Firecrawl Raises $75M for Alexandria and Provider Payments

On September 22, 2026, San Francisco-based Firecrawl announced a $75 million Series B led by Smash Capital and the launch of Alexandria, a retrieval library for AI agents. The service brings the live web, outside data providers and Firecrawl’s own indexes into one interface. The financing is intended to expand that coverage and build a way for more knowledge providers to be paid when agents use their contributions.

As SiliconANGLE’s account describes, Alexandria combines third-party data with Firecrawl-maintained datasets, while a broader self-service licensing system remains planned. Developers can access the retrieval service now; prospective contributors cannot yet enroll in an open, automatic payout program. The distinction matters because existing payments depend on individual provider agreements, not a published set of terms available to every researcher or publisher.

What agents can retrieve through Alexandria

The Alexandria product catalogue lists Firecrawl’s Research, Developer and Government indexes alongside official APIs, licensed publishers and site-specific connectors. It shows ways to connect through an API, MCP or CLI. Those routes give developers a common point of access, but they do not turn the underlying sources into one uniform dataset.

The Research Index is built around scientific paper abstracts, with tools for reading passages and following citations. The Developer Index covers READMEs, issues, pull requests and documentation; the Government Index covers material including laws and permit information. The catalogue also distinguishes official APIs such as FRED and the World Bank from licensed publishers such as Fiscal.ai and Particle. An agent may encounter all of these through Alexandria while receiving records with different origins, scope and usage conditions.

That difference affects what an answer can establish. An abstract can identify a relevant study without standing in for the full paper, while a government record or an official API response carries a different provenance from an extracted webpage. A shared retrieval interface reduces the integrations a developer must maintain; it does not erase the need to know which source supplied a result or what that source actually contains.

Existing provider deals and the proposed payment system

One named provider relationship predates Alexandria. In a March 17 partnership announcement, Wikimedia Enterprise described an arrangement for Firecrawl to obtain Wikimedia project data through its commercial APIs. At that point, routing requests through the direct API was described as forthcoming. The partnership offered a way to reduce repeated page scraping, which both parties described as resource-intensive.

That negotiated arrangement is narrower than the payment model in the headline. Firecrawl has paid agreements with official providers, while its public invitation to people offering data, content or expertise is a waitlist. No general payout formula, provider share of customer charges or date for opening self-service earnings has been disclosed. Those terms will determine whether the proposed system works for contributors beyond organizations able to make individual agreements.

Alexandria’s source categories also raise distinct questions about rights and attribution. Access to an official API can rest on a direct agreement; a Firecrawl-built index and a retrieved webpage have different paths back to their original material. A price shown to a developer for using a capability would not, by itself, reveal how much its information provider receives. The planned marketplace will need to connect usage to the relevant contributor if it is to make payment for that usage meaningful.

What the reported quality gain measures

In its launch account, Firecrawl reports that agents using Alexandria scored 21% higher on answer quality across 845 tasks than agents using built-in web tools. The comparison used the same model and prompts for both groups, with blind AI judging. The figure describes the agent configurations in that evaluation; it is not a measured improvement attributable to any single provider, index or licensed dataset.

The evaluation was conducted by Firecrawl. The published account does not provide the task list, judging rubric, score distribution or breakdown of which source types contributed to the answers. Without those details, readers cannot tell whether the gain was concentrated in particular subjects or how often licensed data, proprietary indexes or the live web made the difference. The result is a specific company-reported comparison, rather than independent evidence that every Alexandria query produces a better answer.

What remains open after the launch

Alexandria is available as a retrieval service, and individual paid provider agreements give the payment ambition an existing foundation. The wider proposition still depends on a self-service system that accepts contributors, tracks demand for their material and states how earnings are calculated. Its reach and economics cannot be assessed from the waitlist or the reported quality score alone.

The next substantive disclosure would be the terms under which a new provider can participate: what material is eligible, how its use is identified, and what payment follows. Until then, Firecrawl’s funded expansion and Alexandria’s current source catalogue are concrete; broad contributor earnings remain a plan.