bodo 2026.4

High-Performance Python Compute Engine for Data and AI

Docs  ·  Slack  ·  Benchmarks

Bodo DataFrames: Drop-in Pandas Replacement for Acceleration and Scaling of Data and AI

Bodo DataFrames is a high performance DataFrame library for large scale Python data processing, AI/ML use cases. It functions as a drop-in replacement for Pandas while providing additional Pandas-compatible APIs for simplifying and scaling AI workloads, a just-in-time (JIT) compiler for accelerating custom transformations, as well as an integrated SQL engine for extra flexibility.

Under the hood, Bodo DataFrames relies on MPI-based high-performance computing (HPC) technology, often making it orders of magnitude faster than tools like Spark or Dask. Refer to our NYC Taxi benchmark for an example where Bodo is 2-240x faster than other systems:

Unlike traditional distributed computing frameworks, Bodo DataFrames:

• Automatically scales and accelerates Pandas workloads with a single line of code change.
• Eliminates runtime overheads common in driver-executor models by leveraging Message Passing Interface (MPI) technology for true parallel execution.

Goals

Bodo DataFrames makes Python run much (much!) faster than it normally does!

1. Exceptional Performance: Deliver HPC-grade performance and scalability for Python data workloads as if the code was written in C++/MPI, whether running on a laptop or across large cloud clusters.

2. Easy to Use: Easily integrate into Python workflows— it's as simple as changing import pandas as pd to import bodo.pandas as pd.

3. Interoperable: Compatible with regular Python ecosystem, and can selectively speed up only the sections of the workload that are Bodo supported.

4. Integration with Modern Data Infrastructure: Provide robust support for industry-leading data platforms like Apache Iceberg and Snowflake, enabling smooth interoperability with existing ecosystems.

Key Features

• Drop-in Pandas replacement, (just change the import!) with a seamless fallback to vanilla Pandas to avoid breaking existing workloads.
• Intuitive APIs for simplifying and scaling AI workloads.
• Advanced query optimization, C++ runtime, and parallel execution using MPI to achieve the best possible performance while leveraging all available cores.
• Streaming execution to process larger-than-memory datasets.
• Just in time (JIT) compilation with native support for Pandas, Numpy and Scikit-learn for accelerating custom transformations or performance-critical functions.
• High performance SQL engine that is natively integrated into Python.
• Advanced scalable I/O support for Iceberg, Snowflake, Parquet, CSV, and JSON with automatic filter pushdown and column pruning for optimized data access.

See Bodo DataFrames documentation to learn more: https://docs.bodo.ai/

Installation

Note: Bodo DataFrames requires Python 3.10+.

Bodo DataFrames can be installed using Pip or Conda:

pip install -U bodo

or

conda create -n Bodo python=3.14 -c conda-forge
conda activate Bodo
conda install bodo -c conda-forge

Bodo DataFrames works with Linux x86, both Mac x86 and Mac ARM, and Windows right now. We will have Linux ARM support (and more) coming soon!

Bodo DataFrames Example

Here is an example Pandas code that reads and processes a sample Parquet dataset. Note that we replaced the typical import:

import pandas as pd

with:

import bodo.pandas as pd

which accelerates the following code segment by about 20-30x on a laptop.

import bodo.pandas as pd
import numpy as np
import time

NUM_GROUPS = 30
NUM_ROWS = 20_000_000

df = pd.DataFrame({
    "A": np.arange(NUM_ROWS) % NUM_GROUPS,
    "B": np.arange(NUM_ROWS)
})
df.to_parquet("my_data.pq")

def computation():
    t1 = time.time()
    df = pd.read_parquet("my_data.pq")
    df["C"] = df.apply(lambda r: 0 if r.A == 0 else (r.B // r.A), axis=1)
    df.to_parquet("out.pq")
    print("Execution time:", time.time() - t1)

computation()

How to Contribute

Please read our latest project contribution guide.

Getting involved

You can join our community and collaborate with other contributors by joining our Slack channel – we’re excited to hear your ideas and help you get started!