Plain-language intake
Customers or staff write the order as an ordinary message. It is parsed against the live product catalogue — products, quantities, dates — and confirmed back before anything is final.
In production Telegram Mini App
Automaton Sales
Orders arrive as chat messages and leave as filed records and real PDF invoices — nothing retyped in between.
The problem
Orders arrive as messages: a text here, a voice note there. Somebody retypes them into a spreadsheet, somebody else makes the invoice, and once a week an order is discovered three days late. The pipeline lives in people's memories.
What we built
Automaton Sales reads the order as the customer wrote it, checks it against your live catalogue, matches the customer and files the order in Postgres — then confirms back. Invoices are generated as real PDFs and delivered straight to the customer's chat, and a nine o'clock digest chases anything still open.
Capabilities
What it actually does
Every item below is a feature of the running system, not a roadmap.
Customers resolved, not retyped
Each order is matched to an existing customer record, so history, pricing and contact details stay in one place instead of in whoever took the order.
An order book you can edit
Every order is listed in a Telegram mini app with its status. Edit, progress or complete orders from your phone; the record updates for everyone at once.
Real PDF invoices
Invoices are generated in code — no template service, no subscription — and delivered straight to the customer's chat, or to you for checking first.
The morning chase
At 09:00 a digest of open orders arrives, oldest first. When there is nothing open it stays silent, rather than training you to ignore it.
A proper schema
Products, customers, orders and configuration live in Postgres as one source of truth — reporting reads the same rows the app writes.
Under the hood
How it is put together
Nothing exotic. A mini app talking to an action-routed API, with state in Postgres and authorisation checked server-side on every call.
- Runs on your server, against your database, with your credentials
- No per-seat licence and no vendor account required
- Documented at handover, so someone else could take it over
| Front end | Telegram Mini App + bot intake |
|---|---|
| Backend | n8n webhooks |
| Data | PostgreSQL — products, customers, orders |
| Parsing | LLM, constrained to the live catalogue |
| Invoicing | PDF generated in-code, delivered to chat |
| Reporting | Daily 09:00 open-order digest |
| Access | Secret-gated endpoints |
Fit
Who this suits
- Suppliers and wholesalers who take orders by message
- Anyone invoicing out of a spreadsheet at the end of the week
- Businesses whose order book is a chat scroll-back
Tell us what keeps going wrong
A conversation costs nothing and often ends with us saying automation is not worth it for a given job. That is a useful answer too.