AgenticSearch

Discover structured entities from the web with per-cell evidence

Examples:

Every cell links to a source URL, evidence snippet, and confidence score.

How it works

  1. Normalize & Type — Clean the query, classify its family, and choose a constrained schema
  2. Search & Scrape — Multi-angle web search via Brave, async scraping with caching
  3. Rerank — Cross-encoder selects the most query-relevant pages for extraction
  4. Discover — Extract broad candidate entities first, then fill missing attributes later
  5. Canonicalize — Merge duplicates and prefer official or canonical sources when they can be resolved
  6. Rank & Verify — Score by evidence, source quality, support, and diversity; prune obvious junk late