Custom Models

Custom Models in Lakehousecat allow Administrators and Builders to create tailored AI interactions by combining a Provider Model with business-specific data and configuration. Once configured and validated, a Custom Model can be shared with end users.

Creating a Custom Model

Navigate to Workspace → Models → Custom Models and click the + icon.

Required Fields

Field	Description
Name	The display name for the model.
Model ID	Auto-generated from the name. Cannot be changed after creation.
Provider Model	The underlying Provider Model (Chat type) that this Custom Model uses.

An optional Description and Profile Image can also be provided during creation.

After clicking Create, the model editor opens for further configuration.

Model Editor Tabs

The Custom Model editor is organized into the following tabs:

1. General

Configure the core settings of the model:

• Name and Model ID
• Provider Model — the linked Provider Model
• System Prompt — instructions that shape the model's behavior (voice input supported)
• Description — a short description of the model's purpose
• Model State — Enabled and Locked switches
• Agent Mode — activates the conversational AI Agent for this model (see Agent Mode)
• Materialization — pre-computes analytical views for faster chart queries (see Materialization)
• Prompt Suggestion Visibility — controls which suggestion types are shown to users
• Profile Image — an optional image for identification in lists

See General for details.

2. Agent Mode

Enable the conversational AI Agent for this model. When active, users can generate charts, manage platform resources, and interact with their data through natural language. The Agent is role-aware — capabilities depend on the user's role.

See Agent Mode for details.

3. Data Sources

Connect structured data sources to the Custom Model. All connected data sources must have completed semantic extraction before the model's semantic layer can be built. For time-based analysis, include a Timeline data source.

See Data Sources for details.

4. Filters

Define data perspectives on the connected data sources. Filters determine which schemas, tables, columns, and rows are exposed to this model's users. This is the second layer of Lakehousecat's two-layer filter architecture — enabling role-based data perspectives without reloading data.

See Filters for details.

5. Operations

Manage the semantic layer built on top of the connected data sources and filters:

• Create Semantic Model — builds the semantic layer for the first time
• Update Semantic Model — refreshes the semantic layer after data or filter changes
• Delete Semantic Model — removes the semantic layer (⚠️ affects all dependent charts and dashboards)

See Operations for details.

6. Parameters

Fine-tune the model's generation behavior with parameters such as Temperature, Top P, Top K, and Stop Sequence. Leave unchanged unless you have a specific reason to adjust them.

See Parameters for details.

7. Suggestions

Manage the prompt suggestions shown to users when they select this model. Suggestions can be added manually.

See Suggestions for details.

8. Hierarchies

View and manage data hierarchies automatically identified from the semantic layer. Hierarchies can be deactivated or extended manually.

See Hierarchies for details.

9. Descriptions

View and manage column- and table-level metadata generated during semantic layer creation. System-generated descriptions are usually sufficient; manual changes are advanced use cases.

See Descriptions for details.

10. Model Map

A visual graph of the data relationships in the Custom Model's semantic layer. Shows how connected data sources, entities, and metrics relate to each other. Available after the first semantic model build.

See Model Map for details.

11. Audit Log

A chronological history of all configuration changes made to this Custom Model — who changed what and when. Useful for troubleshooting, compliance, and accountability around data access.

See Audit Log for details.

Roles and Responsibilities

Role	Can Create	Can Configure	Can Share
Administrator	Yes	Yes (all tabs)	Yes
Builder	Yes (needs shared Provider Model)	Yes (all tabs)	Yes (to end users)
User	No	No	No

Builders need access to at least one shared Provider Model before they can create a Custom Model.