Overview
AI Analyst and the Daasity MCP Server now build views from every project in your Looker model, including a shared project that sits between Daasity's base model and your brand projects. Each view's status now says exactly what was checked when it was built, default filters are described in plain words, and the AI prefers the fields you show in Looker. Refreshes on large accounts no longer fail on wide views, and a few fields and measures that could go missing after a refresh are back.
What's new
- Every project in your Looker model is included.Some accounts layer a shared Looker project between Daasity's base model and each brand's project, which is common for companies with several brands. Explores and views defined in that shared project now reach AI Analyst and the MCP Server. Explores built from another explore come through with all of their joins.
- Default filters are described in plain words.When one of your explores applies a filter by default, the view now states it with the field names and values it uses, instead of raw code. The AI applies the filter and can tell you what it applied.
- The AI prefers the fields you show.Fields your explore hides from its field picker are marked as hidden. When a visible field answers the question, the AI uses it and names it. Nothing is removed, and if only a hidden field can answer, the AI says that it isn't part of your explore.
- View status says what was checked.Each view now carries one of four statuses: it builds and returns data; it works but most of its fields returned no values; its latest date hasn't advanced; or it returned no rows. A separate note says whether the view is built directly on a warehouse table or on a table Looker derives, which can run one to two percent off the version Looker persists. Ask the AI about a view's status and it can tell you both.
- Your SKU attribute names stay put.The names you gave your SKU attribute slots no longer disappear from your views when the lookup that supplies them comes back empty during a refresh. The last names found are kept until new ones arrive.
What's fixed
- Measures filtered to the current period are back.Measures that filter to the current reporting period could be left out of a view after a refresh, which took core sales measures off some views. They are included again.
- Views are no longer marked stale by the wrong date.A view with several date columns was judged by the first one alphabetically rather than its sales date, so current views read as stale. Freshness now follows the latest date across the view's date columns and ignores dates in the future.
- Plan and forecast views are populated.Views whose dates run into the future had been coming through empty. They now carry their data.
- Fields on extended explores are included.An explore that extends a Daasity explore on the same table had its extra fields set aside. They come through.
- Wide views build reliably.A view built on a very large table could fail when one of the checks that runs during the build ran out of time. Every build step now has the same longer time budget, and if a join check still runs out of time, that join is set aside with a reason instead of the whole view failing.
What to do
Run Refresh Semantic Layer, then Refresh Views, from the account menu in AI Analyst. If your company has several brand accounts, run it on each one.
Then try two things. Ask a question against an explore from your shared project, if you have one, and check that it answers from that explore. And ask the AI which filters it applied to any answer; it should name them in your field names.