RunMat
GitHub
MATLAB online alternative

Run MATLAB code in the browser blazing fast

RunMat executes MATLAB-syntax code in your browser with auto GPU acceleration. No account or licence needed.

MATLAB Online has friction

MATLAB Online runs your code on MathWorks' servers, requires an account, and caps free usage at 20 hours/month with 15-minute idle timeouts. Engineers and students hit these limits regularly:

Account barriers

Sign-up process and license requirements create unnecessary friction for quick tasks.

Idle timeouts & hour caps

Sessions timeout after 15 minutes of inactivity. Free tier is capped at 20 hours/month.

Cloud dependency

Code must be uploaded to remote servers, raising privacy and connectivity concerns.

No local GPU access

Code runs on MathWorks' servers, so you cannot use your own GPU for acceleration.

Specialty toolboxes cost extra

Specialty toolboxes (Image Processing, Signal Processing, Optimization, etc.) are paid add-ons on top of base MATLAB.

No built-in agent

MathWorks' Agentic Toolkit is a separate product for local MATLAB installs (Claude Code, Copilot, Codex, Gemini, or Amp). Its MCP returns command-window text — no plot or workspace introspection.

Meet RunMat: no license required

RunMat is an open-source runtime that understands MATLAB syntax and runs it directly in your browser. Your code executes on your own device via WebAssembly, with GPU acceleration in browsers that support WebGPU.

Built-in agent

Describe the math, or paste an existing MATLAB script. The agent writes new code or rewrites around toolboxes and builtins RunMat doesn't yet ship. Every change comes back as a diff.

No account required

Open the sandbox and start coding immediately. No sign-up, no license key.

Client-side execution

Your code compiles to WebAssembly and runs locally in your browser. Nothing leaves your device unless you save to the cloud, and after the initial page load it works offline.

GPU acceleration on any vendor

Metal on Mac, Vulkan on Linux, DirectX 12 on Windows. WebGPU in the browser. No CUDA dependency.

Learn more

Interactive 2D & 3D plotting

GPU-rendered surfaces you can rotate, zoom, and pan. Plots live in the same computation chain as your math.

Type & shape tracking

Hover any variable to see its dimensions. Mismatched matrix sizes get red underlines before you run.

An engineering team, working alongside you

The agent runs your scenarios, reviews the code, and searches the project for what you need. You explore problems that used to take a team.

Runtime-aware inspection

Checks tensors for NaN, verifies shapes against expected dimensions, and reads plot data from any 3D camera angle.

Every change is reviewable

Edits are presented as diffs. Accept or reject each change. Conversations are stored as searchable project files.

How it works

Get started in three simple steps.

1

Open the sandbox

Go to runmat.com/sandbox. The editor loads in your browser; no account or license needed.

2

Write code, or ask the agent

Type MATLAB-syntax code directly, or describe what you want to compute and let the built-in agent write it for you.

3

Run on GPU and iterate

Execute on the same GPU-accelerated runtime. Inspect plots and variables, then iterate until the math is right.

Every change versioned. No git required.

Every save creates a version automatically. Per-file history and full project snapshots are included on all App tiers, starting at $0 with 100 MB on the Hobby tier. Paid plans add project sharing with your team -- no git setup or merge conflicts.

Automatic file history

Browse the timeline and restore any previous state. No commits, no staging area.

Learn more

Project snapshots

Capture your entire project in one click. Restore instantly, even across terabyte-scale datasets.

Learn more

App project sharing

Share projects with colleagues instantly. No shared drives, no emailing files back and forth.

Learn more

The fastest runtime for your math

RunMat runs math faster because of how the runtime is engineered. Fusion merges sequential operations into fewer GPU steps; residency keeps your arrays on-device between steps. That means less memory traffic, fewer program launches, and faster scripts.

RunMat vs. MATLAB Online

RunMat runs in your browser with a built-in agent and GPU acceleration, no account needed. MATLAB Online runs on MathWorks' servers behind a paid license and caps free usage.

RunMat

Open-source runtime with a built-in agent

  • Open-source runtime (MIT)
  • No account or license required
  • Built-in agent
  • Agent reads workspace variables and 2D/3D plot scenes
  • Client-side execution
  • Cross-platform GPU (Metal, Vulkan, DX12, WebGPU)
  • Works offline
  • Interactive 2D & 3D plotting
  • Real-time type & shape tracking
  • Execution tracing & diagnostics
  • Automatic file versioning & snapshots (App)
  • Agent rewrites scripts around unsupported toolboxes & builtins
  • Limited package / toolbox support
  • Subset of MATLAB functions

MATLAB Online

MathWorks' cloud-hosted MATLAB IDE

  • Requires paid MATLAB license
  • Account & sign-in required
  • No built-in agent
  • Free tier capped at 20 hours/month, 15-min idle timeout
  • Cloud-based execution
  • GPU support available
  • Requires internet connection
  • Full MATLAB language
  • Complete toolbox ecosystem
  • Official MathWorks support
  • No built-in file versioning

What works today

Core matrix workflows, plotting, and debugging ship today. Here is what is in progress.

Works well

Matrix and array operations (indexing, slicing, reshaping)

Arithmetic, logical, and relational operators

Control flow (if/else, for, while, switch)

User-defined functions with multiple outputs

Cells, structs, and basic classdef OOP

400+ built-in functions

20+ plot types (plot, scatter, surf, mesh, contour, contourf, imagesc, bar, pie, stem, quiver, area, errorbar, plot3, semilogx/semilogy/loglog, and more) with GPU acceleration, subplots, figure handles, and interactive 3D camera

Real-time type and shape tracking (hover to see matrix dimensions)

Live syntax validation (red underlines for dimension mismatches and errors)

Execution tracing and diagnostic logging

Async code execution (non-blocking runs)

Limitations & future work

In progress

Extensible package support (signal processing, optimization, etc.)

Some edge-case MATLAB semantics

Not supported

Simulink or graphical block diagrams

MATLAB-specific file formats (.slx, .mlapp)

Java/COM interop

For toolboxes and builtins in the “in progress” list above, the built-in agent can often rewrite scripts to use what RunMat already ships, usually preserving the intent.

For a detailed list, see the language coverage guide and function reference. For workflow guides, browse resources or read our deep dives on GPU acceleration and the FFT family in RunMat.

Built for teams

SSO & SCIM

Integrate with your identity provider. Provision and deprovision users automatically.

ITAR-compliant deployment

Self-hosted, air-gapped option for export-controlled environments.

Open source & auditable

MIT-licensed runtime. Inspect every line of code that runs your math.

SOC 2 ready

Built to SOC 2 standards. Audit planned for Q4 2026.

Frequently asked questions

Common questions about RunMat and MATLAB compatibility.

Is RunMat the same as MATLAB?
No. RunMat is an independent project with an open-source runtime that executes MATLAB-syntax code. It is not affiliated with or endorsed by MathWorks, the makers of MATLAB.
Does RunMat have an AI assistant?
Yes. RunMat ships with a built-in agent in the sandbox. You describe what you want to compute; the agent writes MATLAB-syntax code, runs it on the same GPU-accelerated runtime, reads workspace variables and 2D/3D plot scenes, and iterates. Every change comes back as a diff you can accept or reject.
Does MathWorks have an AI assistant?
As of May 2026, MathWorks publishes the MATLAB Agentic Toolkit for Claude Code, GitHub Copilot, OpenAI Codex, Gemini CLI, and Sourcegraph Amp. MathWorks lists MATLAB R2020b or later, a supported AI coding agent, and Git as prerequisites, and its product page says the toolkit requires a local MATLAB installation and an AI service subscription. MathWorks' README says MCP servers must be used with MATLAB under the MathWorks Software License Agreement and must not be shared by multiple users; the repository license also limits the software to use with MathWorks products and service offerings.
How is RunMat's agent different from the MATLAB Agentic Toolkit?
RunMat's runtime is open source under MIT and runs without a MATLAB license. MathWorks documents the Agentic Toolkit as skills plus a MATLAB MCP Core Server connection to a local MATLAB installation and a supported third-party agent. The MCP Core Server exposes five built-in tools for static analysis, code evaluation, file execution, test execution, and toolbox detection; MathWorks also says MCP servers must not be shared by multiple users. RunMat ships the agent and runtime together in the browser, so the agent can inspect RunMat workspace variables, tensor shapes, and plot scenes directly.
Can I run my existing MATLAB scripts in RunMat?
Many MATLAB scripts run without modification, especially those using core language features and common built-in functions. For scripts that use functions RunMat doesn't ship yet, the built-in agent can help adapt the code — running your script, reading the diagnostics, and proposing reviewable edits that replace unsupported calls with supported equivalents. Try your script in the sandbox; if something doesn't run, ask the agent to help adapt it.
Is RunMat really free?
The RunMat runtime is open source under the MIT license. You can run MATLAB-syntax code in the browser without usage fees or time limits. App features like storage, versioning, and project sharing start on the Hobby tier (100 MB). Paid plans add more storage and team features -- see pricing.
Do I need to create an account?
No. The browser sandbox works immediately without sign-up. Creating a free account unlocks cloud storage (100 MB), file versioning, and project sharing.
Does RunMat work offline?
Yes. Once the sandbox page loads, it can run code without an internet connection. For full offline use with local file access, use the RunMat CLI today. The desktop app with a full IDE experience is coming soon.
How does RunMat run code in the browser?
RunMat compiles to WebAssembly, which runs natively in your browser at near-native speed. For GPU-accelerated operations, it uses WebGPU (available in Chrome, Edge, Safari 18+, and Firefox 139+).
Is my code private?
Yes. Code execution happens entirely on your device. Nothing is uploaded to run your script. If you sign in for cloud storage, files sync to RunMat's servers. Using the built-in agent sends context to the configured LLM provider.
Can RunMat use my GPU?
Yes, in browsers that support WebGPU (Chrome, Edge, Safari 18+, and Firefox 139+). RunMat automatically offloads eligible operations to the GPU for faster execution.
What’s the difference between RunMat and GNU Octave?
Both run MATLAB-style syntax, but RunMat is designed for performance (JIT compilation, GPU acceleration) and runs natively in the browser. Octave is a mature desktop application with broader toolbox compatibility but no browser-native execution.
Is there a desktop version?
The RunMat desktop app is coming very soon. It will provide the same interface as the browser sandbox with full local file system access. In the meantime, the CLI is available today for local script execution.
Does RunMat support plotting?
Yes. RunMat supports 20+ interactive plot types including plot, scatter, hist, histogram, surf, mesh, contour, contourf, imagesc, bar, pie, stem, quiver, area, errorbar, plot3, semilogx, semilogy, and loglog, all GPU-accelerated. 3D surface plots can be rotated, zoomed, and panned directly in the browser. See the plotting guide for runnable examples.

Try RunMat for free

Write code or describe what you want to compute. The sandbox is free, no account required.