F04Personal project · 2026 · Runs on my server
AI Mohammed
A Telegram bot that replies in my own texting style. It reads photos, transcribes voice notes, answers with a voice note in my cloned voice when someone asks to hear me, and when one AI provider fails, the next one takes over.
- My role
- Designed, built and run it. AI-assisted build
- Stack
- Python, Flask, Telegram Bot API, several LLM providers, a vision model, a voice-cloning service
- Hosting
- My Oracle server, under its own Linux user. It started on PythonAnywhere
- Running since
- July 2026
How it works
- A message arrivesTelegram delivers it to the bot's webhook: text, a photo, a sticker or a voice note.
- ContextThe bot adds what it remembers about that person and a guide to how I text, so the reply sounds like me.
- The brain ladderIt asks the first AI provider. If that one errors out or hits a limit, the next one answers, then the next. It's the same idea as NetShield-E, applied to AI.
- MediaPhotos and stickers go to a vision model and voice notes are transcribed. Asked to "send a voice note", it answers in my cloned voice. If the voice service is down, the reply arrives as text instead, never as silence.
- Safety and backupsThe model bridge runs as its own Linux user that can't read the bot's chats or keys, with every tool switched off. A daily job copies the bot's memory and settings off-site and keeps the last 30.
Talk to it
Message @MohammedPhase2bot on Telegram. It introduces itself as the AI version of me and answers in English or Egyptian Arabic. If every AI provider is busy, it tells you to try again later.
Conversations are logged on my server so I can improve it, so please don't send anything private.
Its sibling: Rafiq
Rafiq is a private assistant on n8n that handles my tasks, notes and expenses through Telegram. It only answers me, and it runs on the same server.
Live on Telegram
Text the AI version of me
Send it a message, a photo or a voice note. It answers in my style, through whichever AI provider is up.