Skip to content

Introduction ​

Ergalics Studio is a professional scientific-computing workstation that runs entirely in the browser. It pairs a React + TypeScript frontend with a Rust core compiled to WebAssembly and a WebGPU compute pipeline, and it treats third-party extensibility as a first-class concern.

Goals ​

GoalWhat it means
Browser-nativeNo install, no server required; data lives in IndexedDB and can be shared as .clproj files.
Extensible by designEvery visualization is a plugin behind a small, typed contract; third-party packages are isolated from the host page.
Industrial hygieneStrict TypeScript, unit + E2E tests, a deterministic example-data pipeline, and generated, versioned WASM bindings.

Current state ​

The project is under active development and already usable end to end. The following are functional today:

AreaFunctional today
Full workbench loopProjects, a file explorer with per-file editors and a tabbed central area, file routing, plugin registry, parameter panels, status/perf bars.
Four workbench modesStandard (drag → see), Flow (compose a dataflow DAG → run), Block (Scratch-style scripted editor with a Run hat, 40 built-in blocks, shared IR, and 5 sample programs), and Code (free-form Python on a Pyodide CPython Worker, with a REPL console and 9 sample programs). All three scripted modes share one IR, and Block ↔ Flow ↔ Code round-trip is pinned by a sync-threeway test. Files carry the mode: .py/.js/.r open in a Monaco editor, *.flow.json in the Flow DAG editor, *.block.json in the Blockly editor.
163 built-in plugins(153 product plugins + 10 fun/utility), organised into 19 sidebar disciplines, covering the 2D and 3D rendering paths — including a 3-D N-body gravity simulator (astrophysics), a protein interaction network with force-directed layout (systems biology), D2Q9 lattice-Boltzmann fluid / wave-equation / double-pendulum simulations, an offline GeoJSON choropleth map, three interactive physics labs (electromagnetism, optics, structural mechanics), an in-browser AI trainer (linear / non-linear NN / logistic / MNIST CNN on TensorFlow.js), and two field-grade competition solvers: a sparse Hermitian eigenvalue solver (em-eigensolver) and a 1D–3D bidirectional fluid-network coupler (fluid-cfd-coupler), both run inside Pyodide workers with millisecond-level control tooling.
Scientific computing subsystemsPure TypeScript: a statistics kernel (hypothesis tests, effect sizes, corrections, power analysis), scientific binary I/O (HDF5 / NetCDF / FITS / Zarr / Parquet), a publication-grade SVG/PDF plot engine, and a reproducibility kernel (seeded RNG + run manifests).
WebGPU compute pipelineReal compute: GpuBuffer management and kernel compile/dispatch/run from Rust, an api.gpu compute surface for plugins, and reusable WGSL templates with CPU fallbacks — accelerating Particles, N-Body Gravity, the LBM fluid, the wave equation, histogram, heatmap and point-cloud kernels.
Plugin sandboxWorker-based, with a typed RPC protocol.
Git-backed projectsA workspace VFS (src/core/vfs/) shared by the file explorer, per-file editors and the Git working tree; isomorphic-git in the browser and simple-git under Node, with a sidebar Git panel (branch / commit / clone / remote sync / auto-commit / preinstalled hooks).
i18nzh-CN / en-US with reactive locale switching — Block mode uses Blockly's BKY_* key system so block labels re-localise with the rest.
Theming / sharing / CI / testsTheming (dark / light), sharing, GitHub Actions CI, and automated tests (3878 Vitest test cases across 241 test files plus Playwright E2E suites).

Built but not yet shipped as a product: the fully-automated third-party submission & install pipeline for the plugin marketplace (the .cspkg package-signing gate — ed25519, FR-05 — is already live; see SECURITY.md). R for Code mode is a full webR runtime (vendored in public/webr/, falling back to the built-in IR engine when the bundle is absent) — see guide/roadmap.

The intent is that the codebase keeps growing into a production system by adding features into this structure — not by rewriting it.

Released under the MIT License.