OpenMat

OpenMat

The open-source, local-first, AI-native alternative to Mathematica.

Open source. The whole notebook, kernel included, under MIT or Apache-2.0. Read it, fork it, own it.

Local-first. The Rust symbolic kernel runs on your machine: in-process on desktop, as a 530 KB WebAssembly module in the browser. No server, and your notebooks never leave your laptop.

AI-native. Natural language is an input mode, not a plugin: type a sentence and a full notebook writes and evaluates itself, using your own API key or a completely local model. Nobody has shipped this combination before.

Download for macOS Read the docs GitHub v0.01 · Apple Silicon · 5.9 MB · MIT or Apache-2.0

make me a notebook exploring the damped pendulum with a slider for the damping

≡ interpreted · 5 cells

The Damped Pendulum

Friction drains energy each swing, so the oscillation decays. Drag c to change how fast.

In[1]:=NDSolve[{x''[t] + c x'[t] + Sin[x[t]] == 0, x[0] == 2, x'[0] == 0}, x, {t, 0, 20}]

c0.35

1. Energy decay

Type = at the start of any cell, or use the bar at the bottom of the window, and one sentence becomes a real notebook: title, prose, equations, sliders, plots, evaluated in place. Bring your own Anthropic key or run a local Ollama model, fully offline.

Drag it. The math answers.

In OpenMat, every slider re-evaluates through the kernel on each tick. This one is wired up right here on the page: the same damped oscillation the notebook above set up.

In[2]:=
Plot[Exp[-c t] Cos[2 Pi t], {t, 0, 8}]
c = 0.35
Out[2]=

What the kernel does today

v0.01 is a young kernel with honest edges: ODEs are scalar first and second order, Solve reaches quadratics. What is listed here works, in the app and in the browser build. The complete notation guide, with real kernel output for every form, is in the language reference.

Symbolic

D[f, x]
derivatives
Integrate[f, x]
antiderivatives
Integrate[f, {x, a, b}]
definite integrals, with trig rules and a numeric fallback when no closed form exists
Solve[eqn, x]
exact roots through quadratics
Expand · Factor · Simplify
polynomial algebra

Numeric

Plot[f, {x, a, b}]
adaptive sampling, legends, asymptote splitting
ListPlot[data]
discrete point markers
NDSolve[{eqs}, x, {t, a, b}]
first and second order ODEs, Dormand-Prince or SUNDIALS
Manipulate sliders
bind any symbol; the cell re-solves live as you drag

Session

a = 5
definitions persist across cells
f[x_] := x^2
pattern definitions, visible inside Plot and NDSolve
Clear[a]
until you say otherwise
Table[n^2, {n, 1, 8}]
list construction

Natural language

= plot sin x from 0 to 10
one expression, evaluated inline
= make me a notebook about…
a full notebook: title, prose, sliders, plots
Your key or your machine
Anthropic API key, or a local Ollama model with nothing leaving your laptop

Established at Sundai Hack #134

OpenMat was established at Sundai Hack #134 on August 2nd, 2026: the kernel, the notebook, the v0.01 release, and this site, shipped and presented the same day.

The OpenMat team presenting on stage at Sundai Hack #134, with the OpenMat project on the lecture hall projector and the Sundai banner beside them.
Out[3]=Sundai Hack #134, August 2nd, 2026.

Install

macOS app

  1. Download OpenMat_0.1.0_aarch64.dmg
  2. Drag OpenMat to Applications
  3. First launch: right-click OpenMat.app, choose Open, then Open again

The build is unsigned, which is why macOS asks once. Apple Silicon only for now.

From source

git clone https://github.com/dhve/OpenMat.git
cd OpenMat/app && npm install
npm run tauri dev   # desktop app
npm run dev         # browser, same kernel

Rust workspace tests: cargo test --workspace. Needs cmake for the vendored SUNDIALS build.