RunMat

Math and physics calculation engine designed to be backwards-compatible with MATLAB syntax. Automatic GPU acceleration. Browser, desktop, and CLI.

[RunMat @runmat_com]

Try it for yourself at https://t.co/4a7Rdsd8jp
Paste the code below

XRange = -2:0.04:2;

YRange = -2:0.04:2;

[X, Y] = meshgrid(XRange, YRange);

R = sqrt(X.^2 + Y.^2) + 0.01;
for t = 0:0.05:10
Z = sin(pi*R*3 - t*2) ./ (1 + R) + cos(pi*R*2 + t*1.5) ./ (1 + R*2);

surf(X, Y, Z);

shading interp;

colormap("jet");

pause(0.03);

end

[RunMat @runmat_com]

Engineers use wave interference like this to design noise-canceling headphones, radar systems, and 5G antennas.

The key ingredient? #Pi. It controls the frequency, the phase, and the cancellation.

This video shows two pi-frequency waves colliding in real time. GPU-accelerated, running in a browser, 12 lines of code.

Stay tuned for more on Saturday, #PiDay.
#RunMat

[RunMat @runmat_com]

Every #MATLAB #engineer has done this: write 50 lines, hit run, wait, get "Matrix dimensions must agree" on line 6. Go back. Stare. Fix it. Run again. #RunMat catches it before you run.

Hover over any variable and you see its shape. Write a dimension mismatch, and you get the error inline, in real time, with the exact sizes: "left is 1x201, right is 1x201, inner dims 201 and 1." No run. No wait. No guesswork.

Software engineers have had this for decades -- TypeScript, Rust, and Go all catch type errors before execution. #Numerical engineers deserve the same. That's what we're building.

[RunMat @runmat_com]

eig function doc: https://t.co/SJZudYQ7qD

eig source code: https://t.co/nC5wZQUTo0

[RunMat @runmat_com]

MATLAB: eig(A) → eigenvalues. How? Trust us.
RunMat: eig(A) → eigenvalues. How? Here's every step. On GitHub. In Rust.

Eigenvalues tell you the fundamental modes of a system: vibration frequencies, principal components, and stability characteristics. Simple question. Hard computation.

What RunMat does under the hood:
1. Schur decomposition: Reduce to upper-triangular form.
2. Back-substitution: Solve for each eigenvector column by column.
3. Normalization: Handle repeated eigenvalues and near-singular edge cases.
4. GPU awareness: Keep results on-device when a provider is available.

That's 900 lines of Rust. Every edge case (repeated eigenvalues, near-singular matrices, complex spectra) handled explicitly in code you can read.

Engineers should have full traceability: from the function call, through the algorithm, down to the machine epsilon constant. If you can't trace it, you can't trust it. 300+ functions. Zero black boxes.

[RunMat @runmat_com]

Try it for yourself at https://t.co/iM3FzUmnus

% Generate a high-density parametric mesh (250k points)
[u,v] = meshgrid(linspace(0,2*pi,500), linspace(0,pi,500));

% Define the 3D heart surface equations
x0 = 16*sin(u).^3 .* sin(v).^2;
y0 = (13*cos(u)-5*cos(2*u)-2*cos(3*u)-cos(4*u)) .* sin(v).^2;
z0 = 6 * cos(v);
% Animate the heartbeat
for t = 0:0.1:100
% Calculate scale factor for contraction/expansion

s = 1 + 0.05*sin(t*3);
% Render the scaled point cloud in red
scatter3(x0(:)*s, y0(:)*s, z0(:)*s, 2, 'r');
pause(0.01);
end

[RunMat @runmat_com]

Happy #Valentine's Day from the team that loves math. ��️

This isn't a pre-rendered animation. It's live numerical computation.

We generated a parametric heart with 250,000 points and made it beat.

Notice the fluidity? That's what we think modern computation should feel like. Not just solving equations, but exploring them in real-time.

You can rotate, zoom, and dive deep into the structure without it ever skipping a beat.

We put our heart into this runtime so you can put yours into the #math.
(Code in next comment)

[RunMat @runmat_com]

Read the full argument: https://t.co/7z4hn63iMd

[RunMat @runmat_com]

Engineers don't need "better" syntax than MATLAB.

`A * B` is already the perfect way to write matrix multiplication.

The problem isn't the language. It's the runtime.

We built RunMat to fix that: Same whiteboard syntax. Modern, GPU-accelerated engine.

[RunMat @runmat_com]

Introducing RunMat Accelerate: The Fastest Runtime for Your Math
https://t.co/YAZ8oWJGlQ

[Nabeel Allana @nabeelallana]

We may have just built the fastest runtime to run math in the world. The initial benchmarks (on my local MacBook pro) are looking stellar 🔥

Here’s how it works, and how we built it 🧵

[Nabeel Allana @nabeelallana]

@kevinweil We built the entire Matlab runtime (it’s over 3M chatarcters of code) in 6 weeks using it.

https://t.co/V30PWHer2J

It’s open source. Matlab programming is now free — like Python.

We’re making it the fastest math runtime in the world — JIT for CPU logic and auto-GPU shaders.

[Nabeel Allana @nabeelallana]

@DaveShapi Think we may already be at senior. Did a side project to test the bounds — been programming for a very long time, and am a founder.

There’s little to no room for junior / mid. Even senior I think is hairy. I’d only want architects / great supervisors.

https://t.co/GY310H9FZO

[Michael J.J. Tiffany @kubla]

The rust-ification will continue until memory safety improves

[Joris Poort @jorispoort]

Cool new open source MATLAB runtime. Some CUDA acceleration too!

[Nabeel Allana @nabeelallana]

✅ Fully functional MATLAB runtime
✅ All grammar & core semantics implemented
✅ Functional JIT + GPU acceleration framework
✅ Tiny ~5MB binary, runs in 5ms
✅ Container-ready, cross platform (ARM/x86_64/...)
✅ 150-180x faster than Octave
✅ Typed computation graphs (IYKYK)

[Nabeel Allana @nabeelallana]

MATLAB makes $500M annually selling what I just re-engineered in 3 weeks.

Here’s how I demonstrated that enterprise software moats today can evaporate in the same time it typically takes to plan a sprint. 👇🧵

[RunMat @runmat_com]

@FluentInFinance Would add Dystr to the list! Engineers and scientists love it!

[RunMat @runmat_com]

@paulg Classic innovators dilemma!