Open Source Python Stream Processing Tools

Bytewax is a Python framework that simplifies event and stream processing. Because Bytewax couples the stream and event processing capabilities of Flink, Spark, and Kafka Streams with the friendly and familiar interface of Python, you can re-use the Python libraries you already know and love. Connect data sources, run stateful transformations, and write to various downstream systems with built-in connectors or existing Python libraries. Bytewax is a Python framework and Rust distributed processing engine that uses a dataflow computational model to provide parallelizable stream processing and event processing capabilities similar to Flink, Spark, and Kafka Streams. You can use Bytewax for a variety of workloads from moving data à la Kafka Connect style all the way to advanced online machine learning workloads. Bytewax is not limited to streaming applications but excels anywhere that data can be distributed at the input and output.

Apache Spark™ is a unified analytics engine for large-scale data processing. It provides high-level APIs in Scala, Java, Python, and R, and an optimized engine that supports general computation graphs for data analysis. It also supports a rich set of higher-level tools including Spark SQL for SQL and DataFrames, MLlib for machine learning, GraphX for graph processing, and Structured Streaming for stream processing. The SageMaker Spark Container is a Docker image used to run batch data processing workloads on Amazon SageMaker using the Apache Spark framework. The container images in this repository are used to build the pre-built container images that are used when running Spark jobs on Amazon SageMaker using the SageMaker Python SDK. The pre-built images are available in the Amazon Elastic Container Registry (Amazon ECR), and this repository serves as a reference for those wishing to build their own customized Spark containers for use in Amazon SageMaker.

text-dedup is a Python library that enables efficient deduplication of large text corpora by using MinHash and other probabilistic techniques to detect near-duplicate content. This is especially useful for NLP tasks where duplicated training data can skew model performance. text-dedup scales to billions of documents and offers tools for chunking, hashing, and comparing text efficiently with low memory usage. It supports Jaccard similarity thresholding, parallel execution, and flexible deduplication strategies, making it ideal for cleaning web-scraped data, language model training datasets, or document archives.

Fondant is a modular, pipeline-based framework designed to simplify the preparation of large-scale datasets for training machine learning models, especially foundation models. It offers an end-to-end system for ingesting raw data, applying transformations, filtering, and formatting outputs—all while remaining scalable and traceable. Fondant is designed with reproducibility in mind and supports containerized steps using Docker, making it easy to share and reuse data processing components. It’s built for use in research and production, empowering data scientists to streamline dataset curation and preprocessing workflows efficiently.

Lithops is an open-source serverless computing framework that enables transparent execution of Python functions across multiple cloud providers and on-prem infrastructure. It abstracts cloud providers like IBM Cloud, AWS, Azure, and Google Cloud into a unified interface and turns your Python functions into scalable, event-driven workloads. Lithops is ideal for data processing, ML inference, and embarrassingly parallel workloads, giving you the power of FaaS (Function-as-a-Service) without vendor lock-in. It also supports hybrid cloud setups, object storage access, and simple integration with Jupyter notebooks.

Padasip (Python Adaptive Signal Processing) is a Python library tailored for adaptive filtering and online learning applications, particularly in signal processing and time series forecasting. It includes a variety of adaptive filter algorithms such as LMS, RLS, and their variants, offering real-time adaptation to changing environments. The library is lightweight, well-documented, and ideal for research, prototyping, or teaching purposes. Padasip supports both supervised and unsupervised filtering modes and is built to be modular and extensible, making it easy to integrate into larger machine learning pipelines or control systems.

Pathway is an open-source framework designed for building real-time data applications using reactive and declarative paradigms. It enables seamless integration of live data streams and structured data into analytical pipelines with minimal latency. Pathway is especially well-suited for scenarios like financial analytics, IoT, fraud detection, and logistics, where high-velocity and continuously changing data is the norm. Unlike traditional batch processing frameworks, Pathway continuously updates the results of your data logic as new events arrive, functioning more like a database that reacts in real-time. It supports Python, integrates with modern data tools, and offers a deterministic dataflow model to ensure reproducibility and correctness.

Sed.py is a python module to provide a easy way to do text stream processing. Just like the name of module, it likes to do the work that sed can do. But not in sed's way, it's in Python's way. To use this module, the knowledge of regexp is necessary.

Pyper is a Python-native orchestration and scheduling framework designed for modern data workflows, machine learning pipelines, and any task that benefits from a lightweight DAG-based execution engine. Unlike heavier platforms like Airflow, Pyper aims to remain lean, modular, and developer-friendly, embracing Pythonic conventions and minimizing boilerplate. It focuses on local development ergonomics and seamless transition to production environments, making it ideal for small teams and individuals needing a programmable and flexible orchestration solution without the overhead of enterprise systems.

StreamMine is a distributed event processing (streaming) infrastructure. You can create low-latency, fault-tolerant stream processing functionality with any stream-oriented operators that can be implemented in Python.

Wally is a fast-stream-processing framework. Wally makes it easy to react to data in real-time. By eliminating infrastructure complexity, going from prototype to production has never been simpler. When we set out to build Wally, we had several high-level goals in mind. Create a dependable and resilient distributed computing framework. Take care of the complexities of distributed computing "plumbing," allowing developers to focus on their business logic. Provide high-performance & low-latency data processing. Be portable and deploy easily (i.e., run on-prem or any cloud). Manage in-memory state for the application. Allow applications to scale as needed, even when they are live and up-and-running. The primary API for Wally is written in Pony. Wally applications are written using this Pony API.