From Streamlit to Tauri: Bringing Water Engineering Tools to the Desktop

Background

Computational tools in the water conservancy industry come with one peculiar requirement: many users work in environments without internet access (think field surveys at remote sites). The Streamlit version was quick to build, but it can only be used online.

So I started a Streamlit → Tauri migration plan, with the goal of progressively turning 12 hydro-* water computation tools into desktop applications.

Migration Strategy

It's Not a Rewrite — It's a Separation

The core idea is to separate the compute engine from the UI:

Three Phases

1. Extract the compute engine: peel the core computational logic out of the Streamlit code into an independent Python module
2. Rewrite the frontend in React: replicate the Streamlit interface with shadcn/ui, keeping functional parity
3. Tauri integration: the Rust backend invokes the Python compute engine through a sidecar

Pitfalls

1. rsync and symlinks

By default, rsync doesn't follow symlinks during deployment, which causes linked files inside node_modules to disappear.

Fix: rsync -aL (the L flag follows symlinks).

2. Native modules

Native modules like better-sqlite3 behave differently inside Tauri's sidecar environment.

Fix: keep the compute layer pure Python and avoid native modules on the frontend.

3. Webpack crash

Some React components cause webpack memory blowups inside Tauri's webview.

Fix: split chunks and lazy-load the heavy components.

4. CSS hash mismatch

When the standalone build and the static files come from different builds, CSS filenames carry different hashes — and styles vanish.

Fix: rm -rf .next before every deploy to guarantee both come from the same build.

Migration Progress

Two tools have been migrated so far:

• hydro-capacity: Pollutant Assimilation Capacity calculator
• hydro-efficiency: Water Efficiency Assessment system

Each tool ships as both web and desktop. The web version runs on the VPS; the desktop version works offline.

Lessons Learned

1. A standalone compute engine is the key — the difficulty and risk of migration sit in the UI, not the math
2. Don't aim for one-shot completion — migrate the most-used tool first, validate the pattern, then scale up
3. Tauri is far lighter than Electron — packaged binaries are tens of MB instead of 100+ MB
4. Keep the web version online — desktop is a complement, not a replacement