Features

• Your AI assistant can now do anything you can do in the app. Once it's connected to your workspace, it works directly with your real data and models, building, exploring, and analyzing right alongside you. Turn it on from the new MCP Server page in settings: install the connector and paste in the config it generates.
• Move models in and out of Prediction Lab. Export any model to a portable file to hand off or archive, and import models back in, including exports from legacy software.
• See how well your models separate outcomes. A new Lorenz Curve view shows how sharply each fit discriminates, with a Gini score that sums it up in a single number you can compare across fits.
• Feature Importance. A new page ranks your rating factors by how much each one drives the model's predictions, so the biggest contributors stand out immediately.
• Replace a dataset's file in place. Swap in an updated file for an existing dataset without recreating it from scratch.

Improvements

• Distribution expressions are now structured and validated. Building a distribution dimension is more guided, and errors are caught early with a clear message.
• Long-running operations now report their progress in plain terms, like "Building tree 45/100."
• Pages with mathematical notation load noticeably faster.

Fixes

• Lift charts now bin consistently when filters are applied, so the curve no longer shifts as you slice the data.
• The map attribution panel starts collapsed, giving the map more room.
• Removing an unreachable saved cloud connection from the welcome screen now works, and a stale active-organization selection recovers on its own.
• Starting a support chat no longer leaves a half-created conversation behind if something fails partway through.