Simple translate bot for Telegram

A request was submitted to our studio to create a simple Telegram bot that uses an LLM to translate not only text sent by users but also messages from users and Telegram channels. A special requirement was the ability to translate Telegraph articles embedded as links, rapid deployment on a server, as well as privacy protection for other users.

The project was developed in just a few hours using Qwen3.8-27B-GGUF as the main code agent. Through extensive experimentation with local models, our team concluded that modern solutions excel at handling small coding tasks, significantly reducing development costs.

How it works:

1. You send text, a link, or a forwarded message.
2. The bot extracts the content and identifies links to Telegraph.
3. A request is sent to the LLM via an OpenAI-compatible API.
4. The translated text is split into sections for easier reading.
5. You receive the result directly in Telegram.

Features:

Multilingual translation

Support for 8 languages: English, Russian, Spanish, French, German, Japanese, Korean, and Chinese. Simply select a language pair, and the bot will translate any text.

Smart message analysis

• Direct translation — processing regular text.
• Forwarded messages — the bot recognizes messages forwarded from users and channels.
• Caption content — translation of captions for media files.
• Integration with Telegraph
• Automatic extraction and translation of articles.
• Support for Telegraph links in the format telegra.ph and t.me/telegraph: the bot extracts the full article text, removes HTML tags, and translates it.

Response splitting

Long translations are automatically divided into multiple messages to stay within Telegram’s 4,000 character limit.

Access control

With the ADMIN_CHAT_ID setting, the bot will respond only to requests from an authorized user.

Two operating modes

• Local providers (Ollama, llama.cpp, vLLM).
• Cloud providers.

After discussing the work with the client and obtaining their approval, we made the project available in our public GitHub repository under the MIT license.

Here is the link to the project's GitHub