AI Accountant in Telegram with Notion Integration
We built a Telegram bot that turns scattered financial notes into a structured database. Send a photo of a receipt or type "SendPulse subscription, 450 UAH" — the bot reads it, creates a record in Notion, and reminds you about recurring payments.
- Client
- Concept Agency
- Channel
- Telegram
- Database
- Notion
- AI Model
- OpenAI
// Problem
The expenses exist. The bookkeeping doesn't.
Subscriptions, hosting, tools, small payments to contractors — in small businesses and agencies, the finances end up scattered across chats, inboxes, and screenshots. By the end of the month, nobody remembers how much was actually spent, or on what.
Traditional accounting tools mean forms, fields, categories, and yet another app nobody ever opens. Records get made after the fact, and some transactions simply disappear.
We wanted a tool that lives where the conversation already happens — in Telegram. As fast as a message, as structured as a database.
// Solution
A bot that understands transactions in plain language
We built a Telegram bot that takes a photo or a line of text, works out the transaction type, amount, date, and counterparty, and saves it straight to Notion. Five seconds, no forms.
Plain-Language Input
Type "SendPulse subscription" and the bot asks for the amount, builds the transaction, and confirms it with a single tap. No manual forms, no picking categories.
Receipt Photo Recognition
Send a photo of a receipt and OCR plus the LLM pull out the amount, date, counterparty, and category. All you do is tap Save.
Auto-Save to Notion
Every confirmed transaction appears instantly in the Finance database in Notion with all its fields: type, name, amount, date, counterparty, category.
Recurring Payment Reminders
The bot remembers regular payments — hosting, subscriptions, rent — and reminds you before each one is due.
// Why Not a Regular Form
We tried the classic tools. And gave up on them.
Notion, Google Sheets, accounting SaaS — they all work as long as someone disciplined enters the data. In practice, nobody does. So we designed around how people actually behave.
A form takes discipline. A chat doesn't.
Telegram is already where the contractor conversations and the subscription notices land. It makes sense to log the expense in the same window it appeared in.
Picking categories by hand is friction.
The LLM works out the category — Subscriptions, Hosting, Other — from the name and the counterparty. You can override it with one tap, but by default it is already filled in.
Mistakes happen, so everything is editable.
If the bot misreads the name or the amount, the Edit button lets you change any single field without re-entering the whole transaction.
Recurring payments matter most.
Half of a small business's costs are recurring subscriptions. The bot remembers that hosting comes out on the 7th and reminds you ahead of time.
// Technology
Tech Stack
We picked a stack we could ship in a week, without adding more SaaS subscriptions.
Why OpenAI?
OpenAI handles Ukrainian transaction descriptions and receipt photos well — the Vision model reads both handwritten and printed receipts in a single call, with no separate OCR pipeline.
// Key Capabilities
What the Bot Can Do
Text Input
Write it however you like: "SendPulse 500", "hosting, 450 UAH" — the bot reads it and structures it.
Receipt Photos
The Vision model pulls the amount, date, and counterparty out of a photo of a receipt.
Notion Integration
Every transaction appears automatically in the Finance database, with a link to the record.
Recurring Reminders
The bot remembers regular payments and reminds you before they are due.
Edit on the Fly
Save / Edit / Cancel buttons — change any field without re-entering the data.
Auto-Categorization
The LLM sets the category from the name and the counterparty — nothing to pick by hand.
// Results
What changed after launch
внутрішня оцінка — власний інструмент агенції
| Metric | Before | After |
|---|---|---|
| Time to log a transaction | 1–2 minutes in a form | 5 seconds in chat |
| Lost expenses | 20–30% slipped through | 0% — logged on the spot |
| Missed recurring payments | Several a month | Reminders in advance |
| Where records live | Screenshots, chats, Excel | One Notion database |
Bookkeeping stopped being a separate task. Every expense gets logged where it happens — in Telegram. The Finance database in Notion became the single source of truth for every transaction.
Ready to start?
A quick audit to pinpoint where you're losing time and money right now — and whether AI can fix it
