Skip to content
M↓ View as Markdown ↗

Creating Models with Steward

With Steward is the recommended way to create Models in Actian AI Analyst. Steward is the AI assistant in Studio. It analyzes your schema, researches business context, and drafts a complete Model: dimensions, measures and, where it makes sense, filters and relationships to your other Models.

Availability

  • Who can do this: members with the Edit capability in the active Workspace.

When to use Steward

Use Steward when you want it to analyze your table structure and sample data, take your business context into account (what you tell it, plus what's in its memory), suggest names and expressions, propose relationships to your other Models along with the Model, and refine the result through conversation.

If the Model maps directly to one table and you just want a quick starting point, From Table is faster.

Create a Model with Steward

  1. In Studio, click Models in the sidebar.
  2. Click + (New model) at the top of the Models list. The Create Model dialog opens.
  3. Select With Steward (marked Recommended).
  4. The dialog closes and the Steward panel opens on the right with a prompt already filled in.
  5. Replace the placeholder in the prompt, add any business context, and send it.

Starting the conversation

The pre-filled prompt reads:

"I want to create a new model based on [describe the table you want to model]. Please help me generate the model with appropriate dimensions and measures."

The prompt is pre-filled in the Steward panel and is not sent automatically. Replace the placeholder with your specific table and any context about your business domain, then send it.

To point Steward at something that's already in your semantic layer, type @ in the message box and pick a Model, Metric or Glossary term from the suggestions.

Example prompts

With business context:

"I want to create a model from the orders table. We're an e-commerce company and orders are our main transaction entity. Include dimensions for order status, payment method, and shipping region."

With specific requirements:

"Create a model from the customers table. I need dimensions for customer tier (gold, silver, bronze) and signup date. For measures, I need total customers and active customers (those who ordered in the last 90 days)."

Reviewing Steward's changes

Steward doesn't change your semantic layer on its own. It proposes a batch of changes, checks the expressions against your warehouse, and shows them in an approval card headed Steward wants to apply edits, with a summary. Click View details to see each proposed change with its before and after values. The Model is created or updated only when you accept.

On the card, choose one of:

  • Accept: apply the proposed changes.
  • Accept, don't ask again in this chat (when shown): apply the changes, and let Steward apply further changes with the same tool in this chat without asking.
  • Request changes: decline the proposal. Nothing is applied. Tell Steward what to change in your next message.

By default, Steward asks before it changes the semantic layer. Members with the Edit capability can change this per tool: in the Steward panel, open Steward settings → Tools, and set each tool to Always allow, Ask first or Always deny.

Tip

If you accept changes and then change your mind, you can roll a chat back. Hover over the message of yours that led to the changes and click Rollback to here. You're asked to confirm. The rollback undoes Steward's changes from that message onward and removes that message and every later message from the conversation. You can't undo a rollback. To keep the conversation, restore an earlier version of the Model from its version history instead.

Plan mode

For complex Models, turn on Plan mode in the Steward message box before you send your prompt. Steward then drafts a plan for you to review before it changes anything. A plan typically covers:

  • Model name and grain
  • Dimensions with their expressions
  • Measures with their aggregations
  • Relationships to other Models

The plan appears in a Proposed plan card. Choose:

  • Approve to let Steward go ahead. Steward then proposes changes, which you review as described above.
  • Edit to change the plan text yourself, then Save.
  • Keep planning to ask Steward to revise the plan.
  • Cancel to stop.

Iterative refinement

After you accept Steward's proposed Model, you can refine it through conversation:

  • "Add a dimension for fiscal quarter"
  • "Rename the revenue measure to total_revenue"
  • "The active_customers filter should use 90 days, not 30"
  • "Add a relationship to the products model via product_id"

Each change goes through the same review step, unless you set the tool to Always allow or chose Accept, don't ask again in this chat.

Give Steward lasting context

Steward reads its Memory at the start of every session. Use it for business context, naming conventions, or known data quirks you'd otherwise repeat in every prompt. Open Steward settings → Memory in the Steward panel, add your notes and click Save memory. Changes to memory are recorded in the audit log.

Tips for best results

  • Tell Steward about your company and domain, and name the dimensions or measures you need.
  • Start simple and add complexity in small steps.
  • If Steward gets something wrong, say what you expected.
  • Read the proposed changes in the review card before you accept, especially expressions and relationship keys.

For more about working with Steward, see Using Steward.