The product

apps/research-tool is a Next.js App Router app you start with pnpm dev. You create a business from a name and a domain, start a pipeline run, and review the result in tabs: Profile, Overview, Claims, Curation, Captures, Raw Data and Notes.

A settings page shows each source’s status, so you can see which keys are missing before you run.

Each source is served by exactly one connector in packages/connectors. A connector never throws for a missing key: it returns not_configured. Blocked, rate-limited and failed calls get their own source status, so you know what to do next.

Paid sources stay off until you opt in, and they stop at a call ceiling you set.

packages/capture drives a real, locally installed Google Chrome. If a page shows a bot challenge, the capture stops with blocked: no stealth plugins, no proxy rotation, no CAPTCHA solving.

For a blocked page there is a manual capture path: HTML a person saved from their own browser goes through the same extractor and gives the same claims.

packages/resolve turns the claims for each field path into one resolved value. Your curation decision wins; otherwise the claim with the highest confidence wins; a tie between disagreeing claims stays an unresolved conflict for you.

packages/schema defines the file formats with zod. packages/store writes them as deterministic JSON, redacts key- and token-shaped strings before anything reaches disk, and locks a business while a pipeline run writes to it.