Model Builder sessions

A Model Builder session is one build of a semantic model: it works through an approved design in rounds and reports on its card in your chat with the Warehouse Builder. This page follows the session itself. To start a build, see Model Builder; for a guide to every card a session shows, see Model Builder cards.

Where a session shows

• The session card, in the chat at the message that started the build. It stays live while the session works. Its parts, in order: the header (the model as a chip, the state in plain words, a for {name} chip naming who started it), Needs you when something waits (when nothing does, the section folds to Recently decided), the round progress, the last report, the full backfill when one was asked for, and the footer with the session actions.
• The strip above the message box, reading Model Builder · {model} · {round text} · {state}, with a Needs you: {n} badge when something waits on you. Select the strip to open the card: on a desktop the chat jumps to it, on a phone it opens as a sheet. The strip is there only while a session is active or waiting on you.
• The banner in the model's Model AI panel, while the session works: Model Builder is working · round {n} of {m} · started by {name}, with Stop. See Model AI.

What the states mean

The state on the card and the strip says in plain words what is happening:

A state can drop the detail it does not know yet, for example Stopped on its own, or Waiting for the Model AI chat without the activity.

How the rounds run

• A session works in rounds, one after another, inside the model's Model AI chat. It runs on its own as the person who started it, and ordinary changes to the model are applied as the rounds go.
• Changes that need a person stay held and appear on the card under Needs you, where people who can write in the chat approve or reject them. See Model Builder for answering and deciding.
• While the model has no reports yet, the session sets up the fact table and the relationships the design declared without asking you. Anything wider than the approved design still waits for a decision.
• Hover a round's dot on the round track to see its kind, result and duration, for example Round 3 · verify · done · 14 min. Round kinds are build, verify and continuation. The track also shows the session's elapsed time and until {time}, when its window ends.
• A round that ended before it finished shows as partial, with its reason beside it. Its report says what it did reach.

Limits of one session

• 8 rounds, plus one grant of 2 more rounds with Continue with 2 more rounds, which needs the person who started the build. The ceiling is 10 rounds.
• 12 hours from the session's start.
• 30 million tokens, tracked by the bar on the card as {used} of {budget} tokens.
• 2 sessions per model; Start again resets the count.
• One build runs on a model at a time; while a session is open, a second one is refused.

When a session ends on its own

A session that ends by itself says why in plain words. The endings you may meet: it used all its rounds; it reached its 12-hour limit; it used its token budget; it asked for the same thing three times, so a person must decide; two rounds in a row did not finish; the person who started it can no longer use the model's Model AI, or the model was deleted; it finished without a complete model, with the open items naming what is missing; or it could not start or failed while it worked. For the wording of each ending and what to do next, see Model Builder.

The done card

Every ending writes a Model Builder report card with a headline for the outcome:

• Your semantic model is ready when the build finished, or Ready, with charts that failed when a chart was too slow to count.
• Needs your attention when something still waits for a person; Stopped or Failed when the session was stopped or did not finish.

Below the headline the card shows Updated {time}, Owner {role} and the model it built, then:

• Scorecard: Measures (all built, or {n} not built), Tests passed ({passed}/{written}), Model check (Clean, or the error count with its warnings), and Rounds ({used}/{limit}).
• Open items ({n}): up to ten things that still want attention, with blocking items emphasised. For what each kind means, see Model Builder cards.
• Actions: Open the session card while the session still works, then Continue with 2 more rounds, Start again, Open Model AI chat and Open model, as each is offered.

A done card stays current: if the session closes again later for another reason, the card is replaced and reads Needs your attention with a fresh Updated {time}. A session that is continued reads Working again · round {n} of {m} on the card while it works.

Stop, continue or start again

The session card's footer and the done card hold the actions:

• Stop ends the build for now. A dialog asks you to confirm first; it reads "The Model Builder stops working on this model. Anything it made stays; you can start it again."
• Continue with 2 more rounds appears when the session stopped at its round limit. It adds two rounds to the same session, once per session, and belongs to the person who started the build.
• Start again begins a fresh session for the same design, after the current one has ended. It belongs to the person who started the build, and it resets the model's session count. It sends Start the Model Builder again. into the chat.
• Open model opens the semantic model (Semantic models and BI tables); Open Model AI chat opens the model's Model AI panel (Model AI).

Questions, answered

What do the session's state words mean?

The state says in plain words what is happening: Getting ready, Working, Model AI building, round {n} of {m}, Waiting for the Model AI chat, Needs you, Loading full history, Paused: {reason}, Done, Stopped by {name} or Failed. The table above lists each one.

Which of the session's changes happen on its own, and which wait for me?

Ordinary work is applied as the rounds go, so the build does not ask about every change. Anything that needs a person stays held and appears under Needs you on the card. While the model has no reports yet, the fact table and the relationships the design declared are set up without asking; anything wider than the approved design waits for a decision.

Can I use the model before the session finishes?

Yes. The model is created for the build and fills round by round, so you can open it, and its Model AI chat, while the session works. The card, the strip and the panel banner show the progress, and Open model takes you to it.

What does it mean when a round ends as "partial"?

The round stopped before it finished, and its reason follows the name, for example a timeout. What the round completed is kept, and its report says what it reached. The session carries on, unless two rounds in a row do not finish, which ends it and asks for a person.

Why did the done card change after the session ended?

A session can close again after its first done card, for example when it needs a person or its access changed. The card is then updated in place: it reads Needs your attention with a fresh Updated {time} under the headline. A session you continue reads Working again · round {n} of {m}.

Why can't I add more rounds?

Continue with 2 more rounds is offered only when the session stopped at its round limit, it can be used once per session, and it belongs to the person who started the build. Once it has been used, or when a newer session of the model exists, start a new session with Start again.

Where do I find a session after it ends?

Its card stays in the conversation where the build was started, with its last stored status, so you can reopen it from there. The strip above the message box only keeps a session while it is active or waiting on you, so an ended session leaves the strip but not the chat.