Markdown Converter

The one setting that matters

For list and detail pages, use:

{ "content_source": "cleaned_html", "content_filter": "none" }

In the app: Everything on the page.

The filter pruning (and content_source: fit_html) removes repetitive content, which is exactly what a list of items is. It can turn a whole page into 30 characters.

Tip

Before blaming the model, check the length of this node’s output.

Options

Option	Meaning
content_source	Which HTML to convert (default cleaned_html)
preserve_links, preserve_tables, preserve_code	Keep these elements
ignore_images, skip_internal_links, escape_html	Leave things out
body_width	Wrap width; 0 disables wrapping
citations	Also produce numbered references
content_filter	none, pruning, BM25, or llm filtering for fit_markdown
user_query, bm25_threshold	Query and score for BM25
pruning_threshold, pruning_threshold_type, min_word_threshold	Pruning tuning
llm_filter_instruction, llm_filter_model	LLM filter

Input, output, and verification

Consumes fetched raw_pages and produces pages containing Markdown and conversion metadata. Conversion is a separate step from fetching and LLM processing. Inspect a sample page and its text length before extraction. An empty source or aggressive filter can leave nothing useful for the model; a successful conversion does not guarantee useful text.

The example above avoids pruning repeated list items. It is a recommendation for directories, not a statement about every option’s default. See declared defaults and nested Markdown options, then connect LLM Processor.