Librar Labs Thinks School Libraries Are the Key to Organizing the World's Knowledge

The Macro: The World’s Knowledge Is a Mess That AI Cannot Navigate

Here is a problem that gets overlooked in the AI hype cycle. The world is full of massive amounts of unstructured knowledge, and most of it is effectively invisible to computers. Library catalogs, publishing metadata, literary collections, institutional archives. These databases were built over decades by different organizations using different standards with different levels of quality. They are fragmented, inconsistent, and full of errors.

This matters because AI systems are only as good as the data they can access. If you want an AI to help students find books they will actually enjoy, or help publishers understand their catalog, or help researchers navigate literary collections, you need clean, structured, interconnected data. What you actually have is a mess of MARC records, conflicting ISBNs, inconsistent author names, and metadata that was entered by hand in the 1990s.

The library technology market is dominated by a few players. OCLC runs WorldCat. Ex Libris (owned by Clarivate) sells Alma and Primo. Follett dominates K-12 school libraries. These are large, slow-moving organizations with enormous installed bases. None of them have AI-native data infrastructure. They are adding AI features on top of systems that were not designed for it.

The Micro: Self-Healing Data for Libraries and Beyond

Jonathan Gortz, Carl-Hugo Jacobsson, and Kaan Sirin founded Librar Labs. Jonathan is CEO. Kaan calls himself “chief librarian.” The team is four people strong, based in San Francisco, part of YC Winter 2026 with Jared Friedman. They are backed by founders and operators from Palantir, Lovable, and OpenAI.

The company has two faces. The commercial product is for school libraries, where the technology reportedly doubles reading rates. The underlying platform, called OSSUS, is more ambitious. OSSUS describes itself as “self-healing data infrastructure that turns fragmented records into trusted, agent-ready systems of truth.” AI agents continuously audit, merge, enrich, and correct unstructured data in real time.

The school library angle is a clever wedge. School libraries have messy data, limited budgets, and measurable outcomes like reading rates that make ROI easy to demonstrate. If Librar can prove the system works in schools, the same technology applies to public libraries, university libraries, publishers, and eventually any organization sitting on fragmented literary or institutional data.

The “self-healing” framing is interesting. Traditional data cleaning is a one-time project. You pay consultants, they clean the data, and it starts degrading again immediately. A system that continuously maintains data quality is fundamentally more valuable because it solves the problem permanently.

The Verdict

Librar Labs is tackling a real infrastructure problem with a smart go-to-market. Starting with school libraries gives them a beachhead that is small enough to win quickly but large enough to generate meaningful revenue and proof points.

The competitive risk is that the incumbents move. If Follett or Ex Libris builds similar AI data cleaning capabilities, they have the distribution to make Librar irrelevant. But enterprise library technology moves slowly, and Librar’s AI-native architecture is hard to replicate by bolting features onto legacy systems.

In 30 days, I want to see the number of school library deployments. In 60 days, the question is whether the reading rate improvements hold up across different school demographics. In 90 days, I want to know if any public library systems or university libraries are in the pipeline. The jump from school libraries to larger institutions is where the business gets interesting.