Woxi Bets Rust and AI Can Clone the Wolfram Language

Woxi, a solo-maintained Rust interpreter for the Wolfram Language, has returned to Hacker News' front page, with its commit history substantially written by Claude. The AGPL-3.0 interpreter covers about 1,250 of roughly 1,400 tracked Wolfram functions and offers instant startup, a static binary, and WASM support, positioning it as an embeddable runtime rather than a free Mathematica. Its author argues AI agents can improve code quality, backed by a test suite of ~26,000 unit tests and ~900 script snapshot tests.

Dev Tools https://sourcefeed.dev/c/dev-tools Article Woxi Bets Rust and AI Can Clone the Wolfram Language The largely Claude-written interpreter is a real embedding win — and a stress test for AI-scaled reimplementation. Lenn Voss https://sourcefeed.dev/u/lennart voss The Wolfram Language has been reimplemented before. What makes Woxi https://woxi.ad-si.com — back on Hacker News' front page this week with a Show HN — worth your attention isn't that it's a free Mathematica. It's how it exists at all: a solo-maintained Rust interpreter whose commit history is, by the author's own admission and the contributor graph's testimony, substantially written by Claude. Woxi is simultaneously a genuinely useful tool and one of the clearest early tests of whether AI-scaled reimplementation of a decades-old proprietary platform actually works. What it is, concretely Woxi is an AGPL-3.0 interpreter for the Wolfram Language https://github.com/ad-si/Woxi written in Rust. You install it with cargo install woxi or prebuilt binaries, or pip install woxi , or npm install woxi-wasm , and then woxi eval "1 + 2" does what you'd expect — in milliseconds, because there's no kernel to boot and no license server to phone. It runs .wls scripts and, notably, .nb notebook files directly, which wolframscript itself won't do. There's a REPL, a Jupyter kernel, a browser playground running entirely client-side on WASM, and as of this release, Woxi Studio — a native notebook editor built with the iced GUI toolkit that exports to .ipynb , Markdown, LaTeX, Typst, and PDF. Coverage is real but bounded. The project's own tracking file lists roughly 1,400 Wolfram functions, of which about 1,250 are marked implemented. That sounds like 90% until you remember the Wolfram Language ships north of 6,000 built-ins — Woxi tracks a curated subset, and the gaps are telling. FinancialData and WordData are unimplemented because they can't be: they're front-ends to Wolfram's proprietary Knowledgebase, a server-side asset no interpreter can clone. If your notebooks lean on curated data, Woxi will never save you. The "free Mathematica" framing is the wrong one The obvious pitch — open-source Wolfram — is weaker than it sounds. Mathics3 https://mathics.org/ has been an open-source Wolfram Language implementation in Python since 2011 and is still actively maintained. And Wolfram itself gives away the Wolfram Engine https://www.wolfram.com/engine/ free for pre-production development, with an explicit carve-out for open-source projects. Nobody was priced out of experimenting with the Wolfram Language. What you couldn't do before Woxi is treat the Wolfram Language as an embeddable, redistributable runtime. The free Wolfram Engine requires activation against a Wolfram ID, forbids production use, and starts a full kernel — seconds of latency that make it useless for short-lived CLI invocations, cron jobs, or serverless handlers. Woxi's actual wedge is the boring systems stuff Rust buys you: instant startup, a single static binary, and a WASM build that puts a Wolfram-ish evaluator in a browser tab or a Node process with no license entanglement. That's a new capability, not a cheaper old one. If you've got a pile of .wls glue scripts, the migration test is trivial: run them under both binaries and diff the output — which is exactly how the project itself validates. The AI elephant When Woxi first hit HN in March, the discussion got spicy for a reason that had nothing to do with symbolic math. Commenters noticed the GitHub contributor graph attributed on the order of 300,000 lines to Claude, and the author didn't hide it — they argued AI agents can improve code quality and pointed to the test suite as evidence. That suite has since grown to a claimed ~26,000 unit tests plus ~900 script snapshot tests, with a CI rule that every CLI test must pass identically under Woxi and WolframScript. Both sides of that argument deserve to be taken seriously. Differential testing against the reference implementation is the strongest conformance strategy available to a clean-room clone — it's how you'd want this built. But HN's skeptics landed real punches too: thousands of generated, lightly-reviewed tests measure agreement on the cases someone thought to generate, not correctness on the ones nobody did. One commenter with CAS experience found the symbolic integration routine naive — not even attempting partial fraction decomposition for rational functions. That's the pattern you'd predict from AI-scaled development: excellent breadth on shallow, well-specified behavior string functions, list manipulation, syntax , and thin depth where the hard math lives. Mathematica's actual moat was never its parser; it's forty years of algorithmic work on integration, equation solving, and special functions. Snapshot tests can tell you Integrate returned something WolframScript also returns for tested inputs — they can't conjure the algorithm for the untested ones. There's also a deliberate strategic thorn: the AGPL. Asked in March why he didn't pick a license Wolfram Research could use, the author replied, "Why should I want them to use it?" Fair enough — but AGPL cuts users too. Embedding woxi-wasm in a proprietary web product triggers copyleft obligations that a commercial team's lawyers will flag immediately. The license choice optimizes for spite against Wolfram over adoption by industry, and that's a real cost for the embedding use case that is otherwise Woxi's best story. Where this lands My read: Woxi is production-plausible today as a scripting and embedding runtime for the Wolfram subset you can exhaustively test yourself, and nowhere close as a computer algebra system you'd trust with research. Use it to un-license your CI scripts, to run legacy .nb files headless, or to teach Wolfram syntax in a browser with zero installs. Don't use it where a silently wrong integral costs you something — there, the free Wolfram Engine or a mature open CAS like SymPy remains the honest choice. The larger significance is the development model. Reimplementing a proprietary language surface used to be a decade-long community project — Mathics' fifteen years, Octave's thirty. Woxi compressed the breadth phase into months with over 8,000 commits, most of them machine-written, validated by differential testing against the original. Whether the depth phase — the actual mathematics — yields to the same approach is the open question, and I'd bet it mostly doesn't without human algorithmic work. But every proprietary vendor whose moat is a large, well-documented API surface should be looking at Woxi's contributor graph and feeling slightly cold. The breadth moat just got a lot cheaper to cross. Sources & further reading - Show HN: Woxi - Open-source Mathematica / Wolfram Language reimplementation https://woxi.ad-si.com — woxi.ad-si.com - Show HN: Woxi discussion thread https://news.ycombinator.com/item?id=49270040 — news.ycombinator.com - ad-si/Woxi: Wolfram Language / Mathematica reimplementation in Rust https://github.com/ad-si/Woxi — github.com - Woxi: Wolfram Mathematica Reimplementation in Rust March 2026 discussion https://news.ycombinator.com/item?id=47155526 — news.ycombinator.com - Free Wolfram Engine for Developers https://www.wolfram.com/engine/ — wolfram.com - Mathics3: A free, open-source alternative to Mathematica https://mathics.org/ — mathics.org Lenn Voss https://sourcefeed.dev/u/lennart voss · Cloud & Infrastructure Writer Lenn writes about cloud platforms, Kubernetes internals, and the infrastructure decisions that quietly make or break engineering organizations. Based in Berlin's vibrant tech scene, they have a talent for turning dense platform-engineering topics into prose that people actually finish reading. Discussion 0 No comments yet Be the first to weigh in.