Markstream

Most Markdown libraries are judged after the last token lands. Markstream pays attention to the awkward part before that, when code blocks, diagrams, formulas, and ordinary prose are still changing on screen. The framework-specific guidance is a real strength for teams that would otherwise have to work out the client boundaries and parser choices themselves.

Markstream progressively renders Markdown while content is still arriving. The documented content types include Mermaid diagrams, KaTeX, code blocks, and streaming diffs.

It provides packages and quick starts for Vue, React, Svelte, Angular, Nuxt, and Next.js. The docs also cover component overrides, custom trusted tags, parser hooks, migration, SSR, and browser-only dependencies.

Markstream makes the most sense for developers whose interfaces cannot wait for a complete response before rendering readable Markdown.

The framework guides may also help teams replacing another renderer or deciding whether to pass raw content or parsed nodes.

The team behind Markstream says it grew out of their own work on chatbots and AI agents, where the interface had to show a response while its Markdown was still incomplete. They say the markstream-vue package recorded 83,318 npm downloads between September 23 and October 22, 2026, and they point to their Vue streaming chat playground, which replays the same Markdown at different speeds so developers can inspect half-finished states.

Against marked and markdown-it, the team describes the difference as a component layer built for Markdown that's still arriving from a model. It handles partial code fences and other unfinished blocks, renders code, math and diagrams progressively, and supports virtualization for long responses.

They're also clear about where it doesn't fit. If you only need to turn a finished document into an HTML string, they suggest marked or markdown-it is simpler. Mermaid, KaTeX and enhanced code blocks rely on optional dependencies, and your app still has to supply the model connection and the stream transport. Their tip for a first run is to start with the basic renderer and its CSS, then test with a real response that stops partway through a code fence or table before adding anything optional.

The interesting part is not that Markstream renders Markdown. Plenty of libraries do that. It is built around the messy few seconds when an AI response or live feed is still taking shape.

This note reflects the public documentation, not a package installation or performance test.