PandasAI 3.0 is currently in beta. This documentation reflects the latest features and functionality, which may evolve before the final release.

​
Installation

PandasAI requires Python 3.8+ <3.12. We recommend using Poetry for dependency management:

# Using poetry (recommended)
poetry add "pandasai>=3.0.0b2"

# Alternative: using pip
pip install "pandasai>=3.0.0b2"

​
Quick setup

In order to use PandasAI, you need a large language model (LLM). You can use any LLM, but for this guide we’ll use OpenAI through the LiteLLM extension.

First, install the required extension:

pip install pandasai-litellm

Then, import PandasAI and configure the LLM:

import pandasai as pai
from pandasai_litellm.litellm import LiteLLM

# Initialize LiteLLM with your OpenAI model
llm = LiteLLM(model="gpt-4.1-mini", api_key="YOUR_OPENAI_API_KEY")

# Configure PandasAI to use this LLM
pai.config.set({
    "llm": llm
})

​
Chat with your data

import pandasai as pai
from pandasai_litellm.litellm import LiteLLM

# Initialize LiteLLM with your OpenAI model
llm = LiteLLM(model="gpt-4.1-mini", api_key="YOUR_OPENAI_API_KEY")

# Configure PandasAI to use this LLM
pai.config.set({
    "llm": llm
})

# Load your data
df = pai.read_csv("data/companies.csv")

response = df.chat("What is the average revenue by region?")
print(response)

When you ask a question, PandasAI will use the LLM to generate the answer and output a response. Depending on your question, it can return different kind of responses:

  • string
  • dataframe
  • chart
  • number

Find it more about output data formats here.

​
Creating your first data layer

​
1. Define a data source

Start by creating a data schema that describes your dataset:

import pandasai as pai

# Load your data
df = pai.read_csv("data/companies.csv")

# Create the data layer
companies = pai.create(
  path="my-org/companies",
  df=df,
  description="Customer companies dataset"
)

This dataset will be saved in the datasets/my-org/companies folder of your project.

​
2. Define the structure of your dataset

By default, the column will be inferred from the data. For more control, though, you can define explicit column schemas:

# Define a companies dataset with explicit schema
companies = pai.create(
  path="my-org/companies",
  df=df,
  description="Customer companies dataset",
  columns=[
    {
      "name": "company_name",
      "type": "string",
      "description": "The name of the company"
    },
    {
      "name": "revenue",
      "type": "float",
      "description": "The revenue of the company"
    },
    {
      "name": "region",
      "type": "string",
      "description": "The region of the company"
    }
  ]
)

​
3. Load and query data

Once defined, you can easily load and query your datasets:

# Load existing datasets
stocks = pai.load("organization/coca_cola_stock")
companies = pai.load("organization/companies")

# Query using natural language
response = stocks.chat("What is the volatility of the Coca Cola stock?")
response = companies.chat("What is the average revenue by region?")

# Query using multiple datasets
result = pai.chat("Compare the revenue between Coca Cola and Apple", stocks, companies)

​
Next Steps