Streamlit Supabase Connection

Use Supabase Storage, Database, and Auth from Streamlit with Streamlit-aware caching and browser-session-safe authentication.

Open the interactive demo · Read the guides · View the changelog

Why use it?

• Use st.connection() and Streamlit secrets for Supabase configuration.
• Cache Storage and Database reads with familiar ttl values such as "10m".
• Keep Supabase Auth tokens isolated to the current Streamlit browser session.
• Upload st.file_uploader() results without first writing them to the app server.
• Normalize Storage paths and infer common MIME types automatically.

Requirements

Quickstart

Install the package:

pip install st-supabase-connection

Add your project URL and publishable key to .streamlit/secrets.toml:

[connections.supabase_connection]
SUPABASE_URL = "https://your-project.supabase.co"
SUPABASE_PUBLISHABLE_KEY = "sb_publishable_..."

Create the connection and make a cached read:

import streamlit as st

from st_supabase_connection import SupabaseConnection

connection = st.connection(
    "supabase_connection",
    type=SupabaseConnection,
)

buckets = connection.list_buckets(ttl="10m")
st.write(buckets)

The legacy SUPABASE_KEY setting is still supported. New user-facing apps should use a Supabase publishable key. Never use a secret or service_role key for user-scoped operations because those keys bypass Row Level Security.

You can also supply SUPABASE_URL and SUPABASE_PUBLISHABLE_KEY as environment variables, or pass url= and key= to st.connection().

Choose the right client

Streamlit connection objects are cached resources and may be shared between browser sessions. Choose the client based on whether the operation depends on a signed-in user.

Create the session-scoped client inside the Streamlit script:

supabase = connection.session_client()

Reuse that client for Auth and user-specific operations. Do not put it in a module-level variable, st.cache_resource, or another global cache.

Caching behavior

There are two different kinds of ttl:

In most apps, omit ttl from st.connection() and set a finite ttl on reads that can change:

from st_supabase_connection import execute_query

countries = execute_query(
    connection.table("countries").select("id, name").order("name"),
    ttl="10m",
)
st.dataframe(countries.data)

Do not cache Auth calls. For inserts, updates, upserts, and deletes, call Supabase's .execute() directly or pass ttl=0 to execute_query().

See the caching guide for cache scope, user-specific queries, and invalidation considerations.

Common workflows

Upload from st.file_uploader

The uploaded file is handled in memory; it does not need to be copied to the app server first.

uploaded_file = st.file_uploader("Choose a file")

if uploaded_file is not None and st.button("Upload"):
    connection.upload(
        bucket_id="documents",
        source="local",
        file=uploaded_file,
        destination_path=f"uploads/{uploaded_file.name}",
        overwrite="false",
    )
    st.success("Upload complete")

Query the database

Use the shared connection for anonymous reads:

from st_supabase_connection import execute_query

response = execute_query(
    connection.table("countries").select("id, name").limit(20),
    ttl="5m",
)
st.dataframe(response.data)

Use the session client for rows protected by a signed-in user's RLS policies:

supabase = connection.session_client()
response = supabase.table("private_profiles").select("*").execute()
st.dataframe(response.data)

Sign in an existing user

supabase = connection.session_client()

email = st.text_input("Email")
password = st.text_input("Password", type="password")

if st.button("Sign in"):
    supabase.auth.sign_in_with_password(
        {"email": email, "password": password}
    )
    st.success("Signed in")

After sign-in, use the same supabase client for all operations that must carry the user's JWT.

Guides and recipes

• Storage recipes: buckets, uploads, downloads, object management, and signed URLs
• Database recipes: cached reads, joins, filters, writes, RLS, and Data API access
• Authentication and security: keys, sign-in, session isolation, sign-out, and user-scoped queries
• Caching: connection lifetime, result lifetime, and cache-safe writes
• Upgrade notes: behavioral changes and migration guidance

Supported functionality

The connection includes Streamlit-friendly wrappers for commonly used Storage operations:

• Bucket management: list_buckets(), get_bucket(), create_bucket(), update_bucket(), empty_bucket(), and delete_bucket()
• Objects: upload(), download(), list_objects(), move(), and remove()
• URLs: get_public_url(), create_signed_urls(), create_signed_upload_url(), and upload_to_signed_url()
• Database: table() and the cached execute_query() helper
• Auth and other user-scoped Supabase APIs: session_client()

Because session_client() returns the complete Supabase Python client, it can also be used for Functions, Realtime, and other SDK features that must carry the current user's session.

Upgrading to 2.2.0

Version 2.2.0 requires Python 3.10+, Streamlit 1.62.0+, and Supabase Python 2.22.0+.

Apps using Auth should replace process-shared access:

# Deprecated
connection.auth.sign_in_with_password(credentials)

with a session-scoped client:

supabase = connection.session_client()
supabase.auth.sign_in_with_password(credentials)

connection.auth and cached_sign_in_with_password() remain available for compatibility but are deprecated. The latter no longer caches its result and ignores ttl.

See the 2.2.0 changelog for the complete migration notes.

Development

Install the project and demo dependency, then run the library and demo tests:

python -m pip install --editable . streamlit-extras
python -m unittest discover --start-directory tests --verbose

The CI workflow tests Python 3.10–3.14 against the minimum and latest supported Streamlit releases.

Issues and pull requests are welcome in the GitHub repository.

Acknowledgements

This connector builds on Streamlit, Supabase Python, and the work of the Supabase open-source community.

If the project helps you, you can sponsor it on GitHub or buy me a coffee.