vanna 0.0.8

Generate SQL queries from natural language

GitHub	PyPI	Colab	Documentation

Vanna.AI

Vanna is a Python-based AI SQL co-pilot. Our initial users are data-savvy data analysts, data scientists, engineers, and similar people that use Vanna to automate writing complex SQL.

Vanna can:

• Convert natural language to SQL
• Run SQL
• Generate Plotly code
• Run Plotly code
• Get better over time
• Be used in a Jupyter Notebooks, Colab, or other Python environments
• Be used with Snowflake, BigQuery, and other databases
• Be used with Python UIs, such as Streamlit, Dash, and others
• Be used to make Slack bots

Natural Language to SQL

sql = vn.generate_sql(question='Who are the top 10 customers?')

Output:

SELECT customer_name,
       total_sales
FROM   (SELECT c.c_name as customer_name,
               sum(l.l_extendedprice * (1 - l.l_discount)) as total_sales,
               row_number() OVER (ORDER BY sum(l.l_extendedprice * (1 - l.l_discount)) desc) as rank
        FROM   snowflake_sample_data.tpch_sf1.lineitem l join snowflake_sample_data.tpch_sf1.orders o
                ON l.l_orderkey = o.o_orderkey join snowflake_sample_data.tpch_sf1.customer c
                ON o.o_custkey = c.c_custkey
        GROUP BY customer_name)
WHERE  rank <= 10;

Run SQL

This function is provided as a convenience. You can choose to run your SQL however you normally do and use the rest of the downstream functions.

df = vn.get_results(cs, database, sql)

Output:

customer_name	total_sales
Customer#000000001	68127.72
Customer#000000002	65898.69
...

Generate Plotly Code

plotly_code = vn.generate_plotly_code(question=my_question, sql=sql, df=df)

Output:

fig = go.Figure(go.Bar(
    x=df['CUSTOMER_NAME'],
    y=df['TOTAL_SALES'],
    marker={'color': df['TOTAL_SALES'], 'colorscale': 'Viridis'},
    text=df['TOTAL_SALES'],
    textposition='auto',
))

fig.update_layout(
    title="Top 10 Customers by Sales",
    xaxis_title="Customer",
    yaxis_title="Total Sales",
    xaxis_tickangle=-45,
    yaxis_tickprefix="$",
)

Run Plotly Code

fig = vn.get_plotly_figure(plotly_code=plotly_code, df=df)
fig.show()

Output:

Improve Your Training Data

vn.store_sql(
    question=my_question,
    sql=sql,
)

How Vanna Works

flowchart LR
    DB[(Known Correct Question-SQL)]
    Try[Try to Use DDL/Documentation]
    SQL(SQL)
    Check{Is the SQL correct?}
    Generate[fa:fa-circle-question Use Examples to Generate]
    DB --> Find
    Question[fa:fa-circle-question Question] --> Find{fa:fa-magnifying-glass Do we have similar questions?}
    Find -- Yes --> Generate
    Find -- No --> Try
    Generate --> SQL
    Try --> SQL
    SQL --> Check
    Check -- Yes --> DB
    Check -- No --> Analyst[fa:fa-glasses Analyst Writes the SQL]
    Analyst -- Adds --> DB

Getting Started

Install Vanna from PyPI and import it:

%pip install vanna
import vanna as vn

Enter your email to set an API Key

This will send a one-time code to your email address. Copy and paste the code into the prompt.

my_email = '' # Enter your email here
vn.login(email=my_email)

Add Training Data

vn.train(
    question="Which products have the highest sales?",
    sql="...",
)

Generate SQL

sql = vn.generate_sql(question="Who are the top 10 customers?")

Documentation

Full Documentation