Turn meeting recordings into clean summaries and actionable to-do lists — entirely on your own machine 🔒

🔹 About

KhulasaAI is a privacy-first Telegram bot that turns meeting recordings into clean summaries and actionable to-do lists — entirely on your own machine. No paid APIs, no cloud transcription, no data ever leaves your computer.

Send a voice note or audio file to the bot, and get back:

• A concise, adaptive-length summary of the meeting
• A clean list of Action Items, with owners identified when mentioned
• A saved record of every meeting in a local SQLite database

✨ Features

🔒	100% local processing — speech-to-text and summarization both run on your machine
🎤	Accurate Arabic transcription via faster-whisper
🧠	Local LLM summarization via Ollama — works with any chat model you have pulled locally
📏	Adaptive summary length — short meetings get tight summaries; long meetings get organized bullet lists
🧩	Map-reduce chunking for long meetings — avoids context-window truncation issues
🗃️	Local archive — every transcript, summary, and action item list is saved to SQLite
🤖	Simple Telegram UX — just send a voice note or audio file, no commands needed

🛠️ Tech Stack

📋 Requirements

• Python 3.10+
• FFmpeg installed and available on your system PATH
• Ollama installed and running locally
• A Telegram bot token from @BotFather

🚀 Getting Started

1. Clone the repo

git clone [https://github.com/RamiDevX/KhulasaAI.git](https://github.com/RamiDevX/KhulasaAI.git)
cd KhulasaAI

2. Create and activate a virtual environment

python -m venv .venv

# Windows
.venv\Scripts\activate

# macOS/Linux
source .venv/bin/activate

3. Install dependencies

pip install -r requirements.txt

4. Configure environment variables Create a .env file in the project root:

BOT_TOKEN=your_telegram_bot_token_here
WHISPER_MODEL_SIZE=small
OLLAMA_MODEL=llama3.2
OLLAMA_URL=http://localhost:11434/api/chat

5. Run the bot

python bot.py

📂 Project Structure

KhulasaAI/
├── bot.py             # Telegram bot entry point & message handlers
├── transcription.py   # Local speech-to-text via faster-whisper
├── summarizer.py      # Local LLM summarization via Ollama
├── database.py        # SQLite persistence layer
├── config.py          # Environment-based configuration
├── requirements.txt
├── .env.example
└── README.md

📝 Notes

Note

The first run downloads the faster-whisper model weights, so the first request takes longer than usual.

[!NOTE] For better Arabic output quality, consider using an Arabic-tuned model like command-r7b-arabic.

[!WARNING] Never commit your .env file or bot token to version control.

👨‍💻 Developer

Rami Bitar

_{KhulasaAI © 2026}