A turn that went wrong used to end in silence: the Thinking line vanished, the answer stopped mid-sentence, and there was nothing to say whether Axi had finished or fallen over. Now it says so, and the conversation is left in a state you can act on.

Asking Axi something used to mean watching a single word — Thinking — until an answer appeared. Now you see the reads as they finish, and the input they sit above has been rebuilt around them.

Axi could describe your tables and declare their keys. It can now write the two things you actually build dashboards on — measures with their business rule baked in, and metrics — and it checks a metric against the engine before offering it to you.

Axi could describe the table in front of you. It can now work through the whole connection — and the left rail saves the result in one press.

Axi could already put a key in your draft. It now reads your data first, and shows you the count it read.

Axi could already explain your semantic model. It can now change it — by putting the change in your draft, where your own edits live, and leaving Save model to you.

The Check card on a metric's page told you the definition compiles. It now also tells you the number: Run executes the metric — unsaved edits included — and shows its value over the last 30 days.

On a BigQuery connection, the table page in Settings → Admin → Data now shows nested fields as rows of their own and keeps flatten models next to the table they come from.

Selecting a column on a table page now opens an inspector beside the list, where the column's dimension is edited in place — no more switching to the Dimensions tab to rename a field or tag it as mappable.

The connection form is no longer a dialog. Each connection's Settings is a page in three sections, and the Data rail now tells you at a glance how big each connection is and whether it works.