Web Page to Markdown for LLM — Fast & Cheap, CJK-ready

Convert a list of web pages into clean Markdown, plain text or RAG chunks — without a browser, so it is fast and costs $1 per 1,000 pages.

Built to get Japanese, Chinese and Korean pages right:

• No mojibake: legacy encodings like Shift_JIS, EUC-JP, GBK/GB18030, Big5, EUC-KR are detected and decoded correctly
• Furigana (ruby): remove readings (漢字) or keep them in parentheses (漢字(かんじ))
• No stray spaces between CJK characters at line breaks
• Sentence-aware RAG chunks that end at 。！？ as well as . ! ?

And for every language:

• Main content extraction (Readability): navigation, ads, cookie bars and footers removed
• Tables kept as Markdown tables
• Metadata: title, description, site name, author, published time, canonical URL, language, charset
• Polite by default: respects robots.txt and Crawl-delay, waits between requests to the same site, never crawls beyond the URLs you give

What can it be used for?

• Feed web pages to ChatGPT, Claude or any LLM
• Build a RAG / vector database from documentation, news or knowledge-base pages
• Archive or translate Japanese / Chinese / Korean web content without garbled characters
• Use it from AI agents via the Apify API or MCP as a cheap "read this URL" tool

Input

Field	Description
urls	Pages to convert. Only these URLs are fetched — no link following.
outputFormat	markdown (default), text, or chunks
mode	article (main content, default) or fullPage
ruby	remove (default) or parentheses
includeLinks	Keep hyperlinks in Markdown (default: off, cleaner for LLMs)
chunkSize / chunkOverlap	For chunks
respectRobotsTxt	Default on
minDelayPerSiteMs, maxConcurrency, timeoutSecs, maxPageSizeMb	Politeness and limits

{
    "urls": ["https://www.aozora.gr.jp/cards/000148/files/773_14560.html"],
    "ruby": "parentheses"
}

Output

{
    "url": "https://www.aozora.gr.jp/cards/000148/files/773_14560.html",
    "status": "success",
    "title": "夏目漱石 こころ",
    "language": "ja",
    "charset": "shift_jis",
    "extraction": "article",
    "charCount": 162544,
    "markdown": "### 上　先生と私\n\n#### 一\n\n　私(わたくし)はその人を常に先生と呼んでいた。…"
}

status is success, failed (with error), or blocked_by_robots_txt. With outputFormat: "chunks" you get chunks: [{ index, charCount, text }] and chunkCount instead of markdown.

Pricing

Event	Price
Page converted (per successfully converted page)	$0.001 ($1 per 1,000 pages)

Failed pages, pages blocked by robots.txt, PDFs and non-HTML URLs are not charged. Platform usage is included.

Limitations

• No JavaScript rendering. Pages that build their content in the browser (some single-page apps) may return little text — the item then fails with an explanation and is not charged.
• Pages behind logins or paywalls are not supported.
• For PDFs use PDF to Markdown for LLM & RAG; for images and scans use OCR Japanese, Chinese, Korean & English.

Responsible use

Only submit pages you are allowed to process. The Actor identifies itself in the User-Agent (ApifyWebToMarkdownCJK), honours robots.txt by default and spaces out requests to each site.