, TypeScript, Tailwind CSS, NestJS, PostgreSQL, Clerk, Razorpay, Shiprocket, WhatsApp Business API.