I’ve been experimenting with MicroHs, a minimal Haskell implementation, and ended up building a small project on top of it: xeus-haskell, a new Jupyter kernel.

A few things that might interest folks here:

- MicroHs has almost no dependencies, so connecting it to the Jupyter messaging protocol via xeus was surprisingly straightforward. - That minimalism also means MicroHs compiles cleanly to WebAssembly. - As a result, xeus-haskell runs on JupyterLite in the browser — no GHC, no toolchains, no local setup.

The goal is to make Haskell easier to use for scientific and technical computing. Lazy evaluation can be very effective for certain workloads (graph algorithms, recursive structures, etc.), and being able to demonstrate that interactively in a notebook feels like a natural fit.

Demo: https://tani.github.io/xeus-haskell

This works really well, great work! How many MB is the whole bundle?