The result is https://xllify.com

It embeds Luau which tames the very old school C SDK. This means that the same code can run as native add-ins on Windows (XLL) and the more modern-but-far-slower web engine, thanks to WASM. Luau is safe and fast.

The exact same bytecode runs in both styles of add-in. You get the performance of native with no extra work. Many approaches for writing custom functions are unacceptably slow, particularly when their inputs come from fast moving realtime data feeds.

I know Luau is perhaps a left-field choice. However, coupled with a coding assistant and bringing that lightweight conversational dev experience right into Excel (xllify Assistant is an add-in itself) this is perhaps less of a barrier.

Yes, this part is an LLM wrapper. LLMs can reliably write code that targets xllify, given adequate guardrails.

The engine is written in C++ and uses Luau (Roblox's Lua fork) as the sandboxed scripting runtime. Luau might seem an odd choice, but it's genuinely well-suited here, it's fast, sandboxed, and compiles easily for both native and WASM targets. And you don't really need to know Luau to use xllify: the AI writes it for you. We'll see if Luau is a barrier.

Anyway, the same Luau code runs two ways:

.. As a native XLL add-in on Windows, talking directly to Excel's C API. This gives you in-process, multi-threaded function execution with low overhead. If you've ever wrestled with the C API directly, you know why this matters. It's extremely fast but utterly brutal to work with. xllify tames.

.. As a web-based add-in via WebAssembly, using the Office JS platform. Same code, same functions, but now deployable on Mac and Web with all the discoverability and security benefits that come with the modern add-in model.

Luau coroutines are supported, so functions can do non-blocking work (API calls, streaming data) without locking up the calculation chain. The runtime ships with a comprehensive standard library so you're not starting from nothing.

DuckDB is available to do heavy lifting on data analytics problems on data that might be too big to fit into Excel.

LLMs can hallucinate, so xllify doesn't rely purely on generation. The AI will select from a library of verified, tested functions where they exist rather than generating from scratch every time. Industry-specific function packs (finance, healthcare) are on the way.

There are two ways to build:

.. xllify Assistant runs inside Excel itself. You have a conversation with the assistant, build and test functions live, and when you're done you can export a .zip with a ready-to-deploy add-in setup including the package.json Office JS packages.

.. xllify agent tooling for local development if you prefer working in your own editor/terminal.

The add-ins are fully self-contained. No runtime dependencies, no phoning home, no ongoing model costs. Once built, it's yours. Works in air-gapped environments.

Try the online demo and sign up for early access: https://xllify.com - I would love your thoughts on this niche little thing.

Comments URL: https://news.ycombinator.com/item?id=46944111

Points: 2

# Comments: 2

Comments URL: https://news.ycombinator.com/item?id=46823876

Points: 2

# Comments: 0

This exists because I wanted to learn about Zig's C interop and comptime to see if it was possible to make the Excel C SDK slightly nicer to work with.

It's just a demo, I wouldn't use it for anything important, but I learnt a lot doing it. There is also a demo repo: https://github.com/AlexJReid/zigxll-standalone

Comments URL: https://news.ycombinator.com/item?id=45978400

Points: 5

# Comments: 3

Comments URL: https://news.ycombinator.com/item?id=45966406

Points: 1

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=32105936

Points: 1

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=8533759

Points: 1

# Comments: 0