teLLegraM

teLLegraM is a library designed to format LLM (Large Language Model) generated text into Telegram-specific-markdown (MarkdownV2), based on Unified and Remark. It ensures that complex markdown from AI responses is perfectly interpreted by Telegram clients. teLLegraM also supports a syntax-aware and lossless Markdown pagination.

Acknowledgements

This project is based on telegramify-markdown but has been evolved to specifically address the needs of LLM-generated content.

Why teLLegraM?

While the original library provided a solid foundation, teLLegraM introduces several key optimizations for the "LLM to Telegram" workflow:

1. LLM-Specific Optimizations: Tailored handling of common artifacts found in AI responses, ensuring cleaner output.
2. Lossless Pagination: Telegram has strict message length limits. teLLegraM includes a smart pagination feature that splits long text into multiple messages without breaking MarkdownV2 syntax. It ensures bold, italic, or code blocks are correctly closed in one message and reopened in the next, preventing "unclosed entity" errors.
3. Strict MarkdownV2 Compliance: Enhanced escaping rules to handle edge cases often produced by generative models.

Install

npm install tellegram

Usage

Basic Conversion

import { convert } from 'tellegram';

const markdown = `
# Header
## Subheader

[1.0.0](http://version.com)

* item 1
* item 2
* item 3

And simple text with + some - symbols.
`;

const result = convert(markdown);
console.log(result);
/*
 *Header*
 *Subheader*

[1\.0\.0](http://version.com)

 • item 1
 • item 2
 • item 3

And simple text with \+ some \- symbols\.
*/

Pagination (Handling Long Messages)

When dealing with verbose LLM outputs, use the paginate function to safely split text into chunks that respect Telegram's limits (4096 characters) while preserving formatting context.

import { paginate } from 'tellegram';

const longLlmOutput = `... extremely long text with **markdown** ...`;

// Split into an array of strings, each safe to send
const messages = paginate(longLlmOutput);

for (const msg of messages) {
    // Send each part sequentially
    await bot.sendMessage(chatId, msg, { parse_mode: 'MarkdownV2' });
}

Possible options

You can also add unsupported tags strategy as a second argument, which can be one of the following:

• escape - escape unsupported symbols for unsupported tags (default)
• remove - remove unsupported tags
• keep - ignore unsupported tags

convert also accepts a third argument for feature options:

• table: 'list' - convert GFM tables into MarkdownV2-safe hierarchical lists (default)
• table: 'unsupported' - treat tables as unsupported and apply the second-argument strategy

import { convert } from 'tellegram';
const markdown = `
# Header

> Blockquote

<div>Text in div</div>
`;

convert(markdown, 'escape');
/*
*Header*

\> Blockquote

<div\>Text in div</div\>
*/

convert(markdown, 'remove');
/*
*Header*
 */

Convert tables to list

Telegram does not support Markdown tables. By default, teLLegraM converts a table into a vertical hierarchical list using one numbered title line per row plus - lines for the remaining fields. Number markers and - markers are escaped to stay safe for Telegram MarkdownV2.

import { convert } from 'tellegram';

const markdown = `
| Name  | Role  | Score |
| ----- | ----- | ----- |
| Alice | Admin | 95    |
| Bob   | User  | 88    |
`;

convert(markdown, 'escape');
/*
1\.  Name: Alice
•  Role: Admin
•  Score: 95

2\.  Name: Bob
•  Role: User
•  Score: 88
*/

If you want the previous behavior (table treated as unsupported), use table: 'unsupported' with the strategy you need:

convert(markdown, 'escape', { table: 'unsupported' });
// => table is kept as escaped Markdown table text

convert(markdown, 'remove', { table: 'unsupported' });
// => ''

MIT Licence