What the Casabase Cube Agent Can Do

The Casabase Cube agent provides a natural-language interface to Casabase Cube capabilities.

Rather than requiring users to know procedure names, JSON POV syntax, or the physical structure of cube data, the agent can interpret a business request and select an appropriate Casabase Cube tool.

At a high level, agent capabilities fall into four categories:

Casabase Cube Agent
        │
        ├── Discover
        │
        ├── Query and Analyze
        │
        ├── Explain and Assist
        │
        └── Administer
             (when authorized)

The exact capabilities available depend on the Casabase Cube version, the tools configured for the agent, and the privileges of the user.

Discover Cubes

The agent can help users identify the cubes available to them.

For example:

What cubes are available?

Which cube should I use for financial reporting?

Discovery allows the agent to establish the correct analytical context before attempting a data query.

The agent should not assume that a particular cube exists simply because the user’s terminology resembles a common business model.

Explore Dimensions

After identifying a cube, the agent can inspect its dimensional structure.

For example:

What dimensions are in the Finance cube?

Does this cube have a Scenario dimension?

Which dimension contains entities?

This helps translate business terminology into the actual multidimensional model defined in Casabase Cube.

Explore Members and Hierarchies

The agent can use cube metadata to identify members and hierarchy relationships.

For example:

What members are below North America?

Show me the children of Total Product.

What years are available?

Is United States under North America?

Hierarchy discovery is particularly important because natural-language requests often imply hierarchy navigation rather than a simple relational filter.

Conceptually:

"Break North America down"
          │
          ▼
Resolve Entity Hierarchy
          │
          ▼
Select Children of North America

The hierarchy defined in Casabase Cube remains authoritative.

Translate Business Questions into POVs

A central capability of the agent is translating natural language into a Casabase Cube Point of View.

For example:

Show actual revenue for January FY26 in the United States.

can be translated conceptually into:

{
  "Account": ["Revenue"],
  "Scenario": ["Actual"],
  "Years": ["FY26"],
  "Period": ["Jan"],
  "Entity": ["United States"]
}

The user does not need to construct the JSON manually.

The agent can discover the cube structure, resolve the intended members, and build the supported POV required by the Casabase Cube query engine.

See Point of View (POV).

Run Multidimensional Queries

The agent can execute analytical requests through Casabase Cube.

For example:

Show revenue by region for FY26.

Compare Actual and Budget operating income.

Show Gross Margin % by product category.

What was January revenue for the United States?

The resulting values are evaluated by the Casabase Cube query engine rather than independently calculated by the language model. This keeps agent-generated analysis aligned with the same cube semantics used by supported SQL and interactive query paths.

Navigate Hierarchies

The agent can translate natural-language navigation into supported cube selections.

For example:

Show revenue for North America.

followed by:

Break North America down to the next level.

can cause the Entity selection to change from the parent member to an appropriate hierarchy expansion.

Other requests might include:

Show the descendants of Total Product.

Show only the leaf entities under Europe.

Move back to the regional level.

The agent uses Casabase Cube hierarchy metadata and supported POV operators rather than inventing hierarchy relationships.

Change the Analytical Point of View

Users can refine an analysis by changing one or more dimensions.

For example:

Show actual revenue by region for FY26.

then:

Use Budget instead of Actual.

then:

Now show FY25.

then:

Limit it to North America.

Each request changes the effective POV while preserving relevant conversational context.

Snowflake’s agent platform provides the conversational orchestration, while Casabase Cube evaluates each resulting analytical request. Snowflake Cortex Agents can plan requests, select configured tools, and maintain conversational context through threads.

Use Calculated Members

If the cube defines calculated members, the agent can request them just like other cube members.

For example:

Show Gross Margin % by region.

If Gross Margin % is a calculated member defined in Casabase Cube, its result is evaluated by the cube engine.

That evaluation can include:

• Formula dependencies
• Solve order
• Cross-dimensional references
• Hierarchy-aware logic
• Supported time-navigation behavior

The agent does not need to reconstruct the calculation itself.

Use Time-Aware Cube Behavior

The agent can query members whose results depend on the cube’s configured time behavior.

For example:

Show ending balance by quarter.

or:

Compare this period with the prior period.

Where the required behavior is defined by Casabase Cube formulas or time configuration, the query engine applies that behavior.

The agent should use the governed cube model rather than creating a parallel interpretation of time-balance or period-navigation logic.

Compare Scenarios, Periods, and Other Members

Natural language is useful for multidimensional comparisons.

For example:

Compare Actual and Budget revenue by region.

Compare FY26 with FY25.

Show January and February operating income.

The agent can resolve the requested members and include them in the cube POV.

This allows users to ask comparative questions without manually constructing multidimensional selections.

Ask Follow-Up Questions

Users can refine an existing analysis conversationally.

For example:

Show actual revenue by region for FY26.

Which region is largest?

Break that region down by entity.

Now compare it with Budget.

The agent can use conversation context to interpret references such as:

that region

or:

the same period

When additional cube data is required, the agent should execute another governed Casabase Cube query rather than infer values that were not returned by the cube.

Summarize Query Results

After Casabase Cube returns a result, the agent can help explain or summarize it.

For example:

Which region had the highest revenue?

Summarize the Actual versus Budget differences.

What stands out in these results?

This is different from calculating the underlying cube values.

Conceptually:

Casabase Cube
     │
     ▼
Authoritative Result
     │
     ▼
Agent
     │
     ├── Summarize
     ├── Compare
     └── Explain

The analytical values come from Casabase Cube. The natural-language interpretation is provided by the agent.

Visualize Results

Snowflake Cortex Agents can be configured with Snowflake’s Data to Chart capability, which generates visualizations from data returned by other tools.

When this Snowflake capability is enabled for the agent, users may be able to request visualizations such as:

Chart revenue by region.

or:

Show Actual versus Budget as a chart.

Chart generation is a Snowflake Cortex Agent capability, not a separate Casabase Cube calculation engine.

Casabase Cube remains responsible for producing the governed multidimensional data supplied to the visualization.

Explain Cube Structure

The agent can help users understand the model they are working with.

For example:

What does the Entity dimension represent?

What is the parent of United States?

Is Gross Margin a stored or calculated member?

What dimensions are required for this cube?

The available answers depend on the metadata exposed by the configured Casabase Cube discovery tools.

Explain Casabase Cube Usage

If a documentation or knowledge-search capability is configured, the agent can also assist with product questions.

For example:

How do I add a dimension?

What does solve order do?

How do I rebuild a dimension?

How does row-level security work?

These questions should be routed to the configured documentation capability rather than treated as requests for cube data. The current agent setup supports separating documentation questions from cube-data questions.

Assist with Troubleshooting

Where the configured tools expose supported diagnostic capabilities, the agent can help users investigate application behavior.

Examples might include:

Why is this query returning no data?

What should I check if a dimension build fails?

How do I run Health Check?

The agent can combine product guidance with diagnostic tools that the user is authorized to invoke.

The exact diagnostic operations available depend on the installed Casabase Cube version and the agent’s configured tool set.

Administrative Capabilities

Casabase Cube can expose administrative operations through agent tools where appropriate.

These capabilities must remain privilege-controlled.

Conceptually:

General User
    │
    ▼
CUBE_PUBLIC
    │
    ├── Discovery
    └── Querying


Administrator
    │
    ▼
CUBE_ADMIN
    │
    ├── Discovery
    ├── Querying
    └── Authorized Administration

An AI interface does not make an administrative operation less privileged.

A user should not be able to perform an operation through the agent that the user is not authorized to perform through Casabase Cube.

Examples of Administrative Assistance

Depending on the tools exposed by the installed version, an authorized administrator may be able to use the agent to assist with operations such as:

• Reviewing cube configuration
• Reviewing dimension state
• Running supported validation
• Running Health Check
• Reviewing diagnostic information
• Performing other explicitly exposed Casabase Cube administrative operations

Do not assume that every Casabase Cube stored procedure is automatically available to the agent.

Only operations deliberately exposed as agent tools are available.

How Casabase Cube Capabilities Become Agent Tools

Snowflake Cortex Agents support custom tools implemented as stored procedures or UDFs. Snowflake’s agent orchestrator can select these tools based on their names, descriptions, and configured instructions.

This maps naturally to Casabase Cube because application functionality already exists inside Snowflake.

Conceptually:

User Request
     │
     ▼
Cortex Agent
     │
     ▼
Casabase Cube Tool
     │
     ▼
App-Owned Procedure / Function
     │
     ▼
Casabase Cube Engine

Snowflake Native Apps can create Cortex Agents and expose app-owned procedures, UDFs, Cortex Search services, semantic views, and other supported objects through Snowflake’s agent and MCP infrastructure.

Tools Are Explicitly Scoped

The agent does not automatically receive every capability in the application.

Casabase Cube determines which application capabilities are exposed as tools, and Snowflake access controls determine whether the user can invoke the applicable agent and underlying resources.

Snowflake recommends that Native App providers scope tools minimally because every additional tool expands the agent’s access surface. Tool names and descriptions also influence how the agent decides which capability to invoke.

This means the intended model is:

Casabase Cube Functionality
          │
          ▼
Explicitly Exposed Tools
          │
          ▼
User Privileges
          │
          ▼
Effective Agent Capabilities

not:

Install Agent
      │
      ▼
Agent Can Do Everything

Consumer-Owned Objects Remain Controlled

An app-created agent does not receive unrestricted access to customer-owned Snowflake objects.

For consumer-owned objects, Snowflake Native Apps use explicit caller grants where required. App-owned objects receive implicit caller access, while consumer-owned objects require the consumer administrator to grant access to the application.

This allows Casabase Cube agent functionality to remain bounded by the permissions approved by the customer.

See Setting Up the Agent and Governance and Security.

The Agent Does Not Replace Casabase Cube

The agent is an interface to Casabase Cube, not a second analytical engine.

For governed cube analysis:

Agent
  │
  ▼
Interpret Request
  │
  ▼
Use Casabase Cube Tool
  │
  ▼
Casabase Cube
  │
  ▼
Authoritative Result

The agent should not independently recreate:

• Hierarchy aggregation
• Member formulas
• Solve order
• Time balance
• Row-level security
• Other cube semantics

This separation is central to the integration architecture.

The Agent Does Not Invent Governed Metrics

Suppose the cube contains:

Revenue
Cost of Sales
Gross Margin

If Gross Margin is defined as a Casabase Cube calculated member, the agent can query it.

If a user asks for:

Adjusted Strategic Profitability Index

and no such governed metric exists in the cube, the agent should not silently invent a formula and present the result as though it were a Casabase Cube calculation.

It can instead:

• Clarify what the user means
• Identify related existing members
• Explain that the requested governed metric is not defined
• Use another explicitly configured capability when appropriate and clearly distinguish that result from a Casabase Cube metric

This preserves the cube as the authoritative analytical model.

The Agent Does Not Bypass Security

Casabase Cube row-level security continues to apply to supported cube query paths.

Configured security can therefore cause two users asking the same question to receive different valid results because each calculation is based on the data that user is permitted to access.

Security filtering occurs before aggregation and formula evaluation.

As elsewhere in Casabase Cube, a user with application access but no active Casabase Cube security rules for the cube is unrestricted by Casabase Cube row-level security.

See Governance and Security.

The Agent Does Not Override Query Limits

Agent-generated queries remain subject to the same Casabase Cube execution rules as other supported queries.

For example, Casabase Cube can refuse requests that exceed supported calculation semantics or scale.

The current calculation engine supports calculated members on at most two dimensions in a single query, and broad calculated-member selections can also trigger calculation-scale safeguards.

If Casabase Cube refuses the query, the agent should surface or explain the condition and refine the request.

It should not fabricate an approximate result.

What Snowflake Provides Separately

Some capabilities users experience while interacting with the Casabase Cube agent are provided by Snowflake rather than Casabase Cube.

Current Cortex Agent platform capabilities include areas such as:

• Agent orchestration
• Conversation threads
• Data to Chart
• Code execution
• Cortex Search
• Cortex Analyst
• Agent skills
• MCP connectors
• Agent toolsets
• Web search

These are Snowflake platform capabilities and may evolve independently of Casabase Cube.

Their presence in the Cortex Agents platform does not mean every capability is part of the Casabase Cube agent. Only tools deliberately configured by Casabase Cube and authorized in the customer account should be assumed available.

Example Requests

A configured Casabase Cube agent can support conversations such as:

What cubes can I access?

What dimensions are in the Finance cube?

Show the children of North America.

Show Actual revenue by region for FY26.

Break North America down to the next level.

Now compare it with Budget.

Show Gross Margin % instead.

Chart the result.

How is Gross Margin defined?

How does Casabase Cube row-level security work?

The agent may use different tools across that conversation:

Discovery
    │
    ▼
Cube Query
    │
    ▼
Cube Query
    │
    ▼
Snowflake Visualization
    │
    ▼
Metadata / Documentation

The user experiences one conversation, while the agent orchestrates the appropriate governed capabilities behind it.

Key Takeaways

• The Casabase Cube agent provides a natural-language interface to explicitly exposed Casabase Cube capabilities.
• Discovery tools help the agent understand cubes, dimensions, members, and hierarchies.
• Query tools translate business questions into governed multidimensional requests.
• Casabase Cube, not the language model, evaluates cube hierarchies, formulas, aggregation, time behavior, and row-level security.
• Conversational follow-ups can refine the analytical POV.
• Snowflake can provide additional agent capabilities such as visualization, but those are separate from the Casabase Cube calculation engine.
• Administrative operations remain privilege-controlled.
• The agent cannot automatically access every Casabase Cube capability or every customer-owned Snowflake object.
• Unsupported or refused queries should be explained or refined rather than approximated.
• Missing business metrics should not be silently invented and presented as governed cube calculations.

Related Topics

• Intelligence Overview
• How Casabase Cube Works with Snowflake Intelligence
• Setting Up the Agent
• Using Casabase Cube in Snowflake Intelligence
• Governance and Security
• Tips and Limitations
• Point of View (POV)
• Roles and Privileges
• Access Control