Skip to content

atomatic

atomatic

A simple, opinionated, just enough, yet preposterously powerful framework for agents to run automatic (atomistic) calculation campaigns on HPCs.

Warning

Requires Python 3.12; uv is recommended.

Atomatic is a framework for running atomistic simulation campaigns: small, medium, or absolutely massive, on your laptop or on supercomputers - autonomously. This includes

  • creation of specialized & vetted, yet flexible atomistic simulation workflows with atomate2 and other workflow software
  • submission and management of calculations on complex HPC queueing systems, at scale
  • error correction for the interaction between these moving parts.

One prompt launches an ultra-long horizon materials science specialist which can - without intervention - oversee hundreds of thousands of ab initio jobs, correct thousands of disparate errors over weeks or months, and return auditable trails of execution (scripts, outputs) as well as human-interpretable executive results.

In short, it is a computational "proof engine" for hypotheses about the atomistic design of solids.

Example prompts

  • "Calculate 1,000 2nd-order elastic tensors on NERSC perlmutter with PBEsol and compare them to the Materials Projects' matching entries."
  • "Calculate the DFT band structures and DOS for every structure in this generated dataset of 20,000 perovskites, and show me only those with 1.0 ≤ Eg ≤ 1.4 eV."
  • "Calculate the self-diffusion coefficients of these three mpids using MACE MLIP molecular dynamics."

Components

Component Form Use with
Skills SKILL.md files in atomatic/skills any harness: .agents/skills, .claude/skills, .opencode/skills
Tools stdio MCP server, atomatic-mcp any harness which reads MCP
Harness deepagents agent, create_atomatic_agent Python, or a deepagents-compatible GUI
Addons modular extensions of the skills, tools, and harness atomatic addon --enable NAME; see Addons

Use skills alone, skills and tools with your own harness, or all three for the best results.

Documentation

Page Contents
Setup Prerequisites, runner (HPC) and agent computer setup, atomatic configure, .env reference
Run Full-stack agent (CLI, Python, GUI); skills and MCP tools in your own harness
CLI atomatic configure, atomatic addon, atomatic-mcp
Addons Addon use and specification
Develop Debug scripts, langgraph dev server, skill ablations