Parse Sequencial MD Answers from ChatGpt

[dprutean]

I'm using Flexmark to parse Markdown responses from ChatGPT and render them as HTML in a JavaFX application. ChatGPT may return the response all at once, or as a sequence of chunks (streamed tokens). This streaming behavior introduces a challenge when parsing incomplete Markdown constructs.

For example, consider the following input where the Markdown is received in chunks (delimited here by | for clarity - can be in the middle of a text block):

## My | Suggestion
Your query | can __be |improved__ by adding one more table.

If I parse each chunk individually, Flexmark treats '## My' and 'Answer' as separate text nodes instead of one heading, and similarly fails to recognize the emphasis syntax across chunks.

My questions:
Is there a way in Flexmark to detect whether the currently parsed content represents a complete block (e.g., a full heading or paragraph)?

How can I implement a buffering mechanism, collecting incoming chunks until a block is likely complete before passing them JavaFX WebView to render?

I’d appreciate any guidance or best practices for working with incremental Markdown input using Flexmark.