Academy Network of AI Leads in the Mathematical Sciences · Birmingham, 27–28 April 2026

Wolfbook

AI-Friendly Wolfram Language Notebook


Nikolay Gromov
Professor of Theoretical Physics · Mathematics Department · King's College London

Wolfram Language & Wolfbook

🔢Wolfram Language
Symbolic & analytic computation — first class
Powerful numerical methods
Rich built-in visualisation libraries
Comprehensive documentation & vast built-in knowledge
Maths · code · graphics · text — all in one environment
⚠ Key limitations of the classic Mathematica frontend
Binary .nb files — AI cannot read them; no git diff or merging
Weak AI integration — chat-only, no agentic workflows
Primitive LaTeX support
Closed ecosystem — difficult to collaborate and version-control
Wolfbook — the solution
An open-source Wolfram Language frontend running inside VS Code, Cursor or Google Antigravity — platforms with the richest AI ecosystem of any IDE.
Text-based .wb notebooks — full git diff & merge
Agentic AI with live kernel access — GitHub Copilot, Claude, Gemini…
Multi-language workspace: WL, Python, LaTeX, C++ — one window
Full VS Code power: split view, extensions, debugger, formatter
AI can read, write and run Wolfram Language natively
🎓 Free & Open Source (Apache 2.0)  ·  Wolfram Engine: free education licence

VS Code Superpowers — Out of the Box

Since Wolfbook runs inside VS Code, every editor superpower comes for free — version control, split view, debugger, extensions, and full AI integration.
1
Rich LaTeX Rendering
Rich LaTeX formula rendering in Wolfbook — paper-quality typesetting inline
Wolfram outputs render as beautifully typeset LaTeX inline — paper-quality math, straight in the notebook with proper line breaks.
2
Abort Mid-Evaluation
Aborting a long-running Mathematica evaluation mid-way in Wolfbook
Stop a runaway computation instantly.
3
Split View
Wolfbook notebook open in VS Code split view — two panes of the same file
View the same notebook from two positions at once — standard in VS Code, impossible in classic Mathematica.

Ghost Text: AI Reads Your Context

Wolfbook exposes Mathematica code cells and text fields in AI-readable form.
GitHub Copilot reads your comment, surrounding cells and outputs — then predicts the next code block instantly. No copy-paste, no context switching.
💬 Comment → ghost text suggestion
GitHub Copilot ghost text suggestion triggered by a Mathematica comment
✓ Predictive input based on comment context — accepted with Tab
▶ Result: code runs correctly
Wolfbook notebook showing ghost-text-generated code running successfully

Agentic AI — Live Kernel + Full Notebook Access

The most transformative feature: AI agents can read, execute code, and reason over your entire Wolfbook notebook — cells, outputs, and text — using the live Wolfram Language kernel.
Your mathematical work becomes interactive AI collaboration.
🔵 User defines a task in a notebook cell
User's initial prompt to Claude Haiku 4.5 written in a Wolfbook notebook cell
🤖Agent reads the notebook, writes & executes Wolfram code, verifies results, and delivers a structured summary — autonomously.
🟢 Claude Haiku 4.5 — Full Agent Output
Logos of GitHub Codex and Claude Code
Also works with all leading agents — or any model via Cline, supporting 50+ AI providers via OpenRouter — even DeepSeek 🚀

Wolfbook in the Wild — Emerging Workflows

🚀 Just 1 month old · 800+ downloads on Open VSX, VS Code Marketplace & GitHub · workflows are already emerging
1
Verify, Check & LaTeX-ify
Ask agent to check your calculations & boundary cases on the live kernel
Agent restarts kernel, verifies all cells run cleanly (or fixes mistakes), writes a verified summary section
Generate LaTeX notes from the notebook → push to Overleaf via GitHub, merging with your collaborators
2
Prototype → C++ Paclet
Prototype sloppy code in Mathematica — correctness matters, not style
Agent rewrites it as a high-performance C++ Mathematica paclet
Tests against prototype + intensive additional tests; generates documentation
3
Literature Review → Reproduce
Ask for a literature survey — Wolfbook exposes paper-search tools natively
Agent finds key results + BibTeX entries with a per-paper summary
Next prompt: write a notebook section reproducing all key results

Wolfbook Workflows — Research at Scale

4
LaTeX Equation Checker
Ask agent to go through your draft paper
It verifies every equation by running it through the Wolfram kernel
Annotates the LaTeX source with comments wherever errors are found
5
PhD-Level Project — (Advanced)
Start with a well-posed problem (PhD-grade, 1–2 weeks of work)
Prepare a detailed .md plan: step-by-step computation with test criteria per stage — just like briefing a PhD student (using Claud Opus in chat mode for example or planing mode of copilot)
Add the .md to your workspace, direct the agent to it
Agent follows the plan, verifying & testing each stage before proceeding
🍽️ Go for lunch — return to a fully solved, tested & documented result (if all goes as planned)

Case Studies — Research Notebooks from a Single Prompt

Each notebook was generated by Copilot agent mode from one initial prompt + a few steers. No manual code writing.

Spin Chain Spectrum

Heisenberg XXX Model
A chain of spins- with nearest-neighbour exchange , exactly solvable by the Bethe Ansatz. A twisted boundary condition interpolates between periodic and antiperiodic BC — a clean probe of integrability.
💬 PROMPT
“Use Wolfbook tools to implement the Heisenberg spin-chain Hamiltonian with quasi-periodic boundary conditions and plot its spectrum for L = 2, 3, 4 as a function of the twist. Then introduce a small non-integrable deformation and make the same plot and compare it with the integrable case.”
What was built:
siteOp[op,k,L] (KroneckerProduct embedding) + Hham[L_,phi_] builds the full matrix. Spectra for swept over (80 steps). Then NNN deformation (): integrable shows smooth level crossings; non-integrable shows avoided crossings.
Wolfbook tools:
  • wolfbook_insertCells / runCell
  • wolfbook_editCell
  • wolfbook_lookupSymbol — verified KroneckerProduct
Artifacts: spinchain.wb (12 cells) · spinchain_report.tex

psu(2|2) Superalgebra

Beisert Central Extension (AdS/CFT)
The symmetry algebra of the SYM spin chain (Beisert 2006). Has 14 generators: , (bosonic) + , (fermionic). Pure does not close — Jacobi identities force three central charges .
💬 PROMPT
“Explore the psu(2|2) superalgebra and its central extension using Mathematica. Load NCAlgebra and set up the generators: bosonic su(2)×su(2) and fermionic supercharges. Verify Jacobi identities for a representative selection of triples. Introduce the three central elements C, P, Q and show the algebra closes only with the central terms.”
What was built:
NCAlgebra loaded; grading set; super-bracket implemented as GB0[A,B]. Systematic Jacobi check for all generator triples — 7 identities fail without central terms. Adding to anticommutator: all close.
Wolfbook tools:
  • wolfbook_evaluateExpression — kernel scratchpad ×14+ calls to prototype each bracket rule before committing to a cell
  • wolfbook_insertCells / editCell
Artifacts: psu22.wb (multi-section) · psu22_report.tex

Compton Scattering in Scalar QED

Automated Feynman Diagrams with FeynArts
Scalar QED: . Compton scattering has 3 tree-level (seagull, s-, u-channel) and 53 one-loop diagrams. Agent installed FeynArts 3.12 and wrote the model file autonomously.
💬 PROMPT
“Find an online description of FeynArts and produce pictures for 1-loop Compton scattering in Scalar QED in this notebook. Use Wolfbook tools to test your expressions.”
What was built:
Custom ScalarQED.mod model file (SSV 3-point + SSVV seagull vertices). CreateTopologies + InsertFields for scattering. Full 53-diagram 1-loop sheet exported as PNG/PDF; diagrams further split into topology classes (self-energy, vertex, box) with DiagramExtract.
Wolfbook tools:
  • wolfbook_evaluateExpression — probed kernel; debugged model
  • wolfbook_getKernelState / restartKernel
  • wolfbook_insertCells / editCell
Artifacts: Fyanman.wb · ComptonScalarQED_notes.tex

Conclusions & Future Directions

📌 Key Takeaways
Text-based .wb notebooks: fully AI-readable, git-diffable, mergeable
Wolfram Language mathematical power + VS Code AI ecosystem — in one tool
Agentic AI reads, executes & verifies Wolfram code autonomously
Works with GitHub Copilot, Claude, Gemini, DeepSeek — your choice
🎓 Free & Open Source (Apache 2.0) · Wolfram Engine: free education licence
🌊
Vibe Problem Solving
The emerging paradigm where AI agents handle the computational heavy lifting while the researcher stays at the level of high-level problem formulation.

Like “vibe coding” — but for mathematical research.
Wolfbook is the natural home
for this new way of doing mathematics.

Get Started — Everything is Free

1
Wolfram Engine (skip if you have Mathematica)
Free for non-commercial use. Activate with a free Wolfram ID.
wolfram.com/engine
2
VS Code
Free, all platforms (Windows · macOS · Linux).
code.visualstudio.com
3
Wolfbook extension
Search Wolfbook in the VS Code Extensions sidebar, or use a link on the right. Create test.wb — kernel auto-starts on first Shift+Enter.
Optional: Install the GitHub Copilot extension for full agent-mode AI — free with GitHub Education.