A talking ChatGPT butler for your Telegram chats
This bot wires OpenAI's GPT-4, Whisper, and a text-to-speech engine into Telegram so you can argue with an AI by voice message.

What it does
It’s a Telegram bot that forwards your messages to OpenAI’s GPT-4 and pipes the response back. Send a voice note and it uses Whisper to transcribe, then replies with text or—if you prefer—synthesized speech via Play.ht. LangChainJS handles prompt construction, conversation memory, and Google search integration.
The interesting bit
The voice loop is the hook: Whisper in, GPT-4 thinks, Play.ht speaks back. It’s a complete speech-to-speech pipeline crammed into a Telegram bot, which is arguably the most practical UI for personal AI assistants right now.
Key highlights
- Full-duplex voice: Whisper transcription + Play.ht voice synthesis
- Conversation history managed by LangChainJS
- Optional Google search integration via LangChain
- Single-user lockdown via
SERVE_THIS_USER_ONLYenv var - One-click Railway deploy template
Caveats
- Requires four separate API keys/tokens (OpenAI, Telegram, Play.ht x2) plus ffmpeg
- GPT-4 access is assumed; GPT-3.5-turbo is the fallback
- Play.ht voice selection is manual and requires checking external docs
Verdict
Worth a spin if you want a voice-enabled ChatGPT in your pocket without building a mobile app. Skip it if you’re hoping for group chat support or fine-grained access controls beyond a single-user allowlist.