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

• Browser-native: no install, no server required; data lives in IndexedDB and can be shared as .clproj files.
• Extensible by design: every visualization is a plugin behind a small, typed contract; third-party packages are isolated from the host page.
• Industrial hygiene: strict 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:

• The full workbench loop: projects, file routing, plugin registry, parameter panels, status/perf bars.
• Four workbench modes — Standard (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.
• 59 built-in plugins (49 scientific + 10 fun/utility) 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 subsystems in pure 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).
• A real WebGPU compute pipeline: 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.
• A Worker-based plugin sandbox with a typed RPC protocol.
• i18n (zh-CN / en-US) with reactive locale switching — Block mode uses Blockly's BKY_* key system so block labels re-localise with the rest.
• Theming (dark / light), sharing, GitHub Actions CI, and automated tests (2233 Vitest unit tests across 123 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.