scout → read
Scout a page’s links and structure cheaply, then deep-read only what matters — fewer tokens burned.
Marrow
A crawler for AI agents. It scouts, reads and crawls the web into clean markdown — so your agent doesn't have to.
The layer beneath your agent
Section titled “The layer beneath your agent”Claude, Codex and Gemini can already search. Marrow is the layer underneath — it gets the data and hands it over clean. The AI stays with your agent; Marrow stays a crawler.
Cache as standard
Re-reads are near-free. The second fetch of any URL is fast and cheap.
Drive your own browser
The pages worth automating are behind a login. You sign in — 2FA, CAPTCHA, SSO — and the agent works past it, in your Chrome, on your machine. No password ever reaches Marrow. Learn how.
Turn one page into a catalog
One reference page becomes a structured dataset. Written twice — API and agent — so you can compare. Learn how.
Data behind a click
When the rows only appear after “show details”, the cheap path returns more text and none of the data. depth:"deep" renders and expands the page first. Learn how.
Why not just my agent’s search?
Section titled “Why not just my agent’s search?”Search answers questions. Marrow fetches sources. Your agent does the thinking; Marrow feeds it clean, complete pages — including the ones a search result only points at. The AI lives in your agent, not in the crawler.
Built for the loop, not the shot
Section titled “Built for the loop, not the shot”An agent rarely finds what it needs in one search. It loops — scout, read, rephrase, read again — and the answer comes together across those turns, not in any single shot. Marrow makes that loop cheap: re-reads are near-free from cache, so your agent can keep going until it has what it needs. Crawlers that bill per request do the opposite: every extra turn costs more. Marrow is built for the loop.