streamlit-ai-assist

streamlit-ai-assist uses the power of Snowflake Arctic, an open_source LLM, to personalize dashboards and allow all audiences to easily explore new and pre-built data visualizations. It's a built-in data analyst for your Streamlit dashboard.

If you add this component to your dashboard, dashboard viewers will be able to chat with an LLM assistant to get answers to their questions. If there are relevant graphs already in the dashboard, then the assistant will display those. If the question requires a new graph, the LLM is able to write the code to create that. These new graphs are highlighted in yellow, and if GitHub connection is configured in the app, then the user will be allowed to easily create a PR that integrates the code into the dashboard once reviewed and approved by a team member.

If a user has a particular question, the AI assistant can execute SQL SELECT statements that let it access the raw numbers. It can summarize graphs for you and return answers to a quick question you have.

The bottom view of the streamlit component is a preview of the most relevant graphs availabe in the dashboard. If you haven't formulated a question yet, a simple keyword search or "show me graphs about X" could be helpful!

Installation instructions

pip install streamlit-custom-component

Usage instructions

import streamlit as st

from streamlit_ai_assist import streamlit_ai_assit

streamlit_ai_assist(
            graphing_file_path="graphing.py",
            graphing_import_path="graphing",
            database_name="snowflake",
            general_description="This is a database for a company that does X",
            mode="chat",
            key="foo1"
)

Requirements

The following credentials are needed:

Environment Variables (can be specified in secrets.toml or with os)

REPLICATE_API_TOKEN

Database Connection Variables (must be specified in secrets.toml in a section named connections.<database_name>)

e.g.

[connections.snowflake]
type = "snowflake"
user = 
password =
account = 
role = 
warehouse = 
database = 
schema = 
client_session_keep_alive = true

Currently, only Snowflake database connections are tested and fully supported.

Optional Environment Variables

GITHUB_PERSONAL_ACCESS_TOKEN
REPO_NAME
REPO_OWNER

If these are specified, GitHub PR automation will be enabled.