Getting started

Install flowdown, render a document, stream a reply and teach the model the dialect.

1. Install the package
   React 18 or 19 is the only required peer.
2. Import the stylesheet once
   It holds the layout of every form and the light and dark colours.
3. Render Markdown
   <Markdown> takes the text as its child.
4. Stream a reply
   Pass streaming while the text is still arriving.
5. Teach the model
   Add the dialect's instructions to the system prompt.

1.Install#

bun add @localix/flowdown

npm install @localix/flowdown

pnpm add @localix/flowdown

yarn add @localix/flowdown

The package is ESM only and ships its own types. It runs in any bundler that resolves the exports field (Vite, Next.js, webpack 5, Rspack, esbuild, Bun) and in Node 20 or later.

2.Import the stylesheet#

import '@localix/flowdown/styles.css'

Import it once, in your app's entry or root layout. Every rule is scoped to the .flowdown element the renderer draws, so it touches nothing else on the page. Colours, fonts and radii are tokens you can set; Styling lists them.

3.Render#

import { Markdown } from '@localix/flowdown'

export function Doc({ source }: { source: string }) {
  return <Markdown>{source}</Markdown>
}

That is GitHub Markdown — tables, task lists, strikethrough, footnotes, autolinks — plus the dialect, with raw HTML parsed and sanitized.

4.Stream a reply#

While a reply is arriving, pass streaming:

<Markdown streaming={status === 'streaming'}>{text}</Markdown>

• Only the block the new text lands in is parsed again; everything above it is kept.
• The unfinished end is completed so it renders the way it will when finished: **bo is bold, an open code fence is code, a half-typed link is its text.
• A caret follows the text. Turn it off with caret={false}.

Pass the whole text each time, not the new part. When the reply is done, set streaming to false: the final text is rendered as written, with nothing completed. Streaming explains the details and shows how to wire it to the Vercel AI SDK.

5.Teach the model#

A model writes the forms it was told about and no others. Add the dialect's instructions to your system prompt:

import { answerFormatPrompt } from '@localix/flowdown/prompt'

const system = ['You are a helpful assistant.', answerFormatPrompt()].join('\n\n')

Math and diagrams are left out of the prompt by default, as they are off in <Markdown> by default: turn on answerFormatPrompt({ math: true, diagrams: true }) together with the page. Prompting shows the full text and how to adapt it.

Optional: math, diagrams and highlighting#

They are off by default, so a page that does not use them does not ship them:

import { Markdown } from '@localix/flowdown'
import { highlight } from '@localix/flowdown/highlight'
import 'katex/dist/katex.min.css'

;<Markdown
  math={() => import('@localix/flowdown/math')}
  mermaid={() => import('mermaid')}
  highlight={highlight}
>
  {text}
</Markdown>

Install the engines you use: katex, mermaid, highlight.js. See Math, Diagrams and Highlighting.