Introducing Question and Highlights: High-Quality Answers from the Web, 100x Fewer Tokens

Firecrawl has introduced two new formats, "question" and "highlights," to streamline the process of retrieving specific content from web pages using large language models (LLMs). These formats replace the traditional method of scraping full pages, chunking, and running them through a model by offering a more efficient, up to 100x more token-efficient solution. The "question" format provides a concise, grounded answer directly from the page without extraneous content, while the "highlights" format extracts exact sentences, code blocks, and tables verbatim, ideal for compliance and data capture tasks. Both formats run on a managed LLM stack, with built-in protection against prompt injection, ensuring accurate and secure outputs. Firecrawl's enhancements offer significant cost savings and improved efficiency for agents performing numerous lookups by minimizing the amount of data processed, with billing and telemetry integrated into the existing Firecrawl framework.