Case study · Built and run by our team
A private AI assistant that reads email, calendar, team chat and a task list, and reports in the chat app already open on the phone. Built and run by our team for our own work.
The context
The person running a business checks an inbox, a calendar, a team chat and a task list, each in its own app, many times a day. The urgent item is usually in the one they have not opened yet.
Vaflunia is the assistant our team built for itself to fix that. She reads all four and reports in Telegram, so the question “what needs me today?” has one place to go.
The challenge
Answer “what needs me today?” across email, calendar, team chat and tasks, on a phone, in one message.
Speak up when something needs attention without turning into another noisy feed.
Draft replies in the right voice, and never send anything without a tap from the person it speaks for.
Stay private: one person’s data, one authorised account, running on a machine we control.
The solution
Delivered functionality and implementation boundaries.
01
There is no new app to install or learn. You message the assistant in Telegram the way you would message a colleague, from any phone or computer. The bot answers one authorised account only; anyone else gets a one word refusal.
02
A Claude model with tool calling decides which systems to look in. “Brief me” reads the calendar, unread email, team chat mentions and the task board, and comes back with one summary that puts the overdue and urgent items first.
03
When you ask for a reply, she writes the draft and shows Send and Cancel buttons. Nothing is sent until you tap Send, and a cancelled draft stays in the drafts folder.
04
Email and team chat are checked every few minutes in the background. When someone mentions you in a channel, the message reaches your phone with who said it and where. Dismissed items stay dismissed, so the feed does not repeat itself.
05
Past conversations and notes are stored in a local search index, so she remembers context from last week. A small web dashboard on the same machine shows the morning briefing, alerts and the task board.
Core functionality
Technical implementation
One Python service runs the Telegram bot, a small web dashboard and a scheduler for background checks. A Claude model with tool calling chooses which system to read; the tools reach email, calendar, team chat, local files and a task board kept in SQLite. Past conversations live in a local ChromaDB index. Sending email always goes through a Send button in Telegram, and the bot answers one authorised account only.
Python · Claude API · Telegram Bot API · ChromaDB · SQLite · NiceGUI
Business impact
travel and hospitality
AI agent and embeddable booking platform
Built and run by our team
Let an agency offer live hotel availability on its own website, in its own colours, without hiring a developer.
Next.js · TypeScript · PostgreSQL · Prisma
View Case Study
travel and hospitality
Booking app and custom CRM
Built and run by our team
Let members search hotels and pay in one flow on a phone, with an app that installs like a native one.
Next.js · TypeScript · PostgreSQL · Drizzle
View Case StudyTell us where the current process breaks down. We can map the workflow, examine the integration requirements and define a practical scope for the build.