Parameters
The Parameters tab lets you override the AI models used for semantic extraction on this specific data source. By default, Lakehousecat uses the system-wide model configuration. Use this tab when a particular data source requires a different model — for example, a more capable model for complex schemas or a cost-optimized model for large but simple schemas.
Model Settings
Override Chat Model
Selects the language model used to generate descriptions, synonyms, and semantic metadata during extraction.
- Default: Uses the system-wide chat model configured by the administrator.
- When to override: Use a more capable model (e.g., a larger Claude or GPT-4 variant) for data sources with complex schemas, ambiguous column names, or domain-specific terminology that benefits from stronger language understanding.
Override Embedding Model
Selects the embedding model used to generate vector representations of tables and columns for semantic search.
- Default: Uses the system-wide embedding model configured by the administrator.
- When to override: Switch to a domain-specific or higher-dimensional embedding model when semantic search quality is insufficient for specialized content.
When to Use Model Overrides
| Scenario | Recommendation |
|---|---|
| Complex schema with many tables and ambiguous names | Override to a more capable chat model |
| Large, simple schema where cost matters | Keep system default or override to a cheaper model |
| Domain-specific data (medical, legal, financial) | Override to a model fine-tuned or better suited for that domain |
| Embedding search quality is poor | Override to a higher-quality embedding model |
| System default is already optimal | Leave both fields empty (use system default) |
Model overrides apply only to semantic extraction — they do not affect how users query the data in conversations. Query-time model selection is configured at the Custom Model level.
Notes
- Both fields are optional. An empty field means the system default is used.
- Changes take effect the next time you run semantic extraction from the Operations tab.
- If you change the override model, re-run semantic extraction to regenerate the semantic layer with the new model.