Verified knowledge infrastructure for the human record.
QuarterMill turns archival, library, and museum collections into connected, source linked knowledge that people and AI tools can rely on. Institutions keep control of their material, and every answer points back to the original page.
Recognized by
- Tsai CITY
- Startup Yale 2026
- Black Is Tech 2026
- Yale Innovation Summit
- AI at Yale
- Dwight Hall at Yale
The problem
Why this matters now
The answer to a real question is usually spread across many documents in many collections, and no single one holds it.
Most AI tools blend whatever text they were trained on and lose the link back to where any of it came from.
So answers sound certain but cannot be checked, and gaps in the record get filled with guesses. Where being wrong has a cost, that is not good enough.
How it works
Three steps from collection to verified answer
- Step 1
Bring in your collection
We work with what you've already digitized: scanned pages, PDFs, existing transcriptions, and catalog records.
- Step 2
Connect and check
We find the people, places, and events in each document, link them across collections, and tie every fact to the page it came from. When the records don't answer a question, we say so.
- Step 3
Use it anywhere
Your staff and visitors can search and explore your collections, and AI tools can draw on them with sources attached.
What we can read
Handwritten and printed material
From manuscripts and letters to newspapers, government records, and legal documents. We specialize in what most AI reads poorly, including handwritten Arabic script languages such as Jawi, Pegon, and Ajami, and Yiddish handwriting.
Coming soon: audio and video collections.
- Manuscripts and letters
- Newspapers
- Government records
- Legal documents
- Scholarly records
Who it's for
Built for the people who hold the record and the people who build on it
For institutions
Make your collections usable by researchers and AI tools without handing them over.
- You decide who can access your material and how
- Every use is attributed back to your holdings
- Your records stay in your systems
For AI builders
Ground your product in primary sources instead of scraped text.
- Structured, source linked answers through an API
- Citations to the original page for every claim
- A clear signal when the record has no answer
Where we are
Early, and working with a small group of partners
QuarterMill was founded at Yale in 2026 and won the New Haven Civic Innovation Prize at Startup Yale. We are in early work with archives, libraries, and museums, and we are taking on a limited number of pilot institutions.
What we hold to
Four commitments
- Source first.
- Every claim links to the page it came from.
- Owners stay in control.
- Institutions decide who uses their material and how.
- Silence is reported.
- When the record has nothing, we say so instead of guessing.
- Corrections stay visible.
- When new evidence changes a claim, the earlier version is kept and marked.
“AI learned to talk before it learned to cite. We are building the layer that makes knowledge accountable to its sources.”
See QuarterMill on your own collection.
For institutions, AI builders, researchers, and press. Pilot QuarterMill with your own collection, see it working, or ask us anything.