Overview
PandasAI is a Python library that brings conversational capabilities to tabular data analysis. By connecting pandas DataFrames to a large language model (LLM), PandasAI allows users to ask natural-language questions about their data, get analysis summaries, request plots, and produce pandas code snippets or direct answers — all through a chat-like interface.
Key features
- Conversational queries: Ask natural-language questions about your DataFrame (e.g., "What is the average revenue by region?") and receive direct answers or generated pandas code.
- Multi-DataFrame support: Query and relate multiple DataFrames in a single conversational session.
- Visualization generation: Request charts (histograms, bar plots, scatter plots, etc.) in plain language and receive generated visualizations.
- LLM integration: Works with various LLM backends (examples in the repo show integration patterns like LiteLLM and OpenAI models). This allows leveraging models for reasoning, summarization, and code generation.
- Docker sandbox: Optional sandboxed environment to run generated code safely, reducing the risk of executing untrusted code on the host machine.
- RAG (Retrieval-Augmented Generation) compatibility: Designed to be used in workflows where external context or retrieval improves answers to data questions.
- Low barrier to entry: Aimed at both non-technical users (natural-language access to data) and technical users (speeding up exploration and prototyping).
Getting started (summary)
- Install: pip install pandasai (and optional LLM adapters like pandasai-litellm).
- Configure an LLM backend and API key in the library config.
- Load your data into a pandas DataFrame and call df.chat("Your question") to interact.
Example:
import pandasai as pai
from pandasai_litellm.litellm import LiteLLM
llm = LiteLLM(model="gpt-4.1-mini", api_key="YOUR_OPENAI_API_KEY")
pai.config.set({"llm": llm})
df = pai.read_csv("data/companies.csv")
response = df.chat("What is the average revenue by region?")
print(response)Use cases
- Business analysts who want to query datasets without writing pandas code.
- Data scientists who want to prototype analyses quickly or generate code snippets for repetitive tasks.
- Teams building data-driven chat interfaces or LLM-powered analytics products.
Governance & license
PandasAI is published under the MIT license. The repository provides documentation, examples, and a Discord community for discussion.
Project status & links
The project provides docs at the official site and examples in the GitHub repository. It includes CI/CD badges, tests coverage, and installation instructions in its README.
