What it does TeleChat is a Python Telegram bot that pipes OpenAI-compatible LLMs — and a laundry list of others via the author’s companion uni-api project — into group and private chats. It handles text, voice, images, PDFs, and even ArXiv papers, with plugins for web search (DuckDuckGo or Google), URL summarization, and a code interpreter. Deployment is geared toward zero-cost hosts: Koyeb, Zeabur, Replit, plus Docker and fly.io.

The interesting bit The model grouping system is a nice touch of UX in an otherwise config-heavy space. You can bucket models by provider or capability with a semicolon-delimited env string, and stragglers land in an auto-generated “OTHERS” group. The author also built a custom Markdown-to-Telegram renderer (md2tgmd) to avoid the usual formatting carnage.

Key highlights

• Supports multimodal Q&A: voice, images, PDF/TXT/MD/Python uploads
• Per-topic isolation in group chats (separate API keys, history, plugins)
• Inline mode: @ the bot in any chat without switching windows
• Auto-split and auto-merge for Telegram’s message length limits
• Config persistence via mounted CONFIG_DIR; multi-user or global mode
• Whitelist/blacklist/admin controls for access management

Caveats

• The README claims support for “GPT-5” and “Claude 4.1” as defaults, which appear to be speculative or placeholder names; verify actual model availability
• Non-OpenAI providers (Anthropic, Gemini, Groq, etc.) require spinning up the separate uni-api project — this bot alone only speaks OpenAI-format APIs
• Plugin system is all-or-nothing via env flags; no per-user plugin toggles mentioned

Verdict Worth a look if you need a multi-user, multi-model Telegram bot with document support and don’t mind some assembly with uni-api for non-OpenAI backends. Skip it if you want a managed SaaS or native provider SDKs without middleware.