Research Environment Templates for the Catalyst AI Scientist

This repository is used by Catalyst and contains research environment templates for different research areas.

Binary Blobs & Licensing Guidelines

To keep the repository lightweight and ensure compliance with licensing terms, binary files (including PDFs) and assets of unclear licensing must not be checked directly into this git repository.

The blobs.json Convention

Each template folder remains self-contained. To configure binary reference assets for a template, create a blobs.json file inside that template's folder (e.g., bifurcation/blobs.json).

The blobs.json file must contain a flat JSON key-value mapping from the target filename (relative to the template directory) to its source URL:

{
  "there_will_be_a_scientific_theory_of_deep_learning.pdf": "https://arxiv.org/pdf/2604.21691"
}

The root download_blobs.py script dynamically scans the repository for templates, parses these blobs.json configuration files, and downloads or manages the assets.

Usage

To download the binary assets for all templates:

python3 download_blobs.py

To download assets for a specific template (e.g., bifurcation):

python3 download_blobs.py bifurcation

To clean up and delete all downloaded binary assets:

python3 download_blobs.py --clean