New Features

Introduced the SupportedResponseFormats field for AI Models, allowing specification of response formats such as Any, Text, JSON, Markdown, or Model Specific. This enhancement provides flexibility for future expansions.

Added the ResponseFormat field to AI Prompts, enabling users to define desired response formats aligned with the supported formats of the selected AI model.
Implemented the ModelSpecificResponseFormat field (nvarchar(max)) to accommodate custom JSON schemas for structured outputs, supporting models like OpenAI and Google.

Updated BaseLLM to support the new response format features. Subclasses for providers supporting JSON mode (e.g., OpenAI, Groq) have been configured accordingly for immediate use.

ScopeDefault: Introduced a column with options Users, Admins, AI, All, or any combination thereof (comma-delimited). Entities with All or AI in this field are processed accordingly.
RowsToPackWithSchema: Added a column to specify data packing preferences with options None, All, or Sample. For "type tables," setting this to All includes all rows; for others, Sample can be used to include a subset.
RowsToPackSampleMethod: Defined methods (random, top n, bottom n) to select rows when RowsToPackWithSchema is set to Sample.
RowsToPackSampleCount: Specified the number of rows to include in the sample.
RowsToPackSampleOrder: Determined the ordering of rows for sampling, allowing sorting (e.g., by OrderDate DESC) before selecting the top N rows.

ScopeDefault: Applied at the field level to refine data inclusion, enabling exclusion of unnecessary columns.
AutoUpdateRelatedEntityInfo: A boolean field (defaulting to true) controlling whether CodeGen updates the RelatedEntityID and RelatedEntityFieldName columns. For schemas lacking foreign keys, setting this to false allows manual configuration, providing essential relationship information.
ValuesToPackWithSchema: An attribute with options All, None, or Auto (default). Auto includes associated value lists; All runs a DISTINCT query to gather the full range of underlying table values, enhancing model query capabilities.

AutoUpdateFromSchema: Introduced a boolean field (defaulting to true). Setting this to false grants manual control over relationship records, complementing the above features.

To utilize these enhancements in Skip, upgrade to the upcoming 2.21.0 build. Update prompts to enable JSON as needed, and ensure the new metadata is passed into BaseLLM calls.

Changelog