Overview

Streamlit Runtime Environments: Warehouse vs Container

What you will learn

What you will build

Prerequisites

In Snowsight, use Projects » Workspaces for the SQL in this guide.

Step-by-Step guide

Step 1: Create the project objects (SQL)

Tip: Keep everything in a single database/schema so the Streamlit app and the agent share the same context.

-- Use Workspaces (Projects » Workspaces) to run this SQL.
USE ROLE ACCOUNTADMIN;

-- Pick names you’ll reuse throughout this guide.
CREATE OR REPLACE DATABASE AGENTS_STREAMLIT_DEMO;
CREATE OR REPLACE SCHEMA AGENTS_STREAMLIT_DEMO.APP;

USE DATABASE AGENTS_STREAMLIT_DEMO;
USE SCHEMA APP;

-- A small warehouse for queries executed by the app (container runtime uses a warehouse only for SQL queries).
CREATE OR REPLACE WAREHOUSE STREAMLIT_QUERY_WH
  WAREHOUSE_SIZE = 'XSMALL'
  AUTO_SUSPEND = 60
  AUTO_RESUME = TRUE
  INITIALLY_SUSPENDED = TRUE;

-- A compute pool to host the Streamlit container runtime.
-- Increase MAX_NODES if you expect to run multiple container-runtime apps concurrently.
CREATE COMPUTE POOL IF NOT EXISTS STREAMLIT_COMPUTE_POOL
  MIN_NODES = 1
  MAX_NODES = 1
  INSTANCE_FAMILY = CPU_X64_XS;

-- Stage for app files (used by CREATE STREAMLIT ... FROM '@stage/path').
CREATE OR REPLACE STAGE APP_STAGE;

Step 2: Install the Snowflake Documentation CKE (Marketplace)

A shared database SNOWFLAKE_DOCUMENTATION is created A shared search service CKE_SNOWFLAKE_DOCS_SERVICE is created under schema SHARED

USE ROLE ACCOUNTADMIN;

-- The listing creates a database; schemas/services depend on the listing.
SHOW CORTEX SEARCH SERVICES IN DATABASE SNOWFLAKE_DOCUMENTATION;

Step 3: Create the Cortex Agent (SQL)

USE ROLE ACCOUNTADMIN;

USE DATABASE AGENTS_STREAMLIT_DEMO;
USE SCHEMA APP;

CREATE OR REPLACE AGENT SNOWFLAKE_DOCUMENTATION
  COMMENT = 'Answers questions using the Snowflake Documentation CKE'
  PROFILE = '{"display_name":"Snowflake Documentation","avatar":"snowflake.png","color":"blue"}'
  FROM SPECIFICATION
$$
instructions:
  system: |
    You are a Snowflake documentation assistant.
    Use ONLY the Snowflake Documentation search tool to answer the user.
    If the documentation does not contain the answer, say you do not know.
    When you use retrieved content, include citations/attribution if available.

tools:
  - tool_spec:
      type: "cortex_search"
      name: "snowflake_docs_search"
      description: "Search Snowflake Documentation (CKE) to ground answers."

tool_resources:
  snowflake_docs_search:
    -- Fully-qualified Cortex Search Service name from the CKE database:
    -- Example format: SNOWFLAKE_DOCUMENTATION_CKE.SHARED.<SERVICE_NAME>
    name: "SNOWFLAKE_DOCUMENTATION.SHARED.CKE_SNOWFLAKE_DOCS_SERVICE"
    max_results: "5"
$$;

SHOW AGENTS IN SCHEMA AGENTS_STREAMLIT_DEMO.APP;

This guide uses ACCOUNTADMIN and does not add custom roles or grants.

Step 4: Download the apps files

streamlit_app.py

This is the streamlit app that calls:

• Threads API: POST /api/v2/cortex/threads
• Agent Run API: POST /api/v2/databases/<db>/schemas/<schema>/agents/SNOWFLAKE_DOCUMENTATION:run

The app reads the container session token from /snowflake/session/token.

Step 5: Deploy the Streamlit app (Container Runtime)

5A) Upload files to the stage (Snowsight UI)

5B) Create External Access Integration and Streamlit object (SQL)

USE ROLE ACCOUNTADMIN;

USE DATABASE AGENTS_STREAMLIT_DEMO;
USE SCHEMA APP;

-- Setup External Access Integration (EAI) to enable custom installation of packages in the streamlit app
-- 1) Allow egress to the PyPI domains Streamlit dependency installs need
CREATE OR REPLACE NETWORK RULE pypi_network_rule
  MODE = EGRESS
  TYPE = HOST_PORT
  VALUE_LIST = ('pypi.org', 'pypi.python.org', 'pythonhosted.org', 'files.pythonhosted.org');

-- 2) Create an External Access Integration using that rule
CREATE OR REPLACE EXTERNAL ACCESS INTEGRATION pypi_access_integration
  ALLOWED_NETWORK_RULES = (pypi_network_rule)
  ENABLED = TRUE;

-- Create Streamlit app using the files and EAI
CREATE OR REPLACE STREAMLIT AGENTS_DOCS_CHAT_APP
  FROM '@APP_STAGE'
  MAIN_FILE = 'streamlit_app.py'
  RUNTIME_NAME = 'SYSTEM$ST_CONTAINER_RUNTIME_PY3_11'
  COMPUTE_POOL = STREAMLIT_COMPUTE_POOL
  EXTERNAL_ACCESS_INTEGRATIONS = (pypi_access_integration)
  QUERY_WAREHOUSE = STREAMLIT_QUERY_WH;

The container runtime requires both RUNTIME_NAME = 'SYSTEM$ST_CONTAINER_RUNTIME_PY3_11' and a COMPUTE_POOL.

Step 6: Try it out

Troubleshooting

Cleanup

USE ROLE ACCOUNTADMIN;

DROP STREAMLIT IF EXISTS AGENTS_STREAMLIT_DEMO.APP.AGENTS_DOCS_CHAT_APP;

DROP AGENT IF EXISTS AGENTS_STREAMLIT_DEMO.APP.SNOWFLAKE_DOCUMENTATION;

DROP STAGE IF EXISTS AGENTS_STREAMLIT_DEMO.APP.APP_STAGE;

DROP COMPUTE POOL IF EXISTS STREAMLIT_COMPUTE_POOL;

DROP WAREHOUSE IF EXISTS STREAMLIT_QUERY_WH;

DROP SCHEMA IF EXISTS AGENTS_STREAMLIT_DEMO.APP;
DROP DATABASE IF EXISTS AGENTS_STREAMLIT_DEMO;

-- Optional: also drop the Marketplace-installed database if you no longer need it.
-- DROP DATABASE IF EXISTS SNOWFLAKE_DOCUMENTATION_CKE;

Conclusion and Resources

What You Learned

Related Resources: