TheoremDB opens as a public workspace for machine mathematics
TheoremDB has gone live at theoremdb.org as a public workspace for machine mathematics, currently in alpha. The site's stated problem is that research agents working on mathematics often repeat effort because earlier attempts, partial results, and failed approaches are scattered and hard to find. TheoremDB's answer is a shared record that agents and humans can search and extend rather than starting from scratch each time.
The workspace is organized around reviewed problems, each with a defined target. Every problem has a card that opens into a packet showing what has already been proved, which routes have failed, and the code behind every computation submitted for it. Solutions can be submitted at several evidence grades, and a Lean-verified proof receives the highest grade available, giving the database a machine-checkable standard for its strongest claims.
Public writes are already live, including Lean proof contributions submitted through a companion tool called TheoremDB Researcher. One feature described as part of the long-term design, semantic expansion, is explicitly not yet enabled, so problems and prior attempts must currently be found by browsing rather than by meaning-based search.
The open problems already listed on the site range across number theory, combinatorics, complexity theory, and dynamical systems, including long-standing questions such as whether P-vs-NP-adjacent complexity classes collapse, whether the Chirikov standard map has positive entropy for some parameter, and sharp constants for specific finite combinatorial structures. The site frames its ambition explicitly: over time, its records could become for mathematical research what OEIS, the Online Encyclopedia of Integer Sequences, already is for integer sequences: a searchable index of problems, approaches, evidence, and results.
Key facts
- TheoremDB has launched at theoremdb.org as a public workspace for machine mathematics, currently in alpha.
- Public writes are already live, including Lean proof contributions submitted through a companion tool called TheoremDB Researcher.
- Semantic expansion, a planned meaning-based search layer, is explicitly not yet enabled.
- Each problem card records what has been proved, which approaches failed, and the code behind every computation; solutions can be graded by evidence, with a Lean-verified proof ranking highest.
- The site states its long-term ambition is to become for mathematical research what OEIS is for integer sequences.
Why it matters
Research agents working on mathematics tend to repeat work because earlier attempts, partial results, and failed approaches are hard to find once they scatter across papers, forums, and private notes. TheoremDB's premise is that giving those attempts a shared, searchable record lets agents and humans build on prior work instead of rediscovering the same dead ends, the way OEIS already does for integer sequences.
Who it affects
The site is aimed at research agents and human mathematicians working on open problems in areas such as number theory, combinatorics, complexity theory, and dynamical systems. It also targets anyone contributing formal proofs, since the highest evidence grade on the site is reserved for proofs verified in the Lean proof assistant.
How to use it
TheoremDB is live at theoremdb.org in alpha, with public writes already open. Contributors can submit Lean proofs through the companion tool TheoremDB Researcher, and each reviewed problem has a card documenting what has been proved, which routes failed, and the code behind every computation. Semantic expansion, a planned search or discovery feature, is explicitly disabled for now, so navigation depends on browsing the listed problems rather than searching by meaning.
How solid is it
The claims here come directly from the site's own copy: it describes itself as in alpha, with public writes and Lean submissions live and semantic expansion disabled. The source gives no launch date, names no individual or organization behind the project, and states no total count of open or solved problems, so the project's scale and backing cannot be independently assessed from the site text alone.
Risks and caveats
As an alpha product with public writes already open, the database's content and grading system may change, and quality control on submissions is untested at scale. The site does not disclose who built or funds it, nor how many contributors or proof submissions it has received so far. With semantic search disabled, the practical usefulness of the shared record depends on how easily problems can be browsed until that feature ships.
“Over time, those records can become for mathematical research what OEIS is for integer sequences: a searchable index of problems, approaches, evidence, and results.”
— TheoremDB site copy