llm-scraper

Why we included this project

Writing scrapers the old way means maintaining CSS selectors that break whenever a site changes its markup. This library flips that: you define the shape of the data you want as a Zod or JSON Schema, and it drives a real Playwright browser to the page, feeds the rendered content to an LLM, and hands back objects validated against your schema. It works across providers, from OpenAI and Anthropic to local models through Ollama, and you can feed the model pre-processed HTML, markdown, extracted text, or a screenshot for multimodal models. Streaming output helps on large pages, and the code-generation feature can emit reusable Playwright scripts that skip the LLM call for repeated extractions. For teams that regularly turn web content into JSON, whether for product catalogs or research datasets, that saves a lot of brittle selector-writing.