Skip to content

Firecrawl launches paid agent search, then open sources a 37x-faster scraping engine

· by

The short version

Firecrawl shipped UltraSearch, a paid search product enriched with data on 250 million-plus people and 75 million-plus companies, on November 17, then open-sourced Firecrawl Local, a Rust rewrite it says scrapes 37 times faster and uses 230 times less memory, one day later.

Firecrawl launches paid agent search, then open sources a 37x-faster scraping engine

Firecrawl spent launch week selling access to a database of more than 250 million people and 75 million companies, then turned around and gave away the scraping engine that pipeline runs on, for free.

What happened

UltraSearch for agents went live on November 17, 2026, returning search results pre-loaded with "data snippets that matter" instead of raw links, per Firecrawl's changelog. A single call can surface more than 250 million people with job titles, 75 million companies with funding data, and 150,000 stocks with live pricing, pulled from Firecrawl's Alexandria data platform, according to that same entry. Firecrawl Local followed one day later, on November 18: a rebuild of the company's scrape, crawl, and map primitives in Rust, per Firecrawl's changelog. Per that entry, the new engine runs 37 times faster than the self-hosted version, idles at 11 MB of memory (230 times less than before), and ships "100% open source" under the project's existing license, while staying compatible with current Firecrawl MCP servers and SDKs.

Neither release arrived in isolation. Five days before UltraSearch, on November 12, Firecrawl shipped MCP v3, adding access to more than 3,000 tools through automatic provider waterfalls, 60 times higher rate limits, four times more usage, and the ability to spawn parallel subtasks, per Firecrawl's changelog. One day after that, on November 13, the company expanded its Alexandria company database to more than 75 million firms, pitching headcount, revenue, funding, and tech-stack lookups in roughly half a second at a price it says undercuts data-enrichment vendor Clay by 88% at scale, per Firecrawl's changelog.

Self-hosted scraping gets free while agent-ready enrichment stays a paid product

Running the core scrape, crawl, and map job, turning a URL into clean markdown, now costs nothing and needs no Firecrawl account. That closes off the case for paying a hosted plan just to do raw extraction. Enrichment is the part that still gets metered: Firecrawl's cloud plans charge one credit per page scraped, crawled, or mapped, two credits per ten search results, two credits per minute of browser interaction, and one credit per page on each monitoring check, per Firecrawl's pricing page. Firecrawl Local sidesteps all of it, since it runs the same primitives on a developer's own machine instead of calling the API. Knowing who holds a given job title, what a company raised, or what a stock trades at right now is still UltraSearch's and Alexandria's selling point, and that is something Local does not touch. A team that only needs page text now has a free, open-source path that did not exist a week ago. Whoever needs entity context attached to a search result, though, is still paying Firecrawl for it, through the same credit-metered plans the November 13 company-data release and the September Series B were built to grow.

Context and reactions

Alexandria is the data platform underneath both headline launches, introduced on September 22 alongside a $75 million Series B led by Smash Capital, with Altos Ventures, Nexus Venture Partners, Y Combinator, Freestyle, and Offline Ventures also participating, per Firecrawl's own announcement. Caleb Peffer, writing that announcement, said the company would spend "a good chunk" of the round "buying knowledge from people," the rationale he gave for Alexandria's data-provider integrations, custom connectors, and Firecrawl's own indexes. Steady changelog activity since then tracks that bet rather than the scraping core: MCP v3 on November 12, the company-data expansion on November 13, and UltraSearch on November 17. Local, arriving a day after UltraSearch, is the first release in that stretch pointed at the free, self-hosted side of the product instead of the paid data layer.

What to watch next

Firecrawl has not said whether it is tracking Firecrawl Local's adoption against hosted-API revenue, and it has not published usage data comparing the two paths. Both the 37x speed figure and the 230x memory figure come only from Firecrawl's own changelog so far, so an independent benchmark against the prior self-hosted build or against rival open-source scrapers would test whether the numbers hold outside the company's own test setup. Pricing is the clearer business signal: whether the credit-based cloud plans that fund Alexandria and UltraSearch change in the months after a free local alternative ships, or whether Local works as a funnel that still points power users toward paid enrichment once raw scraping stops being the thing worth charging for. Developers can try the free local engine or the hosted plans at Firecrawl.

Sources