Web to Markdown for LLMs

Pricing

Rating

Developer

Actor stats

Categories

Share

Convert any URL to clean, structured markdown optimized for LLM consumption. 85% average token savings vs raw HTML. The open-source Firecrawl alternative on Apify.

Why This Actor?

LLMs choke on raw HTML. Scripts, styles, navigation, ads — all noise that burns tokens and confuses models. This actor strips all that away and returns clean markdown that your AI can actually reason about.

Raw HTML:67,841 tokens → costs $0.068 per page(GPT-4)
Markdown:6,176 tokens → costs $0.006 per page(GPT-4)
 ↑ 91% savings

How It Works

┌──────────┐ ┌─────────────────┐ ┌──────────────┐
│ Any URL │────▶│ Puppeteer │────▶│ Clean │
│ │ │ renders page │ │ Markdown │
└──────────┘ │ (JavaScript, │ │ + metadata │
 │ SPAs, dynamic) │ │ + stats │
 └─────────────────┘ └──────────────┘
 │
 ┌───────┴───────┐
 │ Cheerio parses │
 │ Turndown │
 │ converts to MD │
 └───────────────┘
Noise removed: scripts, styles, nav, footer, ads, popups, modals
Kept: headings, paragraphs, lists, tables, links, images, code blocks

What Data Does It Extract?

Use Cases

1. RAG Pipelines — Feed clean web content into vector databases (Pinecone, Weaviate, Chroma). 85% fewer tokens = 85% lower embedding costs.

2. AI Agent Tool Use — Give your agent a "read the web" tool. Pass any URL, get structured content back. Works with LangChain, LlamaIndex, CrewAI, AutoGen.

3. Content Repurposing — Convert any article/blog into markdown for your CMS, newsletter, or documentation site.

4. Training Data — Build LLM training datasets from web content. Clean markdown = higher quality training data.

5. Competitive Intelligence — Monitor competitor websites and extract structured content for analysis.

Input Parameters

*Provide either url or urls

Output Example

{
"url":"https://blog.example.com/article",
"sourceUrl":"https://blog.example.com/article",
"title":"How AI Agents Read the Web",
"description":"A guide to building web-reading capabilities for AI agents",
"author":"Jane Doe",
"publishDate":"2026-03-25T10:00:00.000Z",
"language":"en",
"markdown":"# How AI Agents Read the Web\n\n**Author:** Jane Doe\n**Published:** 2026-03-25\n\n---\n\nAI agents need structured data to reason about web content...",
"wordCount":2450,
"links":[
{"text":"LangChain docs","href":"https://docs.langchain.com"},
{"text":"Vector databases","href":"https://www.pinecone.io"}
],
"images":[
{"src":"https://blog.example.com/diagram.png","alt":"Architecture diagram"}
],
"tableOfContents":[
{"level":1,"text":"How AI Agents Read the Web"},
{"level":2,"text":"The Problem with Raw HTML"},
{"level":2,"text":"The Markdown Solution"}
],
"stats":{
"htmlSize":245000,
"markdownSize":12400,
"htmlTokensEstimate":61250,
"markdownTokensEstimate":3100,
"tokenSavingsPercent":95,
"renderTimeMs":4200
}
}

Performance Benchmarks

Tested across 60 diverse websites:

Comparison vs Firecrawl

Standby API (Instant Response)

This actor supports Apify Standby mode for instant HTTP responses:

# Health check
curl"https://george-the-developer--web-to-markdown-llm.apify.actor/"\
-H"Authorization: Bearer YOUR_TOKEN"
# Convert a URL
curl"https://george-the-developer--web-to-markdown-llm.apify.actor/convert?url=https://example.com"\
-H"Authorization: Bearer YOUR_TOKEN"

Pricing

Pay Per Event: $0.003 per page converted

No monthly subscription. Pay only for what you use.

Integrations

Works with any tool that can call an HTTP API:

• LangChain: Use as a custom tool in your agent chain
• LlamaIndex: Feed markdown into document loaders
• n8n / Make: HTTP request node → markdown output
• Python: requests.get() → JSON with markdown
• Node.js: fetch() → structured response

FAQ

Q: Does it handle JavaScript-rendered pages? A: Yes. Uses Puppeteer with full Chrome to render JavaScript, SPAs, and dynamic content.

Q: What about pages behind logins? A: Currently extracts public content only. Authenticated scraping is on the roadmap.

Q: How accurate is the token estimate? A: Uses the ~4 chars/token heuristic for English text. Actual token counts may vary by model.

Q: Can I process multiple URLs at once? A: Yes. Pass an urls array in batch mode for multiple pages.

Support

• GitHub: the-ai-entrepreneur-ai-hub
• Apify Store: george.the.developer
• Twitter: @ai_in_it

Found a bug? Open an issue or DM on Twitter.