Firecrawl Raises $75M, Launches Alexandria AI Agent Library
Firecrawl Raises $75M, Launches Alexandria AI Agent Library
AI agents are only as accurate as the data they can reach. On September 22, 2026, Firecrawl announced a $75 million Series B round and launched Alexandria, a unified knowledge library designed to give agents reliable, structured access to more than 100 data sources through a single API. The company says internal tests show agents using Alexandria answered 21% more questions correctly than agents relying on built-in web tools alone.
What Is Alexandria?
For most enterprise teams building agents today, accessing external data requires gluing together three or four separate connectors: a web search API, a database of research papers, a code documentation scraper, and maybe a government records feed. Each connector breaks independently, each has different rate limits, and none of them share a common schema.
Alexandria tries to solve that fragmentation. It combines four types of access into one interface:
- Firecrawl’s proprietary indexes. The Research Index covers tens of millions of scientific paper abstracts, the Developer Index spans tens of millions of code repositories and documentation READMEs, and the Government Index includes laws, regulations, court records, and SEC filings.
- Official data providers. Firecrawl already pays Wikimedia Enterprise for licensed Wikipedia access. Alexandria routes those requests through the same API, with more providers planned.
- Custom connectors. Teams can plug domain-specific data sources, such as internal CRMs or third-party SaaS exports, into the Alexandria interface.
- The live web. Firecrawl’s existing scraping infrastructure, which the company says has served more than 5 billion requests, remains the fallback for pages not covered by the indexed sources.
The result is a catalogue an agent can query once to get structured results from whichever source is most appropriate for that query, instead of making parallel calls to separate endpoints and then reconciling conflicting formats.
The Accuracy Claim
Firecrawl says that across 845 tasks, tested with the same model and prompts and judged blindly by AI evaluators, agents using Alexandria scored 21% higher on answer quality than agents using built-in web tools.
That number deserves close reading. Firecrawl ran the test itself, on verticals it chose, using its own definition of answer quality. No independent party has reproduced the result, and the specific tasks and comparison baseline have not been published in full. Read it as a directional vendor claim rather than a controlled benchmark.
That said, the underlying mechanism is defensible. Web search tools retrieve whatever pages rank highest for a query, which can be a news aggregator, a paywalled abstract, or an outdated forum thread. A curated index built from primary sources tends to return more authoritative content for domains like scientific research or regulatory compliance, where the right answer is a specific record, not a summary of summaries.
What This Means for Enterprise AI Teams
Enterprise AI builders should read the Alexandria launch as a data infrastructure bet, not a feature announcement.
Most production agents fail not because the model is too weak but because the context they receive is incomplete or unverified. A sales agent that pulls company data from a third-tier news aggregator instead of the SEC’s primary filings makes different decisions than one drawing from authoritative sources. The same applies to engineering agents looking up API documentation, or compliance agents checking regulatory requirements. For teams where agent accuracy matters for business outcomes, the quality and provenance of data in context is a first-class engineering concern.
Alexandria sits in the same infrastructure layer as Pinecone’s Nexus, which targets retrieval from private enterprise knowledge stores. The distinction is directionality: Nexus optimizes recall from internal data the enterprise already owns; Alexandria broadens what agents can reach externally. In production, enterprise agents usually need both, pulling from internal memory and external reference data in the same response.
The OpenAI Managed Agents API and similar orchestration layers assume the agent already has access to the tools it needs. Alexandria is one of the tools those orchestration platforms would route through, handling the data-retrieval step that makes tool calls meaningful.
The Data Economy Angle
Firecrawl’s CEO Caleb Peffer wrote that a significant share of the $75 million will go toward paying the people whose knowledge Alexandria distributes. The company already pays Wikimedia Enterprise directly for Wikipedia access. A self-service licensing system for researchers, publishers, creators, and domain experts is planned, though the timeline and revenue-share terms have not been announced.
This is the part of the Alexandria launch with the most long-term strategic weight. Every AI product running agents against web data is implicitly competing for access to that data. Legal challenges from publishers and rights holders are accelerating. The companies that establish legitimate licensing relationships early, and structure those relationships to attract high-value contributors, will have a structural advantage over those that depend entirely on open crawling as the legal landscape tightens.
The economics of that marketplace will determine whether Alexandria becomes infrastructure or another wrapper. If the licensing terms attract authoritative sources at scale, the index becomes genuinely defensible. If they do not, competitors with better data relationships will undercut it.
By the Numbers
| Metric | Value |
|---|---|
| Series B round size | $75 million |
| Lead investor | Smash Capital |
| Co-investors | Altos Ventures, Nexus Venture Partners, YC, Freestyle, Offline Ventures |
| Users on Firecrawl platform | 1.5 million |
| Requests served to date | 5 billion |
| Data providers in Alexandria | 100-plus |
| Internal accuracy lift (vendor claim) | 21% across 845 tasks |
| Access methods | API, MCP integration, CLI |
What Comes Next
Firecrawl says the Series B will fund three areas: improving search relevance within Alexandria’s indexes, building deeper retrieval systems, and connecting more first-party data providers through the self-service licensing program.
The MCP integration is important for enterprise adoption timing. Most enterprise AI platforms are already building MCP-compatible tool registries, which means an agent framework that already supports MCP can route to Alexandria without a custom integration. That lowers the barrier for teams evaluating the platform.
For enterprise AI and GTM teams running complex agentic workflows, the practical question is whether Alexandria’s indexed sources cover the domains their agents actually operate in. Research, code, and government documents are specific verticals. Teams in financial services, healthcare, or supply chain should evaluate whether their primary reference data types are in the catalogue before committing to the integration.
If you are mapping your enterprise AI data infrastructure, talk to Enera about how leading teams are structuring agent context pipelines and choosing between proprietary indexes, licensed data providers, and open web access.
Sources: Firecrawl Alexandria launch blog, SiliconANGLE, The Latent, RuntimeWire