Colombus - The ML pipelines exploration platform

Requires Docker and Docker Compose.

Deploy

git clone git@github.com:colombus-dev/colombus.git && cd colombus
cp .env.sample.dev .env # then edit .env
docker compose -f docker-compose.prod.yml up -d --build

Update:

docker compose -f docker-compose.prod.yml pull colombus_app colombus_ui
docker compose -f docker-compose.prod.yml up -d

Develop

cp .env.sample.dev .env # then edit .env
docker network create colombus-dev_network  # once
docker compose -f docker-compose.dev.yml up --build

UI → http://localhost:5173 API → http://localhost:8180

Configuration (.env)

The application works out-of-the-box with the default settings provided in .env.sample. If you do not configure optional external services, their corresponding features will simply be disabled in the UI.

Kaggle Integration (Optional)

To enable importing notebooks directly from Kaggle competitions, provide your Kaggle API credentials in your .env file:

KAGGLE_USERNAME=your_kaggle_username
KAGGLE_KEY=your_kaggle_api_key

(You can generate these credentials by going to your Kaggle account settings and clicking "Create New Token".)

Contributing

uv run --with pre-commit pre-commit install

Cite

@article{JOT:issue_2026_03/a10,
  author = {Nicolas Lacroix and Mireille Blay-Fornarino and Philippe Collet and Frédéric Precioso and Sébastien Mosser},
  title = {A Model-Driven Approach To Support The Understanding Of Machine Learning Pipelines},
  journal = {Journal of Object Technology},
  volume = {25},
  number = {3},
  issn = {1660-1769},
  year = {2026},
  month = march,
  editor = {},
  note = {ECMFA 2026 Journal First},
  pages = {3:127-140},
  doi = {10.5381/jot.2026.25.3.a10},
  url = {http://www.jot.fm/contents/issue_2026_03/a10.html}
}