ChatGPT's Web UI Is Laggy. This Telegram Bot Isn't.

What it does This Python bot proxies OpenAI’s models—GPT-4, GPT-4 Turbo, GPT-4 Vision, DALL-E 2, and Whisper—into Telegram chats. It handles text, voice messages, and image generation, while letting owners restrict access to specific users and track API spend. Custom personas like Code Assistant or Artist are loaded through YAML config files.

The interesting bit Instead of buffering the full reply, it streams tokens into Telegram message bubbles as they arrive. It also supports group chats and can be repointed to OpenAI-compatible APIs such as LocalAI, so you’re not strictly locked to OpenAI’s servers.

Key highlights

• Streams partial GPT-4 replies live rather than waiting for the full completion.
• Bundles DALL-E 2 image generation, Whisper voice-to-text, and syntax highlighting for code blocks.
• Group chat support, user allowlists, and per-user balance tracking for API costs.
• Chat modes (e.g., Psychologist, SQL Assistant) defined in config/chat_modes.yml.
• Configurable API base URL for self-hosted or alternative backends like LocalAI.

Caveats

• The repo description mentions Claude, but the README only documents OpenAI models; Anthropic support is either absent or undocumented.

Verdict This is essentially a polished wrapper around OpenAI’s API, ideal if you want a private ChatGPT instance in Telegram without browser tabs or rate limits. Skip it if you need deep custom logic beyond pre-set chat modes and API proxying.