A fragrance brand, run from one app and from a chat
The Mantra makes small-batch perfumes in Malaysia and sells them through a network of agents. We built the system that runs it: batches, stock, agents, sales and commissions in one app. The public shop runs on the same data, and the owners can ask their AI assistant how the business is doing.
- every bottle, back to the batch it was poured from
- Tracedevery bottle, back to the batch it was poured from
- every screen updates the moment a sale is recorded
- Liveevery screen updates the moment a sale is recorded
- for an AI assistant to read the whole business
- ~160 tokensfor an AI assistant to read the whole business
- each agent sees just their own stock and sales
- Own data onlyeach agent sees just their own stock and sales
Perfume made in batches, sold by a team of agents
Every perfume is blended in a batch and left to mature before it is bottled. Agents across Malaysia take stock from HQ, sell to their own customers, and earn commission. That means a lot to keep track of: which batch is ready, who holds which bottles, who has paid, and who is owed.
HQ
Makes a batch
- Blended and matured
- Bottled into sizes
- Oldest batch sold first
The Mantra app
Tracks every bottle
- Stock at HQ and with each agent
- Stock requests
- Sales and payments
- Commissions
Customers
Buy
- From an agent
- From HQ
- Browse the public shop
Everything the business runs on, in one place
Admins see the whole business. Agents log in to the same app and see only their own stock, sales and commissions, on any phone.
Inventory
Sales
Agents
Commissions and payouts
Stock requests and interest
The screens the team uses every day

Ask your AI assistant how the business is doing
We gave The Mantra its own MCP server, the open standard AI assistants like Claude and ChatGPT use to connect to software. The owners can ask “how did we do this month?” or “what’s running low?” and get an answer from the live numbers, in plain language, on their laptop or phone.
On claude.ai it connects through a proper sign-in, with nothing to paste. Desktop and coding tools use a personal key. A snapshot of the whole business costs about 160 tokens. The AI can file a stock request or note a customer’s interest, but it can never move stock or take money.
You ask your AI assistant
- How did sales go this month compared to last?
- Which products are running low?
- Who are our top agents, and who still owes us?
- Request 10 more Black Velvet 30ML for me.
Claude, ChatGPT or any MCP-ready assistant picks the tool, calls the-mantra and answers from what comes back, on a laptop or a phone.
Read
- overviewrevenue, orders, what’s owed, low stock
- salessales by period, agent or channel
- inventorystock at HQ and with each agent
- batchesbatches and when they mature
- paymentswho has paid and who owes
- rankingsbest products and agents
Act · limited on purpose
- create_stock_requestask HQ for more stock
- record_interestnote a customer’s interest
Nothing to install and no developer needed. Each key belongs to one person and one device, and the usage panel shows exactly what the AI looked at.
Acts as the person signed in
Keys you can switch off
Every question logged
Write once in Notion, publish everywhere
The Mantra’s content lives in a Notion board. Once a post is approved, our Automator platform publishes it to Instagram, Facebook and Threads on schedule, records what went out, and messages the team on Telegram if anything fails. It also warns them twice a day when the queue of approved posts is running low.
One backend, four ways in
The app, the public shop, the AI connection and the automations all work from the same live data, so there is one version of the truth and nothing to sync.
The app
The shop
MCP
Automations
- Next.js
- React
- Convex
- TypeScript
- Tailwind CSS
- shadcn/ui
- MCP · OAuth 2.1
- Automator (ours)
- Notion
- Meta Graph API
- Telegram