Porous Materials

Extracts synthesis procedures and adsorption isotherms from papers on MOFs, zeolites, COFs and related frameworks.

This is the simplest of the three case studies — no domain metric extractors, no custom writer — which makes it the best template to copy when starting a new domain. Building your own case study walks through exactly how it is assembled.

Source — examples/scripts/case_study_porosity/run.py

Run it#

uv run examples/scripts/case_study_porosity/run.py <pdf_dir> <output_dir> \
    --skip-existing

Requires GEMINI_API_KEY, ANTHROPIC_API_KEY (plot reading) and MISTRAL_API_KEY (OCR). Defaults are gemini-3.0-flash for synthesis, gemini-3.0-pro for material extraction and claude-sonnet-4-20250514 for plots; change the constants at the top of run.py.

Results land as one JSON per material under <output_dir>/<paper_id>/, written by AnnotatedJsonWriter — see Output Format.

An interactive version is in examples/scripts/case_study_porosity/porosity_extraction_gemini.ipynb.

What makes this domain different#

Only two things, both of them strings and lists:

• PlotFilterConfig keeps pressure on the x-axis (bar, kPa, p/p₀, …) and uptake on the y-axis (mmol/g, cm³/g, wt%, …), while vetoing temperature, enthalpy, selectivity and permeability plots — which share vocabulary with isotherms but are not isotherms.
• The material prompt asks for each framework variant separately (different linker, metal node, or activation condition), because porosity is exactly what varies between them.

Both are reproduced in full in Building your own case study.