Streamlit integration¶

Streamlit can render interactive tables from objects that implement the Python DataFrame Interchange Protocol (__dataframe__). As of 0.21.0, pydantable implements __dataframe__ on DataFrame (and DataFrameModel via delegation), so you can pass a typed frame directly to st.dataframe.

Install¶

You need Streamlit plus the pydantable Arrow extra:

pip install streamlit
pip install 'pydantable[arrow]'

Usage¶

import streamlit as st
from pydantable import DataFrameModel


class User(DataFrameModel):
    id: int
    name: str
    age: int | None


df = User({"id": [1, 2], "name": ["a", "b"], "age": [10, None]})

st.write(df)
st.dataframe(df)

# `st.data_editor` currently expects a concrete frame type (Arrow/pandas/Polars),
# so use an explicit conversion:
st.data_editor(df.to_arrow())

Fallbacks (when interchange is unavailable)¶

If you don’t have pyarrow installed, __dataframe__ will fail. Use one of:
st.dataframe(df.to_arrow()) (requires pydantable[arrow])
st.dataframe(df.to_polars()) (requires pydantable[polars])
For editing, prefer st.data_editor(df.to_arrow()) (requires pydantable[arrow]) or st.data_editor(df.to_polars()) (requires pydantable[polars]).
st.write(df) will still show either the HTML preview (_repr_html_) or the plain repr depending on the Streamlit rendering path, but it won’t be an interactive table unless Streamlit can treat it as a dataframe-like object.

Costs and limitations¶

pydantable’s __dataframe__ path materializes the current lazy plan (same cost class as EXECUTION → to_arrow()), then delegates to PyArrow’s protocol implementation. It is not a zero-copy export of internal engine buffers.
For large frames, prefer applying a lazy head(...) / slice(...) before displaying.