Woxi 0.3 runs Mathematica notebooks on a Rust engine, and bets on distribution

Adrian Sieber released Woxi 0.3.0 on August 6th, moving his open-source Wolfram Language reimplementation from accumulating functions to executing complete Mathematica notebooks ^[1][2][3]. The release lets the native Woxi Studio editor open and run a selection of interactive Wolfram Demonstrations end to end, and adds prebuilt Windows binaries alongside the existing Linux and macOS builds ^[4][5]. For teams that want reproducible Wolfram Language scripts without a Wolfram kernel, the interesting part is where the engine can run, not what it is named after.

Sieber, a Frankfurt-based developer and the CEO of the hosted SQLite service Airsequel, is taking on a proprietary language that Wolfram has developed since 1988 ^[6][7]. The distribution strategy is the bet. The repository packages the interpreter as a command-line tool, a Jupyter kernel, a Python package, an npm package and a WebAssembly module, so the same Rust engine runs inside a shell script, a notebook, a Python process, a Node application or a browser ^[8][9]. The browser playground executes locally rather than sending computations to a remote server, according to the documentation ^[10]. The Python package reached 0.3.0 on August 7th with prebuilt wheels for Windows, macOS and Linux, exposing symbolic evaluation, persistent session state and SVG graphics through Python calls ^[11][12]. The whole thing is licensed AGPL-3.0-or-later, which keeps modifications to deployed versions subject to the license's source-sharing requirements ^[13].

The 0.3.0 work concentrates on what happens when you open a real .nb file. Woxi Studio can reconstruct stored Manipulate widgets, interpret typeset notebook input as executable code, and render interactive controls including locators, buttons, menus and animations ^[14]. Sieber used demonstrations covering Kepler's second law, parabolic mirrors, Doyle spirals and damped pendulums as end-to-end tests ^[15]. The graphics additions reach past basic plots to error bars, fills, plot markers, three-dimensional surfaces, polygon holes, SVG font embedding and Wolfram-style automatic ticks, alongside more symbolic integrals, differential-equation cases, constrained optimization and equation solving ^[16][17]. The release notes say several rounds of differential fuzzing against wolframscript exposed silent differences in arithmetic, expression ordering and simplification ^[18]. That comparison work matters more than the function count, because a matching function name is worth little if evaluation or output diverges from existing Wolfram code.

Woxi also makes deliberate departures: its command-line program can evaluate input cells directly from .nb notebooks, which WolframScript does not support, and Woxi Studio automatically executes preceding cells rather than allowing out-of-order evaluation that depends on hidden kernel state ^[19][20]. The editor drops Mathematica's % shorthand for the previous result ^[21].

The gap remains large. Woxi's documentation lists the Wolfram Knowledgebase, machine learning, neural networks, finite-element methods, chemistry, video computation and signal processing as mostly unimplemented, and Mathematica 15.0 additions such as GPU kernels, WebSocket connectivity, a rebuilt time-series engine and agent tooling are outside its coverage ^[22][23]. The project says it fully supports Mathematica 1.0's original 554 built-in functions and symbolic core, a useful baseline but a decades-old one ^[24]. Woxi is also not the first alternative engine here: Mathics3, a Python implementation, dates to 2011 ^[25].

What to watch: whether Woxi supports a narrow set of reproducible notebooks well enough that developers embed it, and how the AGPL terms sit with commercial deployment plans.