What are Firecrawl Alexandria Datasets?

Ready-made bulk datasets cut from Firecrawl Alexandria's indexes and public sources in legal, finance, healthcare, and code and engineering. You pick a slice, a file format and a destination, and every row keeps its source URL, crawl date, license and rights basis.

What formats and delivery options are there?

Parquet, JSONL or CSV, with Markdown, the original HTML, PDF and docx files, or both. A delivery arrives as download links or in your own Amazon S3, Google Cloud Storage or Cloudflare R2 bucket, once or on a weekly or monthly refresh.

How is licensing handled?

Each row records its license and rights basis. By default a delivery holds only public-domain and openly licensed rows, and you can choose to include rows with no stated license, flagged for your review. Names, emails and phone numbers can be redacted on request.

What is a custom dataset?

A custom dataset is built to your own schema from sources you pick. Firecrawl builds it from live web pages, PDFs and licensed data providers, and each record keeps the URL it came from.

Can Firecrawl build a dataset that isn't in the catalogue?

Yes. Send our data team a list of domains, a URL list or a description of the data. We scope the fields and volume with you, deliver a sample, then run the full collection and any refreshes you choose.

How do I build a dataset from websites myself?

Collect the URLs with /search or /map, then run a batch scrape with a JSON schema that names your fields. Each page comes back as one record in that shape. For an entire site, /crawl applies the same scrape options to every page it reaches.

Can Firecrawl turn PDFs into structured data?

/parse takes PDFs, Word files and spreadsheets up to 50 MB and returns Markdown, per-page content or structured JSON. /scrape also reads a PDF at a public URL.

How do I keep a web dataset up to date?

Firecrawl Monitor runs scheduled checks on a list of pages, a site or an always-on web search. Each check flags pages that were added, changed or removed and can notify a webhook, email or Slack, so the pipeline re-collects only those records.

Can an AI agent build a dataset when I don't know the sources?

/agent takes a prompt, optional starting URLs and a JSON schema, then searches and navigates the web on its own and returns data in that schema. maxCredits caps what one run can spend.

Where can I get licensed data to add to a dataset?

Firecrawl Alexandria connects data providers such as SEC EDGAR, Fiscal.ai and Wikipedia through Wikimedia Enterprise, and Firecrawl's Research Index, Developer Index and Government & Legal Index, to the same API. It needs an API key on a team with Firecrawl Alexandria enabled and isn't available in keyless sessions. Each tool lists its credit price before you call it.

What security and procurement options do enterprise teams get?

Firecrawl Enterprise includes SAML SSO, SCIM, a DPA, zero data retention, US data processing and custom concurrency. Firecrawl is SOC 2 Type II compliant, and the annual renewal audit is in progress. Teams can also buy Firecrawl through AWS Marketplace or Google Cloud Marketplace.

Datasets for AI training and data teams

The data you need,
delivered to your bucket.

Get ready-made datasets in legal, finance, healthcare and code, or have our team build a custom one. Every row keeps its source URL and license.

Request a dataset

Delivery alexandria_datasets.parquet

5 datasets · 1 schema

Rows from five catalogue datasets in one delivery schema, with the license and source of each
doc_type	license	doc_format	source_name
10-K filing	Unstated	html	SEC EDGAR
Court opinion	Public domain	pdf	CourtListener
Drug label	CC0 1.0	json	openFDA
Trial record	Registry terms	json	ClinicalTrials.gov
Final rule	Public domain	xml	Federal Register

5 rows · license and rights_basis on every rowPARQUET

Trusted by 150,000+
companies of all sizes

Read story

Read story

Read story

Read story

[ 01 / 08 ]

Catalogue

Ready-made datasets

Bulk datasets cut from Firecrawl Alexandria's indexes and public sources, with the original files kept next to the text.

US court opinions

Federal and state opinions with court, case type and citation graph.

htmlpdf

CourtListener · Official court sites · GovInfo

since 1950 · monthly updates

Chapter 11 court records

Bankruptcy opinions, first-day filings and local rules from D. Del., S.D. Tex. and S.D.N.Y.

pdfhtml

U.S. Bankruptcy Courts · GovInfo

since 2005 · monthly updates

Statutes, codes and session laws

U.S. Code, state codes, constitutions, court rules and session laws.

htmlpdfxml

GovInfo · State legislatures · State courts

since 1990 · monthly updates

Regulations and agency guidance

CFR, Federal Register, state administrative codes and agency guidance.

htmlpdfxml

eCFR · Federal Register · Federal agencies

since 1994 · monthly updates

Municipal and county codes

Local ordinances, code sections and council legislation.

htmlpdf

Municode · eCode360 · American Legal

since 2000 · monthly updates

Need something else?

Describe the sources and fields, and our team builds a custom dataset from web pages, PDFs and licensed providers.

Request a custom dataset

[ 02 / 08 ]

Custom datasets

Not in the catalogue? We'll build it.

Our data team builds datasets to your spec from web pages, PDFs and licensed providers, then delivers and refreshes them.

[ YOU SEND ]

A list of domains, a URL list or a description of the data

The fields each record needs
How many records, roughly
How often it refreshes

[ WE BUILD ]

Collection from web pages, PDFs and licensed providers

Fields and volume scoped with you
A sample in your schema first
The full collection after you approve it

[ YOU GET ]

Records in your schema, delivered to your bucket

Parquet, JSONL or CSV, or raw HTML and PDFs
The source URL on every record
Weekly or monthly refreshes

Requests teams bring us

01Technical documentation in non-English languages as raw HTML and PDFs, refreshed monthly with changed and new URLs
02Every page on a list of healthcare provider websites, crawled in full and dropped in S3
03Store-level menu prices for named competitors, collected every week
04Filings, court records and news tied to the tickers on a coverage list

Tell us the sources and fields, and we'll scope the dataset with you.

Talk to our team

[ 03 / 08 ]

Delivery

Delivered your way

Pick a slice by year, document type or jurisdiction, then choose how it arrives.

[ FILE FORMAT ]

Parquet, JSONL or CSV

parquetjsonlcsv

[ CONTENT ]

Markdown, the original files, or both

markdownhtmlpdfdocx

[ DELIVERY ]

A download link or your own bucket

downloads3gcsr2

[ UPDATES ]

Once, monthly or weekly

oncemonthlyweekly

[ LICENSING ]

Open licenses only, or unstated licenses flagged

open-onlyflagged

[ PERSONAL DATA ]

As published, or names, emails and phones redacted

as-publishedredact

Every row carries

doc_idurlsource_nametitledoc_typedoc_formatjurisdictionpublished_atcrawled_atmarkdownraw_htmlpdf_uritoken_countrights_basislicense

[ 04 / 08 ]

How it works

From request to delivery

Our data team confirms the size of every slice before anything is delivered.

[ 01 ]
Pick a dataset or describe one
Choose from the catalogue, or tell us the sources, fields and volume you need for a custom build.
[ 02 ]
Choose the slice and delivery
Narrow it by year, document type or jurisdiction, then pick the format, licensing filter and destination.
[ 03 ]
Review a sample
Ask for about 1,000 rows first, so you can check the fields and coverage before the full delivery.
[ 04 ]
Get the full set and refreshes
The dataset lands in your bucket or as download links, then refreshes weekly or monthly if you choose.

[ 05 / 08 ]

Build it yourself

Or build it on the API

Send a JSON schema and every page returns one record in that shape. Run the same endpoints our data team uses.

/searchFind pages, PDFs and papers by query /scrapeOne record per page in your JSON schema /parsePDFs, Word files and spreadsheets to JSON /crawlEvery reachable page on a site /interactPagination, forms and tabs in a browser /agentA prompt and a schema when sources are unknown /monitorRe-collect only pages that changed Firecrawl AlexandriaLicensed providers for data beyond the web

46x

knowledge base growth, Sep 2025 to Feb 2026

10,000+

domains covered

~800

web sources processed a day

Read the Stanford story

The free plan includes 1,000 credits a month.

Start for free

[ 06 / 08 ]

Rights and security

Data you can use

What legal, security and data teams check before a dataset goes into training or production.

Rights-cleared sources

Catalogue datasets are built from government works, public filings and openly licensed material. Publisher terms are reviewed per source before delivery, and each source is fetched under its own access rules.

Provenance on every row

Each row records its source URL, crawl date, license and rights basis. By default a delivery holds only public-domain and openly licensed rows.

Zero data retention

Enterprise plans can turn on zero data retention for /scrape, /interact and /search, so Firecrawl keeps no page content or extracted data after a request ends.

Ready for procurement

SOC 2 Type II compliant, with the annual renewal audit in progress. SAML SSO, SCIM, a DPA and US data processing, and you can buy through AWS Marketplace or Google Cloud Marketplace.

[ 07 / 08 ]

Compare

How Firecrawl compares with other ways to get data

Feature	Firecrawl	Managed data services	In-house scrapers	Dataset marketplaces
Ready-made datasets in legal, finance, healthcare and code	Yes	No	No	Yes
Custom datasets built to your schema	Yes	Yes	Yes	No
License and rights basis on every row	Yes	Varies	No	Varies
Original HTML and PDFs kept next to the Markdown	Yes	Varies	Yes	Varies
Delivery to your S3, GCS or R2 bucket	Yes	Yes	Yes	Varies
No scraper code or parsers to maintain	Yes	Yes	No	Yes
Self-serve API for your own pipelines	Yes	No	Yes	No
Callable by AI agents through MCP, CLI and SDKs	Yes	No	No	No

[ 08 / 08 ]

FAQ

Frequently asked questions

Ordering, delivery, licensing and building datasets on the API.

Datasets

Building on the API

Security

Enterprise plans can turn on zero data retention for /scrape and /interact, so Firecrawl doesn't keep page content or extracted data after the request ends. /search has two ZDR modes: Firecrawl-side only, or end to end, where the upstream search provider also retains nothing. ZDR searches leave out Firecrawl Alexandria provider results, and ZDR /agent runs call only ZDR-capable providers. A ZDR /agent result stays retrievable for 24 hours and is then deleted.

Get started

Tell us the dataset you need.

Our data team scopes it with you and sends a sample before the full delivery.

Request a dataset Start for free

Are you an AI agent? See setup options

Datasets for AI training and data teams

The data you need,delivered to your bucket.

Ready-made datasets

US court opinions

Chapter 11 court records

Statutes, codes and session laws

Regulations and agency guidance

Municipal and county codes

SEC periodic and current reports

SEC exhibits and long-tail forms

Restructuring and distress disclosures

Company and market research

Clinical trial registry records

FDA drug labels

FDA recalls

Medicare coverage determinations

CDC public health reports

Biomedical research, open access

Price transparency and compliance

Code repository issues, pull requests and docs

iOS app and Swift repositories

Industrial and infrastructure engineering

CAD models and engineering drawings

Not in the catalogue? We'll build it.

A list of domains, a URL list or a description of the data

Collection from web pages, PDFs and licensed providers

Records in your schema, delivered to your bucket

Delivered your way

From request to delivery

Pick a dataset or describe one

Choose the slice and delivery

Review a sample

Get the full set and refreshes

Or build it on the API

Data you can use

Rights-cleared sources

Provenance on every row

Zero data retention

Ready for procurement

How Firecrawl compares with other ways to get data

Frequently asked questions

The data you need,
delivered to your bucket.