Diagonal

Diagonal, which trades as diagonal.sh, runs a hosted service for putting linear and logistic regression models into production. A team trains its model as it normally would, for example with scikit-learn, then calls Diagonal's Python package to publish it as a scalable inference endpoint that any application can query over REST.

What Diagonal offers

• One-step deployment — a single deploy call turns a trained model into a secure, always-on API endpoint.
• Atomic model updates — calling deploy again with the same model ID swaps in new weights with no downtime, so retraining pipelines can push changes quickly.
• Stack-neutral integration — training can run anywhere Python runs, from a laptop notebook to a streaming cluster, while the endpoint stays the same.
• Online learning support — suited to models that are retrained often, including contextual multi-armed bandits.

Who it works with

The service is aimed at teams building machine learning products. Examples it gives include payment providers and banks flagging fraudulent transactions, gaming and customer-facing platforms spotting bots, market participants predicting short-term volatility from price and volume data, and security firms.

Key facts

• Model types: linear and logistic regression
• Interface: Python package plus a REST inference API