PhoenixStreamdown

Streaming markdown renderer for Phoenix LiveView, optimized for LLM output.

Inspired by Streamdown and vue-stream-markdown, but built to fully leverage LiveView — server-side rendering, automatic DOM diffing, and zero client-side JavaScript.

Installation

def deps do
  [{:phoenix_streamdown, "~> 1.0.0-beta"}]
end

Usage

use PhoenixStreamdown

<.markdown content={@response} streaming={@streaming?} />

That's it. Pass content as a markdown string, set streaming to true while tokens are arriving. Completed blocks are frozen with phx-update="ignore" — only the last block re-renders on each token.

How it works

1. Remend — auto-closes incomplete syntax (**bold → **bold**, unclosed fences, partial links)
2. Blocks — splits into independent blocks so earlier ones are stable
3. MDEx — renders each block to HTML server-side (Rust-backed)
4. LiveView — diffs only the active block, skips the rest

On a 56-block document, this is ~7x less server work and ~460x smaller diffs per token compared to re-rendering the full document each time.

Example

A full chat app with ReqLLM streaming is in the example/ directory. Run it:

cd example
cp .env.example .env  # add your OpenRouter API key
mix setup
mix phx.server

Documentation

HexDocs — attributes, customization (themes, CSS classes, stable IDs, MDEx options).

License

MIT