designing-real-world-ai-agents-workshop: Skill for Claude Code

Streamlit Snowflake connection

Connect your Streamlit app to Snowflake the right way.

Use st.connection

Always use st.connection("snowflake") instead of raw connectors.

import streamlit as st

conn = st.connection("snowflake")

# Query data
df = conn.query("SELECT * FROM my_table LIMIT 100")
st.dataframe(df)

Why st.connection:

• Automatic connection pooling
• Built-in caching
• Handles reconnection
• Works with st.secrets

Caller's rights connection (Streamlit 1.53+)

For apps running in Snowflake, use caller's rights to run queries with the viewer's permissions instead of the app owner's:

conn = st.connection("snowflake", type="snowflake-callers-rights")

This is useful when:

• Different users should see different data based on their Snowflake roles
• You want row-level security to apply based on the viewer
• You don't want the app to have elevated permissions

Cached queries

Use the built-in ttl parameter to cache query results:

from datetime import timedelta

conn = st.connection("snowflake")

# Cache for 10 minutes
df = conn.query("SELECT * FROM metrics", ttl=timedelta(minutes=10))

# Cache for 1 hour
df = conn.query("SELECT * FROM reference_data", ttl=3600)

Configure with st.secrets

Store credentials in .streamlit/secrets.toml (never commit this file).

CRITICAL: Derive the account and host values from the user's Snowflake CLI connection config. Run snow connection list and use the exact values. A wrong account will redirect to the wrong login page.

# .streamlit/secrets.toml
[connections.snowflake]
account = "ORGNAME-ACCTNAME"            # from `snow connection list`
host = "myaccount.snowflakecomputing.com"  # from `snow connection list` (include if present)
user = "your_user"
authenticator = "externalbrowser"
warehouse = "your_warehouse"
database = "your_database"
schema = "your_schema"

Add to .gitignore:

.streamlit/secrets.toml

Parameterized queries