Alexandria Has Papers, Code and Law. Here's the Missing Index: Time
On September 22, Firecrawl launched Alexandria alongside a $75M Series B (press release). The pitch is simple and useful: one interface that gives AI agents access to official data providers, Firecrawl's own indexes and the live web, instead of a pile of separate integrations (Firecrawl).
The library is already broad. There is a Research Index of AI and ML papers and their code (launch post), a Developer Index of more than 70 million READMEs, docs, issues, pull requests and OpenAPI specs (launch post), a Government Index of laws, regulations and ordinances, 88 official providers exposing 504 capabilities across 28 categories, and Firecrawl's own search and scrape for the live web.
Everything in it is present tense
Look at that list again and notice what every shelf has in common: it describes the world as it is now. The Developer Index has today's docs. The live web is, by definition, today's web. A regulation index holds the current text of the law.
That is exactly right for most agent tasks. But a whole class of questions can only be answered by looking backwards, and there is no shelf for them:
What did this page say on a past date? "What did Acme's pricing page say in March 2023?" A live scrape returns today's price. The answer you need was overwritten long ago.
When did it change? "When did they drop the free tier?" or "When did this refund policy go from 30 days to 14?" You need every version of the page, with dates, not the latest one.
What did the docs say for the version I'm using? Coding agents regularly work against a library version that is two releases behind. Current docs describe APIs that don't exist yet in that version, and omit the ones that were removed. Version-pinned docs are a real gap in any current-docs index.
The archive that already carries the name
There is a fitting history here. The press has called the Internet Archive the digital Library of Alexandria for years, and the Archive had a real partnership with the Bibliotheca Alexandrina in Egypt, which hosted a mirror of its collection. The Wayback Machine holds hundreds of billions of captures of the web as it was.
What it doesn't do is answer questions. It gives you a calendar of snapshots per URL, which is perfect for a person spot-checking one page and awkward for an agent that needs a direct answer in one call.
What we built: web history for agents
arkibber turns Internet Archive captures into answers. Four tools, available over MCP and a REST API:
page_at_date returns the text of a page as it was on a given date, from the closest capture at or before it. page_changes lists every distinct version of a page between two dates. docs_at_version returns a documentation page, and the docs pages it links to, as they were on a release date. domain_timeline shows month-by-month change activity for a whole site, with its most frequently changed pages.
Every answer includes a link to the exact Wayback snapshot it came from. That matters for trust (anyone can check the source) and for the Archive, which gets the attribution and the traffic.
Complement, not competitor
We think of arkibber as a missing shelf, not a rival library. An agent can use Alexandria for the present and arkibber for the past in the same session: pull a company's current pricing from the live web, then ask arkibber what it was a year ago and when it changed. Firecrawl has said it pays providers whose data agents use, and plans a self-service provider program. A web-history index is the kind of provider we'd like Alexandria to have, and we'd like it to be us.
To be clear about the relationship: arkibber is independent and isn't affiliated with or endorsed by Firecrawl. The two simply work well side by side.
Built to support the Internet Archive, not strain it
Access to the Archive is under real pressure. More than 240 sites, including the New York Times and USA Today, now block its crawler over AI concerns (Fortune), and the Archive added rate limiting against automated traffic in September (Internet Archive blog).
So we built arkibber to be the opposite of the scrapers causing that pressure. Fetching is cached and rate-limited through a single identified crawler with a contact address. We keep a shared index, so a site is fetched once no matter how many people ask about it. We serve derived data (text, changes and timelines), not bulk copies of captures. And every answer sends readers back to the Wayback Machine.
Try it
You can see how any website changed over the last three years on the web history page, no account needed. If you build agents, the MCP tutorial gets you from zero to your first historical answer in a few minutes. Alexandria gives agents the world's knowledge as it is. arkibber gives them the web as it was.