Expand launches Fetch URL-to-Markdown API for AI agents

What did Expand announce with Fetch?

Expand announced Fetch, a URL-to-Markdown API for AI agents and developers that the company says is live now and priced at $1 per 1,000 fetches. The YC launch post attributes the product to founder Tim Suchanek and describes Fetch as the company’s core web-reading product.

The launch post says a single Fetch call returns Markdown written by Expand’s Iris model, along with hidden JSON and a citable snapshot. Expand’s website says the same API returns three artifacts from one URL: page.md, state.json and a snapshot, and that the service can also crawl sites and monitor changes.

How does Fetch turn web pages into agent-ready data?

Expand says Fetch uses its Iris vision model to inspect rendered webpages, classify page elements and write the main content as Markdown while preserving tables. Navigation and footer material is placed in an appendix, which is meant to keep the primary output cleaner for downstream agent use.

The company says each fetched line links back to its location on the source page, and every fetch includes a Wayback Machine snapshot. Expand also says the system handles JavaScript rendering, residential proxies and bot protection automatically on every request.

Beyond standard webpages, Expand says Fetch can convert PDFs, images, office files, ebooks and data files into Markdown. The company’s pricing page says PDF conversion costs one credit per page on top of the Fetch, while office files, ebooks and data files convert locally at no extra charge. Scanned pages and images incur a five-credit-per-page vision-model surcharge, and YouTube URLs return transcript, timestamps, chapters and a frame for every moment for five credits per video, with videos without captions billed at 10 credits per minute to transcribe. Audio transcription is billed at 10 credits per minute, rounded up, with a one-minute minimum.

What else is in Expand’s API package?

Expand says Fetch includes two additional API features: X-Ray and Highlights. X-Ray extracts hidden JSON and is meant to keep question-relevant state while dropping analytics and configuration data.

Highlights returns passages that answer a user’s question. The company says Highlights can be used alongside Fetch for one extra credit, or independently of Fetch at one credit per request. Expand’s pricing page says successful Fetch requests cost one credit each, with 1,000 credits priced at $1, while failed, blocked and empty requests cost nothing. The company also offers 10,000 free credits, says credits do not expire, and charges crawling at the same one-credit rate per page with no extra crawling cost.

Expand says its crawler can scale to millions of pages, and its WebSync product streams ordered page-added, page-changed and page-removed events that can be resumed from the prior offset. The company’s website also lists browser usage at $1 per browser per month, with an example of 1,000 browsers costing $1,000 monthly.

What does Expand say about performance and pricing?

Expand is positioning Fetch as both a cheaper and more effective way for agents to read the web. On its site, the company says its token-efficient format uses 24% fewer tokens at scale and gives an illustrative savings estimate of about $576,000 a year at 5 million requests a month on Gemini 4. Those savings are Expand’s estimate, not an independently verified benchmark result.

The company also published benchmark claims. Expand says agents answered 74% of questions using Expand Markdown versus 55% for the next best API, and that in its MDBench testing it was the only API to return hidden JSON. Expand says MDBench Hard v1 contains 1,094 questions across 326 hard pages and was tested on Oct. 8, 2026.

On that benchmark, Expand says it recorded 74.1% accuracy, compared with 55.3% for Exa and 27.2% for Firecrawl, with confidence intervals shown. The company also says Iris beat Jev on its validation set, with an F1 score of 92.7% versus 69.6%. As with any vendor-provided benchmark, those numbers reflect Expand’s own testing and methodology.

Who is behind Expand?

Expand founder Tim Suchanek describes the company as an applied research lab aiming to make the web queryable like a database. Suchanek says Iris took a year to build and cites his background as a founding engineer at Prisma and founder of Stellate.

Y Combinator’s Launch YC page attributes the Fetch announcement to Suchanek and says the launch post was published four days before the captured page. The YC launch post also describes the broader Fetch package as Iris, X-Ray, Wayback Machine and Highlights, plus Crawl and WebSync.

For founders and AI builders, the product sits in a growing category of web-to-agent infrastructure: APIs that convert live sites into cleaner, structured inputs that model-powered systems can reason over. Expand’s pitch is that Fetch reduces the amount of scraping, parsing and cleanup developers need to do themselves, while preserving source links and snapshots for traceability.