Expand Fetch

Not yet independently verified. The launch date rests on the company's own Y Combinator launch post of 12 October 2026. The benchmark figures and the founder background come from the maker's launch post only. Other products called Expand exist; this listing is the web data API at expand.ai. No independent report was found. We will update this when it can be confirmed, and remove this note.

Expand Fetch is a web data API from Expand, the name used on its site for ExpandAI, Inc., a Delaware corporation with an address in San Francisco. A developer sends a URL and receives Markdown produced by the company's own vision model, called Iris, which reads the rendered page rather than applying fixed cleaning rules. The founder, Tim Suchanek, writes that the product came from building research agents that lost footers, tables and pages that never loaded. The launch post says he previously worked at Prisma and founded Stellate.

What it does

Iris returns three things for a page: Markdown that keeps tables and icons, any useful JSON embedded in the page, and an appendix holding navigation, footers and similar page chrome. Each fetch is stored as a snapshot in which lines link back to their position on the page, so an answer can be checked against its source. A query option returns only the passages that answer a stated question, and the site offers Fetch through an API, SDK, command line tool and MCP server. The company handles JavaScript rendering, proxies and bot detection for each request. It also converts PDFs and other documents, reads scanned pages and images with a vision model, and transcribes YouTube video and audio. Its benchmark figures, including 74 percent of questions answered from its Markdown against 55 percent for the next API, are the maker's own and were not independently checked.

Who it is for

The product is meant for engineers building AI agents and research tools that need to read web pages reliably, and for teams that want to cite exactly where an agent found an answer. The site names Mixedbread among the engineers who use it.

Key facts

• Maker: Expand (ExpandAI, Inc.), San Francisco
• Pricing: one credit per successful fetch, 1,000 credits for $1; failed, blocked and empty requests cost nothing
• Free tier: 10,000 credits worth $10, with 2 concurrent requests and 7 days of storage
• Extra credits: Highlights one credit per request, PDFs one credit a page, scanned pages and images five credits a page, video transcripts five credits, audio ten credits a minute
• Plans: Free, pay as you go with prepaid credits that do not expire, and Enterprise on request
• Backers named: Y Combinator, Conviction, Pioneer Fund, system.one and basecase
• Questions to ask: the benchmark method and test pages, data retention for stored snapshots, and rate limits