Fluidize

An Open Framework for AI-Driven Scientific Computing

fluidize-python is a library for building modular, reproducible scientific computing pipelines. It provides a unified interface to a wide range of physical simulation tools, eliminating the need to navigate the inconsistent, incomplete instructions that often vary from tool to tool.

This library marks our first step toward AI-orchestrated scientific computing. By standardizing tools and practices within our framework, AI agents can automatically build, configure, and execute computational pipelines across domains and simulation platforms. Our goal is to improve today’s simulation tools so AI can assist researchers and scientists in accelerating the pace of innovation and scientific discovery.

Quick Start

Installation

Prerequesites:

• Python 3.9+

• Docker Desktop (for local execution). Download and install Docker Desktop from https://docs.docker.com/desktop/.

  After installation, verify with:

  docker --version
  

From PyPI

pip install fluidize

From Source

git clone https://github.com/Fluidize-Inc/fluidize-python.git
cd fluidize-python
make install

Run Examples

Example projects are located in this folder: example/

The Problem

Students and researchers face significant barriers when working with different simulation tools:

• Setup overhead – Installing and configuring someone else’s research code can take an enormous amount of time.
• Diverse architectures – Scientific software is built using a wide range of tools and architectures, each with its own complexities and quirks.
• Time drain – Good software engineering practices are important, but in practice they often slow down the process of getting immediate results.
• Reproducibility issues – Sharing and reproducing experiments is frequently cumbersome and error-prone.
• Scaling friction – Moving from a local prototype to a cloud environment or dedicated compute cluster can be slow and difficult.

The Solution

Fluidize provides a standardized wrapper that turns complex scientific software into modular components. This makes it possible to:

• Expose a single API endpoint for any scientific computing software—any language, any tool, any complexity.
• Easily connect tools that were never designed to work together.
• Adopt consistent I/O patterns across all simulations.

All of this works with minimal or no changes to the existing codebase, allowing our framework to scale effortlessly to any repository.

Architecture

Nodes

The foundational building blocks of Fluidize. Each node encapsulates a computational unit with:

Key Features:

• Predictable input/output paths
• Modular and extensible design
• No source code modification required
• Automated node generation support (Public launch soon)

Projects

The project currently hosts a simple layer for composing and managing multiple nodes:

Docker engine is used for local execution. With API calls, we use the Kubernetes engine with Argo Workflow Manager.

Documentation

Comprehensive documentation is available at https://Fluidize-Inc.github.io/fluidize-python/

• Getting Started Guide
• Node Creation Tutorial
• Project Orchestration
• API Reference

Contributing

We would love contributions and collaborations! Please see our Contributing Guide for details.

Also - we would love to help streamline your research pipeline! Please reach out at henry@fluidize.ai or henrybae@g.harvard.edu.

Roadmap

This is just the beginning of what we think is a really exciting new era for how we learn science and do research. We will be releasing the following tools built from this framework:

• Fluidize Playground: Automatically explore and build simulation pipelines with natural language.
• Auto-Fluidize: Automatically convert obscure scientific software to run anywhere

License

This project is licensed under the MIT License - see the LICENSE file for details.