Process library by

AI-generated research output, not engineering advice. These designs were produced by AI coding agents (Claude Code) under human direction and are checked by automated tools; they have had no independent human review. They are not approved for construction, operation, procurement or safety decisions and come with no warranty. Check the design maturity and review status in each process's process.yaml before relying on any number.

50 processes, 182 recorded agent runs in 43 of them, about US$650 of recorded agent cost; 49 at concept stage and 1 at screening, none independently reviewed. The figure counts recorded runs only (processes/*/runs/*/run.yaml), with costs as the runs reported them: one run records no cost, and agent work outside a recorded run is not included.

Overview

The process library is a collection of 50 chemical processes kept as code, from air separation, ammonia and methanol to chlor-alkali, sulfuric acid and the utilities a plant needs. Each process is one folder with a manifest, process.yaml: what goes in, what comes out, how far the design is taken and who has reviewed it. Most folders are also ChemE-as-Code engineering projects, checked by the cheme toolkit: each has a block and a process flow diagram in the DEXPI exchange standard, a flowsheet and an equipment list; the most advanced (asu-1, ammonia-1) also have P&IDs and a 3D layout. The home page explains what "chemical engineering as code" means.

The library is the source of the Process Atlas (a preview; it shows the processes marked public, which is all 50, as unreviewed research output), which links the processes through the chemicals they make and use. From the same files it builds open life cycle inventory (LCI) datasets for Brightway and openLCA. All of it is plain text in git, so every change is reviewed and versioned. This is research software in active development.

Features

• One folder per process: 44 engineering projects (plant.yaml) and 6 library utilities (steam, cooling water, chilled water, instrument air, wastewater treatment, flaring).
• A manifest per process (process.yaml, with a JSON Schema): families mapped to the ISIC Rev. 4 and NACE Rev. 2 industry classifications, and separate fields for design maturity and review status.
• A chemicals registry with CAS, InChIKey and ChEBI identities and product grades; it links the processes to each other in the Atlas.
• LCI database builds: a Brightway workbook, an openLCA JSON-LD package and a file for Brightway's multifunctional package (the last two keep the unallocated multi-output datasets), plus a one-call Brightway installer (python/). Background data (e.g. ecoinvent) are referenced by name; no licensed data are shipped.
• Markets and regional scenarios: supply chains per product and region, and one design assessed against several regional backgrounds without copying its folder.
• Benchmarks per process family, with reference ranges from cited open sources.
• Continuous checks: CI validates every plant with cheme, runs the tests and builds the catalog and the LCI database; only processes marked public enter the public build.

Quick start

The Process Atlas shows the published processes. To open every plant of the library in the ChemE-as-Code Viewer on your own machine (cheme serve in library mode, core 1.1.0 or later), you need uv and Node.js. core is still a private repository: until it opens, the install below needs read access to it. core installs pyDEXPI from PyPI.

git clone https://github.com/ChemE-as-Code/core.git && uv tool install './core[serve]'
(cd core/viewer && npm install && npm run build)
git clone https://github.com/ChemE-as-Code/process-library.git && cd process-library
cheme serve --repo . --ui ../core/viewer/dist   # http://127.0.0.1:8765, each plant under /p/<id>/

To work on one process, run cheme in its folder (cd processes/asu-1). The agent skills are not in git: run cheme skills update once in a process folder after cloning, so that Claude Code finds them there. To add a process, copy templates/process/ and follow CONTRIBUTING.md.

Documentation

• Technical documentation of the toolkit: docs.cheme-as-code.org. This library's methodology and governance pages are in docs/ (an MkDocs site, built in CI but not yet published).
• Plain-language pages: cheme-as-code.org and the Process Atlas.
• The full reference of this repository (paths, processes, builds, CI, repository size): REFERENCE.md. The reference sections of the earlier README now live there.
• Adding and publishing a process: CONTRIBUTING.md. The LCI export: lci/README.md.

Citation

Both papers are in preparation and their titles are working titles; neither has a DOI yet. If you use the library, please cite the Process Atlas paper, and ChemE-as-Code for the toolkit. If you use the LCI database, cite the Process Atlas paper (in preparation, 2026), by Artur M. Schweidtmann and Jana M. Weber, and the database version. See also cheme-as-code.org/about/cite.

@misc{schweidtmann_process_atlas,
  author = {Schweidtmann, Artur M. and Weber, Jana M.},
  title  = {Process Atlas: an open, git-native, reproducible library and life cycle inventory of chemical processes (working title)},
  url    = {https://atlas.cheme-as-code.org},
  year   = {2026},
  note   = {in preparation}
}

@misc{schweidtmann_cheme_as_code,
  author = {Schweidtmann, Artur M.},
  title  = {ChemE-as-Code: An Open-Source Platform for AI-Accelerated Chemical Process Engineering (working title)},
  url    = {https://cheme-as-code.org},
  year   = {2026},
  note   = {in preparation}
}

Contributors

	Name	Role
	Artur M. Schweidtmann (LinkedIn)	ChemE-as-Code and Process Atlas (principal investigator); Process Intelligence Research, Department of Chemical Engineering, Delft University of Technology
	Jana M. Weber	Process Atlas; Delft Bioinformatics Lab, Department of Computer Science, Delft University of Technology
	Achmad A. Alimin (Angga)	Safety and HAZOP; layout STL export; Process Intelligence Research, Department of Chemical Engineering, Delft University of Technology
	Sophia Rupprecht	Inherited thermodynamic model selection, phase calculations, units and skills; design case study; Process Intelligence Research, Department of Chemical Engineering, Delft University of Technology
	Elise van Voorbergen	Inherited reaction identification, kinetics retrieval, filtering and unit conversion

Everyone who contributes is credited in the git history: see the contributors of this repository.

License

The process data and designs are licensed under CC BY 4.0, unless a process states otherwise in its process.yaml; third-party data keep their own terms. The code (scripts, schemas and tests) and the skills and scripts inside process folders are licensed under AGPL-3.0-or-later. The one exception is the small LCI data installer in python/, which is under the MIT License so that you can use it in any workflow. Copyright (C) 2026 Artur Schweidtmann; TU Delft disclaims copyright interest (see NOTICE).

Contact

Questions, or working with industry: Artur Schweidtmann, [email protected]. Ways to collaborate: pi-research.org/industry and cheme-as-code.org/about.