Features

• See your teammates' work as it happens. New and archived branches, reviews, feature and model changes, and fits starting and finishing all appear live, with no refresh. Your chats and tool approvals also stay in step across every session you have open.
• The model review is one page. Summary, Dataset, Fit, Lift, and Terms sections scroll in order, and a link can open the review at any of them. Lift and Lorenz curves compare the two fits, and each factor's relativities are charted with confidence intervals, so you can see which effects moved and by how much.
• The dataset review has an Overview tab. Features, geographies, and hierarchies get their own detail panes, and the Columns tab profiles each column's values with an optional target lens. Filter and partition details show the rows they cover, and review maps say how many values match each region.
• Review objects open in a list beside their detail, with unified diffs and the unchanged objects listed too.
• On the desktop, changes made by an MCP agent are attributed to that agent, and you and your local agents can edit each other's branches.
• You can opt in to error reporting from the new Privacy page in settings.
• The chat panel shows the tool approvals you've already given.
• Admins can list members, change their roles, and remove them through the API, and you can list your chats through the API.

Improvements

• When someone else saves first, the editor names who it was and lets you reload their version or overwrite it with yours.
• Modelers can fit, predict, and branch from work they didn't create.
• Distribution recomputes keep running when you leave the chart, and the chart shows when a dataset file is being replaced.
• Review tabs have their own links.
• Type, icons, charts, and status colors across the app follow one design system, and keyboard focus and screen reader announcements are more consistent.
• Fits that take offsets from another model are faster when those offsets span many groups.
• The window controls style setting is hidden where there are no window controls.

Fixes

• The datasets list shows a healthy dataset's status correctly, and replacing a dataset's file no longer leaves it stuck as failed.
• Lift charts and Lorenz curves that stopped loading after the app quit mid-compute load again.
• A job that times out, runs out of memory, or crashes now fails with an error, and jobs that hit a transient error before starting are retried.
• Jobs interrupted by quitting the desktop app recover when it restarts.
• Review diffs include changes to dependent objects and feature changes between model versions.
• An open review refreshes when its branch is edited, approved, closed, or merged.
• Live updates reconnect after the connection drops.
• Sending a message in the desktop AI chat works, and the assistant's tool calls are kept across turns.
• Creating, rotating, and deleting an organization's AI configuration works.
• Agent and AI configuration lists load past the first page.
• Desktop sign-in completes reliably, and the sign-in window's Cancel button closes it.
• Dropping files onto a desktop upload area works.
• GBM fit progress no longer stalls in Safari.
• Distribution charts show an error you can retry when a target or partition fails to load.
• List pages show their empty state again.
• The titlebar clock no longer overlaps the forward arrow when the sidebar is collapsed, and the favicon is visible on dark browser tabs.