Sheaf

Drop in a paper — or just name what you want to understand — and get a problem sheet cut for the mathematics you have already met, sharper each time you come back.

Give Sheaf a paper, a blog post, or just a concept you want to better understand — an arXiv link, or the simple text black holes — and it'll write a problem sheet, or a course of them: problems come with staged hints and solutions, you can type your attempts in rough notation and have them marked, and there is a chat window with an AI supervisor — useful for asking after references to similar work, or good explanatory writing. Sheaf builds a profile of your skills and understanding, uses it to generate problems that suit your learning style, and tries to avoid problems you've already solved. Where a problem admits a faithful formalisation it also ships a Lean 4 statement, which Sheaf can compile against mathlib to certify that the statement type-checks.

Sheaf is currently a local-only tool: every problem sheet and all of your profile information is stored on your machine. It's free, and talks to a model through Claude Code, Codex, the Anthropic API or the OpenAI API — or you can hack in your own backend for any other completion source. I ask that early testers who are willing and able opt into usage pings, so I can understand simple usage patterns.

I wrote Sheaf while I was trying to learn more about quantum computing, as I engage much better with problems than with reading, and translating results into puzzles is the nicest way to feel their texture.

Demo sheet

Sheaf's own output, cut from Peter Shor's original paper on factoring integers into primes with a quantum computer: five problems that build the algorithm in the order thirty years of hindsight settled on — the classical reduction to order finding, the quantum Fourier transform, and recovering the order by continued fractions.

Shor's Algorithm: Order Finding, Fourier Transforms and Continued Fractions

Every sheet is a single self-contained HTML file — mathematics and fonts embedded, hint ladders and solutions folded shut, print-ready.

Install

Sheaf is in a private pilot while the first cohort works through real papers. Write to [email protected] for an invite — say roughly what you'd want to read with it. With access, the install is two minutes:

git clone https://github.com/grsmith44/sheaf.git
cd sheaf/web
npm install
npm start

Node 22 or newer required. Full install instructions and model-backend setup.