Self-healing markdown. Intelligently parses and styles incomplete Markdown blocks.
Overview
Remend is a lightweight utility that handles incomplete Markdown syntax during streaming. When AI models stream Markdown token-by-token, you often get partial formatting markers like unclosed **bold** or incomplete [links](. Remend automatically completes these unterminated blocks so they render correctly in real-time.
Remend powers the markdown termination logic in Streamdown and can be used standalone in any streaming Markdown application.
Inline math: $formula → $formula$ (opt-in, see inlineKatex)
Single tilde escape: 20~25 → 20\~25 (prevents false strikethrough)
Installation
npm i remend
Usage
import remend from "remend";// During streamingconst partialMarkdown = "This is **bold text";const completed = remend(partialMarkdown);// Result: "This is **bold text**"// With incomplete linkconst partialLink = "Check out [this link](https://exampl";const completed = remend(partialLink);// Result: "Check out [this link](streamdown:incomplete-link)"
Configuration
You can selectively disable specific completions by passing an options object. Options default to true unless noted otherwise:
import remend from "remend";// Disable link and KaTeX completionconst completed = remend(partialMarkdown, { links: false, katex: false,});
Available options:
Option
Description
links
Complete incomplete links
images
Complete incomplete images
bold
Complete bold formatting (**)
italic
Complete italic formatting (* and _)
boldItalic
Complete bold-italic formatting (***)
inlineCode
Complete inline code formatting (`)
singleTilde
Escape single ~ between word characters to prevent false strikethrough (e.g. 20~25)
strikethrough
Complete strikethrough formatting (~~)
katex
Complete block KaTeX math ()
Custom Handlers
You can extend remend with custom handlers to complete your own markers during streaming. This is useful for custom syntax like <<<JOKE>>> blocks or other domain-specific patterns.
import remend, { type RemendHandler } from "remend";const jokeHandler: RemendHandler = { name: "joke", handle: (text) => { // Complete <<<JOKE>>> marks that aren't closed const match = text.match(/<<<JOKE>>>([^<]*)$/); if (match && !text.endsWith("<<</JOKE>>>")) { return `${text}<<</JOKE>>>`; } return text; }, priority: 80, // Runs after most built-ins (0-70)};const
Built-in handlers use priorities 0-75. Custom handlers default to 100 (run after built-ins):
Handler
Priority
singleTilde
0
comparisonOperators
5
htmlTags
10
setextHeadings
15
links
20
boldItalic
30
bold
35
italic
40-42
inlineCode
50
strikethrough
60
katex
70
Exported Utilities
Remend exports utility functions for context detection in custom handlers:
import { isWithinCodeBlock, isWithinMathBlock, isWithinLinkOrImageUrl, isWordChar,} from "remend";const handler: RemendHandler = { name: "custom", handle: (text) => { // Skip if we're inside a code block if (isWithinCodeBlock(text, text.length - 1)) { return text; } // Your logic here return text; },};
Usage with Remark
Remend is a preprocessor that must be run on the raw Markdown string before passing it into the unified/remark processing pipeline:
import remend from "remend";import { unified } from "unified";import remarkParse from "remark-parse";import remarkRehype from "remark-rehype";import rehypeStringify from "rehype-stringify";const streamedMarkdown = "This is **incomplete bold";// Run Remend first to complete incomplete syntaxconst completedMarkdown = remend(streamedMarkdown);// Then process with unifiedconst file = await unified() .use(remarkParse) .use(remarkRehype) .use(rehypeStringify)
This is important because Remend operates on the raw string level, while remark/unified work with abstract syntax trees (ASTs). Running Remend after parsing would be ineffective.
How It Works
Remend analyzes the input text and:
Detects incomplete formatting markers at the end of the text
Counts opening vs closing markers (considering escaped characters)
Intelligently adds closing markers when needed
Respects context like code blocks, math blocks, and list items
Handles edge cases like nested brackets and word-internal characters
The parser is designed to be defensive and only completes formatting when it's unambiguous that the block is incomplete.
Performance
Remend is built for high-performance streaming scenarios: