Managing Instructions
Use the Instructions page to control how AI generates BQL (Birst Query Language) queries in your workspace.
You can select a query generation method and provide custom instructions that help the AI interpret user requests and generate more accurate queries.
You can access the Instructions page from: .
Note: Longer instructions can increase processing costs. Keep your instructions concise and focused on information the AI needs to generate queries effectively.
Instruction Settings
The Instructions page includes these settings:
- BQL Generation: This section controls how the AI converts natural language requests into BQL queries across your workspace.
- Method Selection: This section determines which AI model and approach is used to generate BQL queries for standard features, such as the Report Builder, AI Search, and Prompt Playground.
| Method | Description | Best For |
|---|---|---|
| Foundation Model V1 | Best for more complex queries with clean, presentation-ready output. Supports attribute filter value matching. | General-purpose use; recommended starting point for most workspaces. |
| Analytics Agent V1 | Fast and reliable for straightforward queries. Supports attribute filter value matching. Optimized for Analytics Agent workflows. | Workspaces where speed is a priority and queries tend to be simple. |
| Custom Model V1 | Fine-tuned text-to-BQL model with hybrid retrieval. Attribute filter value matching not supported. Optimized for English. | English-only workspaces with well-structured data models and no need for embedding-based filtering. |