Models
In Lakehousecat, models are central to every interaction. There are two types of models: Provider Models and Custom Models.
Provider Models are configurations that connect Lakehousecat to an external AI provider such as OpenAI, Anthropic, Google, Azure, AWS, or xAI. They define which underlying model to use, which model type (Chat or Speech-to-Text), and the provider credentials. Provider Models are created and managed exclusively by Administrators, since they involve API keys and incur costs.
Custom Models are built on top of Provider Models and tailored to specific use cases. They connect to data sources, define a system prompt, and expose curated suggestions to users. Custom Models can be created by Administrators and Builders. Once configured and validated, they are shared with end users.
Provider Models
Provider Models configure the connection to a supported AI provider. The administrator selects:
- Provider Type — OpenAI, Anthropic, Google, Azure, AWS, or xAI
- Model Type — Chat or Speech-to-Text (STT)
- Configuration Name — a unique name for this configuration
- Provider credentials — API keys and provider-specific connection settings
After creation, Provider Models can be shared with Builders so they can use them as the base for Custom Models.
Custom Models
Custom Models build on a Provider Model and add business-specific configuration:
- General — name, linked Provider Model, system prompt, description, and visibility settings
- Data Sources — connects the model to structured data sources
- Operations — creates, updates, or deletes the semantic layer over the connected data
- Parameters — fine-tunes model behavior (temperature, top-p, etc.)
- Suggestions — prompt suggestions displayed to users when selecting the model
- Hierarchies — data hierarchies generated from the semantic layer
- Descriptions — column and table-level metadata from the semantic layer
For details on each section, explore the dedicated pages for Provider Models and Custom Models.