Agentic scrapers now support URL-typed parameters (type: "url") — assigned automatically to params named url or ending in _url. The legacy top-level url field is still accepted and is auto-promoted into parameters for back-compat. See Custom Parameters.
Agentic runs are now single-URL. Each agentic run handles one URL per request — to scrape multiple URLs, call the endpoint once per URL. Multi-URL fan-out via url: [...] remains supported by extractor scrapers.
March 2026
API restructured into three sub-routers: /v1/extractor/, /v1/agent/, /v1/scrapers/
New Agent API for AI-powered one-shot extraction and reusable agentic scrapers
Asynchronous scraper execution with run_async and polling
Code generation is now asynchronous (returns 202)
Unified GET /v1/scrapers lists both extractor and agentic scrapers
Old scraper CRUD endpoints deprecated (still functional with _warning)
scraper_id renamed to template_id (scraper_id still accepted)
Callback support (callback_url) for async runs and agent extraction
