whatisgithub

What is fairchem?

facebookresearch/fairchem — explained in plain English

Analysis updated 2026-07-05 · repo last pushed 2026-07-05

2,173PythonAudience · researcherComplexity · 4/5ActiveSetup · hard

In one sentence

Fairchem is Meta's open-source AI library that predicts how molecules and materials behave, replacing slow physics simulations with fast machine learning models for chemistry and materials science research.

Mindmap

mindmap
  root((repo))
    What it does
      Predicts molecule properties
      Simulates atom interactions
      Replaces slow physics sims
    Tech stack
      Python
      ASE toolkit
      Multi-GPU support
    Use cases
      Discover new catalysts
      Study metal-organic frameworks
      Design new molecules
    Audience
      Materials scientists
      Chemists
      Deep tech founders
    Scale
      100000 atoms plus
      Fast molecular dynamics
      Auto parallel computing
Click or tap to explore — scroll the page freely

Code map

Detail Auto

An interactive map of this repo's files and how they connect — its source is parsed live in your browser. Click Visualize to build it.

filefunction / class

What do people build with it?

USE CASE 1

Speed up discovery of new catalysts by simulating reactions with AI instead of slow physics calculations.

USE CASE 2

Study metal-organic frameworks and crystal structures quickly using a single AI model.

USE CASE 3

Design and test new molecules by predicting their energies and forces in a fraction of the time.

USE CASE 4

Run large-scale molecular dynamics simulations with over 100,000 atoms across multiple GPUs.

What is it built with?

PythonASEPyTorchCUDA

How does it compare?

facebookresearch/fairchemgoogle-deepmind/science-skillshughyau/academicforge
Stars2,1732,2022,095
LanguagePythonPythonPython
Last pushed2026-07-052026-07-01
MaintenanceActiveActive
Setup difficultyhardmoderateeasy
Complexity4/52/52/5
Audienceresearcherresearcherresearcher

Figures from each repo's GitHub metadata at analysis time.

How do you get it running?

Difficulty · hard Time to first run · 1h+

Requires high-end GPUs for meaningful simulations and familiarity with the ASE scientific toolkit and Python.

So what is it?

Fairchem is an open-source library from Meta's FAIR Chemistry team that uses machine learning to predict the properties and behavior of molecules and materials. Normally, figuring out how atoms interact, like how a catalyst speeds up a chemical reaction or how a crystal structure settles into shape, requires slow, computationally expensive physics simulations. This project replaces a lot of that heavy lifting with fast AI models, letting researchers simulate chemistry much more quickly. At the core of the project is a model called UMA. Instead of needing a different AI for every type of chemistry problem, you use this one model and simply tell it which "task" you are working on. For example, you set the task to "omol" for molecules and polymers, or "omat" for inorganic materials. It plugs into an existing popular scientific toolkit called ASE, so you set up your atoms in Python, tell the AI to calculate the energies and forces, and then simulate how the system evolves over time or settles into a stable state. The people who would use this are materials scientists, chemists, and researchers working on things like discovering new catalysts, studying metal-organic frameworks, or designing new molecules. For a founder or product manager in deep tech or climate tech, this tool could drastically speed up an R&D pipeline, instead of waiting days for traditional simulations to tell you if a material is viable, the AI can give you answers in a fraction of the time. It also scales up to handle massive simulations across multiple high-end GPUs. One notable thing about the project is its focus on scale and speed. The team highlights that it can handle systems of over 100,000 atoms and run simulations at speeds relevant to real-world molecular dynamics. It also manages all the complex parallel computing under the hood with a single setting, so the researcher does not have to be an expert in distributed systems to use it. However, the models were trained on specific types of physics data, so users need to be careful not to mix and match them with incompatible external datasets without adjusting for the differences.

Copy-paste prompts

Prompt 1
Using fairchem's UMA model with the ASE toolkit, how do I set up a molecule in Python, set the task to omol, and calculate its energies and forces?
Prompt 2
I want to run a molecular dynamics simulation on a system with over 100,000 atoms using fairchem. How do I configure it to use multiple GPUs with a single setting?
Prompt 3
How do I choose the correct task setting in fairchem UMA for simulating inorganic materials versus organic molecules and polymers?
Prompt 4
What are the limitations of using fairchem's pretrained models with my own external datasets, and how do I adjust for differences in the physics data?

Frequently asked questions

What is fairchem?

Fairchem is Meta's open-source AI library that predicts how molecules and materials behave, replacing slow physics simulations with fast machine learning models for chemistry and materials science research.

What language is fairchem written in?

Mainly Python. The stack also includes Python, ASE, PyTorch.

Is fairchem actively maintained?

Active — commit in last 30 days (last push 2026-07-05).

How hard is fairchem to set up?

Setup difficulty is rated hard, with roughly 1h+ to a first successful run.

Who is fairchem for?

Mainly researcher.

Open on GitHub → Ask about another repo

This repo across BitVibe Labs

Verify against the repo before relying on details.