Reader is a web-to-markdown conversion service that transforms any URL into LLM-friendly input through a simple proxy prefix.
Reader solves the problem of feeding raw web content to language models and retrieval-augmented generation systems, which struggle with HTML formatting, ads, and irrelevant page elements. It works by accepting a URL prefixed with https://r.jina.ai/ and returning cleaned, markdown-formatted content optimized for LLM consumption. The service also includes a search capability via https://s.jina.ai/ that lets language models query the web for current information. The tool intelligently selects between headless Chrome rendering for JavaScript-heavy sites and lightweight curl-based fetching for simpler pages. Beyond web pages, it handles PDFs via PDF.js, Microsoft Office documents through LibreOffice conversion, and images through vision-language model captioning.
Teams building AI agents, RAG systems, or applications that need to feed web content to language models should consider this tool. It eliminates the need to build custom web scraping and content cleaning pipelines. The hosted service at r.jina.ai is free and production-ready with documented rate limits, making it suitable for both prototyping and deployed systems. The open-source branch runs locally in stateless mode with optional S3-compatible bucket caching via Docker Compose, giving teams the flexibility to self-host if needed. The README does not compare Reader to alternative solutions.
The project maintains active synchronization between its open-source branch and the production SaaS codebase, with recent updates decoupling the storage layer and adding support for direct file uploads of PDFs and Office documents. Development has evolved from Firebase-backed infrastructure to a Cloud Run deployment with MongoDB, reflecting a shift toward more portable and self-hostable architecture. The tool is positioned as a core product of Jina AI with explicit commitments to stability and ongoing maintenance.
