About TheoryBench.
TheoryBench brings a paper draft, PDF preview, computations, references, and Theoria chat into one local-first workbench for theoretical physics.
What it is
A paper workspace
Edit project files, compile locally, inspect the PDF, and keep review comments beside the source.
A place to compute
Run Python and Wolfram cells as part of the working draft, with outputs promoted only when you choose.
An advisor layer
Theoria can find references, flag likely mistakes, and ask for the next useful check using the AI account you connect.
A share path
Share an inspectable bundle, or connect Overleaf when collaborators need the paper there too.
What stays local
Runs on your machine
- Project files and private working notes
- TeX compile and PDF rendering
- Python cells and local packages
- Mathematica or Wolfram Engine license
- Claude/Codex accounts you connect
Shared only when you choose
- Read-only export bundles
- Overleaf sync from the local app
- Shared accounts later
- No upload of private project data unless you choose it
How it works
-
Open a project folder
Open an existing paper folder or a sample. TheoryBench reads its source, assets, bibliography, and notes.
-
Check what is installed
It reports what's available — TeX, Python, Wolfram, PDF, Claude/Codex — before you rely on it.
-
Write and compute in context
LaTeX stays the source of truth; run cells to check a derivation, and promote only the results that belong in the paper.
-
Ask Theoria in the draft
Theoria attaches feedback to specific lines, equations, citations, and cells, then points toward the next computation rather than replacing the work.
-
Share deliberately
Share a read-only bundle, or sync TeX, bibliography, and figure files with Overleaf. Your local notebook is not run from Overleaf.
Current status
You can preview the workbench in the demo right now. Soon: a downloadable Mac app to run it locally: edit, compile to PDF, run cells, pull references, get a review from Theoria, and sync an Overleaf Git project. Later: shared accounts and secure remote computation.