Build, simulate and understand complex systems.
OpenMathLab brings numerical computing, modeling, simulation and advanced engineering workflows into one browser workspace.
OpenMathLab brings numerical computing, modeling, simulation and advanced engineering workflows into one browser workspace.
Choose a cloud API or a local model running on the OpenMathLab server computer. You control when a message or code is shared.
Keys stay in server memory only, and are cleared on restart or by the Remove key button.
Your conversation appears here. Messages are not saved to SQLite or project files.
A response containing a fenced MATLAB or Octave code block can be previewed. AI cannot change or execute your project without your explicit approval.
No code change proposed.
AI calls can incur provider charges. Include code only when you intend to share it with the selected provider.
Ollama, LM Studio and vLLM addresses refer to the machine running OpenMathLab. If OpenMathLab is hosted remotely, localhost on your personal PC is not accessible to the server. Remote-user local models need a separate trusted local connector. Each visitor has separate private AI preferences in no-login mode.
Your GPU, your choice. GPU operations run in your browser or on your own separate GPU worker — never on the machine hosting OpenMathLab.
The GPU token is never sent to OpenMathLab's Python server, SQLite database, or project ZIP. Keep it private. For remote GPU workers use HTTPS and explicit allowed origins.
Select a GPU mode, detect hardware, then run a sample.
WebGPU supports compatible browsers and hardware. The separate GPU worker requires a GPU-enabled PyTorch installation on your machine and your configured CORS origin. Browser WebGPU currently supports matrix multiplication and vector addition; the separate worker supports eight bounded operations. General MATLAB scripts, arbitrary Simulink models and full GPU toolboxes still run through the existing CPU runtime and are not silently moved to a GPU. The browser-only status journal does not independently verify GPU worker claims, and running GPU kernels cannot be force-cancelled. Public Internet deployment still requires additional security hardening.
No accounts, registration, or passwords. Each browser has a private visitor identity. Share revisioned projects using one-time invitations; ZIPs always remain portable.
Start a collaboration project, then generate a one-time invitation code for a teammate. The teammate pastes the project ID and invite code to join. This shares revisioned text changes, not the full project ZIP or binary datasets.
Published text is journaled in SQLite. Use Download snapshot for a portable ZIP; binary files must still be exported from the main project manager.
Project members and revision details will appear here.
This is revision-based collaboration with conflicts detected, not simultaneous CRDT co-editing. Keep downloading ZIPs to preserve complete projects.
Account and workload counts without revealing anyone's scripts, API keys, files, or GPU credentials.
Select Refresh metrics to inspect this installation.
This dashboard does not enable public sandboxing, automatic remote GPU execution, or distributed scheduling. Public access requires additional hardening.
Preferences persist on this OpenMathLab installation in SQLite; your project data remains file-based.
Auto-save does not save to a server project database. Download your project ZIP to preserve work permanently.
A clear view of what lives where.
Projects stay in your browser and downloaded ZIPs; optional small text-only recovery drafts are stored in this browser when autosave is enabled. Binary files are not in drafts. No SQLite project storage.
Stores appearance settings and AI provider URL/model metadata by anonymous browser profile. No AI keys.
Held temporarily in Python server memory and sent only to the configured AI provider. Never exported in a ZIP.
Chat text is held in the open page; requests are sent to your configured model endpoint. Remote providers may retain data according to their policies.
Open engineering computing in a single Python-served environment.
Python, FastAPI, HTML, CSS, JavaScript, Apache Arrow/Parquet (optional), and SQLite preferences. No Node/npm required.
Switch workspaces with the top navigation. In Workspace, select main.m, Command Window, or Result. Settings are separated from engineering project files.
OpenMathLab has free no-login private guest workspaces. Public shared hosting requires hardened isolated workers, restricted network access and independent security verification. Guest session isolation alone is not a sandbox.
>> OpenMathLab ready. >> Run a script or enter commands from the workspace.
No results yet.
Ready.
Browse CSV without optional packages. For Parquet and Arrow, install Apache PyArrow.
Review your dataset at a glance. Use the left navigation to inspect the schema, browse a sample or export to another format.
Types are inferred by Apache Arrow when available. Standard CSV inspection without PyArrow reports string types.
| # | Column | Type |
|---|
Showing up to 50 records; filtering applies to the preview only.
Query the opened CSV, Arrow, or Parquet dataset using safe, bounded SQLite SELECT statements. The table is named data; all processing is in memory.
Conversions happen in memory. Files are downloaded directly to your computer.
Digital twins, experiment design, uncertainty, surrogate models, provenance and agent-assisted engineering workflows.
Write. Execute. Explain. Export.
Review isolated CPU jobs and client-reported personal GPU jobs. GPU requests never pass through the hosting compute server.
Uses the script from Workspace → Code. The worker has no network, host mounts or GPU access. It is stateless and cannot change your live Workspace.
Personal GPU statuses are reported by your browser, not independently verified by this server. Continue checking your own GPU worker for authoritative status; this journal never saves tokens or numerical inputs.