ScrapeGraphAI

Rather than writing CSS selectors, you give ScrapeGraphAI a URL or a local file plus a plain-English request, and a language model reads the fetched content and hands back structured JSON. Under the hood each job is a graph of nodes for fetching, parsing, chunking, and prompting. Ready-made pipelines cover a single page (SmartScraperGraph), answers gathered from search results (SearchGraph), batches of sources processed in parallel, generation of a reusable Python scraper (ScriptCreatorGraph), and narration of results as audio (SpeechGraph). The library sits on LangChain and drives Playwright for JavaScript-heavy pages, so setup is pip install scrapegraphai followed by playwright install. Models can come from OpenAI, Azure, Gemini, Groq, Mistral, AWS Bedrock, or a local Ollama server for fully on-machine runs. The MIT license permits commercial use. Anonymous telemetry is on by default; set SCRAPEGRAPHAI_TELEMETRY_ENABLED=false to turn it off. The same team sells a hosted API with proxies and crawling, but the open library runs without it. Suits data engineers and RAG builders tired of brittle parsers.