Headless content rendering

Single-page applications (SPAs, such as those built with React) require JavaScript to render the content of each page — a process known as DOM hydration. A direct curl or fetch of these websites will just leave you with an empty shell of HTML.

Fetching through pure.md, on the other hand, hydrates the DOM of SPAs in the background so that pages render completely.

Similarly, PDFs are parsed as pure markdown automatically. Images run through AI models for object detection and summarization. Excel and Numbers spreadsheet documents can also be converted into markdown.

Markdown written for LLMs

Cut your inference costs and speed up your agents' workflows. Powered by HTMLRewriter, our URL-to-markdown service is optimized for low latency and low token output. We remove superfluous fluff from web pages — while also adding page metadata as frontmatter — so that LLMs have the most context in the fewest characters possible.

           |
 r.jina.ai |░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░  143K tokens
           |
           |
tavily.com |░░░░░░░░░░░░░░░ 55K tokens
           |
           |
   pure.md |████████ 28K tokens ✅
           |_______̩_______̩_______̩_______̩_______̩_______̩_______̩___

                 25K           75K           125K          175K
        

           |
 r.jina.ai | ❌ Verifying you are human...
           |
           |
tavily.com | ❌ Failed to fetch content
           |
           |
   pure.md |█████ 22K tokens ✅
           |_______̩_______̩_______̩_______̩_______̩_______̩_______̩___

                 25K           75K           125K          175K
        

Knowledge in real-time

Make your AI apps aware of recent events. With our built-in search engine result page (SERP) crawling, you can turn user queries into a concatenated markdown string of answers that you can feed directly into your prompts. Your model will think it was trained yesterday.

Inference when you need it

Extract data from any page or search simply by changing from GET to POST. pure.md offers a selection of generative AI models for extracting structured or unstructured data from web pages. Stream back responses in markdown for tasks like summarization, or generate JSON that conforms to a custom schema.

POST https://pure.md/reuters.com

{
  "prompt": "What are the top 5 headlines from today?",
  "model": "meta/llama-3.1-8b",
  "schema": {
    "type": "object",
    "properties": {
      "headlines": {
        "type": "array",
        "items": {"type": "string"}
      }
    },
    "required": ["headlines"]
  }
}

Works on select social media sites

Prefix social media URLs with `pure.md/` like you normally would, and it will just work. Behind the scenes, pure.md transparently pulls data from data enrichment providers.

Pricing as a feature

Simple, easy-to-understand pricing for projects of any size. All plans are available for commercial use. Pay for what you need, and cancel at any time.