ponia 0.1.2

Query pandas DataFrames with natural language using OpenAI function calling

Ponia

Query pandas DataFrames with natural language using OpenAI function calling.

Installation

pip install -e .

Usage

import pandas as pd
from ponia import ponia

# Set your API key as environment variable
# export OPENAI_API_KEY="your-api-key"

df = pd.DataFrame({
    'product': ['apple', 'banana', 'orange'],
    'monday': [10, 15, 8],
    'tuesday': [12, 18, 9],
    'wednesday': [8, 20, 11]
})

# Query in natural language
answer = ponia(df, "Which column has the largest value?")
print(answer)
# "The column 'wednesday' has the largest value, which is 20..."

answer = ponia(df, "What is the average of monday sales?")
print(answer)
# "The average of column 'monday' is 11.0"

# Get raw JSON result instead of natural language
result = ponia(df, "What is the max value?", raw=True)
print(result)
# {'value': 20, 'column': 'wednesday', 'row': 1}

Features

• Zero data leakage: Data is never sent to OpenAI, only the question
• No eval(): Does not execute AI-generated code, only predefined functions
• Minimal: Minimal dependencies (pandas, openai)
• Extensible: Easy to add new functions

Available Functions

• find_max_value_location - Global max value and location
• find_min_value_location - Global min value and location
• get_column_max/min/sum/mean/median/std - Column statistics
• count_rows/columns/unique - Counts
• filter_by_value/comparison - Row filtering
• filter_and_aggregate - Filter then aggregate (e.g., "average salary for engineers")
• group_aggregate - Group by and aggregate
• get_correlation - Correlation between columns
• get_top_n_rows/bottom_n_rows - Top/bottom N rows
• And more...

Tests

pip install -e ".[dev]"
pytest