Every AI chat app re-parses the entire markdown on each token—O(n²) slowdown. After 500 tokens, that’s 500 full re-renders. StreamMD fixes this with incremental block parsing: only new text is processed, completed blocks are memoized (no re-render), and only the active block updates. Result: ~300x less work. Built-in syntax highlighting (15 langs, ~3kB, zero deps). Drop-in:

AppSignal — Get full visibility into app health, errors, and performance

Get full visibility into app health, errors, and performance

Hey Product Hunt! 👋 I'm Jai. I built StreamMD because I noticed every AI chat app has the same hidden performance bug: they re-parse the entire markdown string on every single token. After 500 tokens, you're re-parsing a 2,000+ character string 500 times. The total characters processed grows QUADRATICALLY. At 100 tok/s, that's insane. StreamMD's incremental parser only processes the NEW characters. Completed blocks (headings, code fences, paragraphs) are frozen via React.memo and never re-render. Best way to see it: https://altrusian.com/stream-md Watch the "Chars parsed" counter — naive hits 400K+ while StreamMD stays at ~1.3K for the same document. It's the companion to ZeroJitter (our canvas renderer). Together they own streaming LLM display. Would love your feedback! 🙏

AppSignal — Get full visibility into app health, errors, and performance

Get full visibility into app health, errors, and performance

StreamMD

Streaming MD for LLMs. 300x fewer chars parsed per token

Streaming MD for LLMs. 300x fewer chars parsed per token