About Dataembed

Dataembed searches the web like a database. You describe the list you want in plain English, and an agent builds it for you as a table: real companies, people, papers, articles or posts, each one checked against your conditions and backed by a quote from the page it came from.

It exists because search engines return pages, not answers to set-shaped questions. "Every seed-stage climate startup in Europe with a technical founder" is a filter over the web, and doing it by hand means dozens of tabs and a spreadsheet you can't trust. Dataembed does the tab-opening for you and shows its work.

How a webset gets built

• Plan — the agent turns your description into search queries, yes/no criteria and enrichment columns you can edit
• Search — each query fans out across Exa, Parallel, Firecrawl and search-grounded models via the Vercel AI Gateway, and the ranked lists are merged
• Verify — every candidate page is fetched and judged against each criterion, with a verbatim evidence quote per verdict
• Export — rows stream into the grid as they finish; download the webset as CSV or JSON, or read it over the API

Open source

Dataembed is MIT licensed. The whole product — the Next.js app, the agent, the search pipeline — is one repository you can run on your own Postgres with a single AI Gateway key. The hosted version at dataembed.com is the same code.

Who makes it

Dataembed is built by Kaiyu Hsu, an independent software engineer. It is a small project, not a company with a sales team: questions and bug reports reach the person who wrote the code.

• Email — kai@kyh.io
• GitHub issues — bugs, feature requests and self-hosting questions, answered in public
• X — @kaiyuhsu