Streamlit for internal tools nobody else will ever see

A recurring bad trade in internal tooling: spinning up a React frontend, a REST API, and real auth for a tool that three people on one team will use to eyeball some data or trigger a script. The tool that actually gets used is often the one that took an afternoon, not the one that took a sprint.

What Streamlit actually gets you

Streamlit turns a Python script into a web UI with widgets, tables, and charts, using plain Python control flow — no separate frontend build, no API layer to maintain in sync with a UI, no deployment pipeline beyond running the script.

import streamlit as st

st.title("Eval Results Browser")
dataset = st.selectbox("Dataset", ["v1", "v2", "v3"])
results = load_results(dataset)
st.dataframe(results)
st.line_chart(results["score"])

That's a functioning internal tool — pick a dataset, see a table, see a trend line — in about as much code as it takes to describe what it does, with no frontend framework decision, no component library, and no API contract to keep in sync with a UI built separately.

What you're trading away, on purpose

Streamlit reruns the entire script top to bottom on every interaction, which is a real constraint for anything with expensive computation or complex, deeply nested interactive state — it's not the right tool for a polished multi-step wizard or a UI with intricate client-side state management. Layout customization is comparatively limited next to a real frontend framework, and it's not meant to serve thousands of concurrent external users. None of these are bugs; they're the trade Streamlit makes deliberately, in exchange for the entire frontend layer disappearing as a concern.

Where it stops being the right call

The moment this stops being the right tool: the internal tool needs to become customer-facing, needs genuinely complex interactive state, or needs to serve real concurrent load rather than three people occasionally checking a dashboard. Building it in Streamlit first is still frequently the right call even then — it validates whether the tool is worth the investment of a real frontend before spending a sprint building one nobody ends up using, and "nobody ended up using it" is a much cheaper thing to learn from an afternoon's Streamlit script than from a properly built app three weeks in.