Knowledge Base - Smarter retrieval with chunk enrichment
Chunk enrichment involves populating Knowledge Base content with meaningful context, such as source, dates, tags, or other relevant information, to help retrieve more accurate, filtered, or ranked results when performing searches or asking questions.
The benefits of chunk enrichment include improved grounding, fine-grained filtering or scoring, reduced hallucinations, and better confidence when teams rely on AI-generated responses.
You can perform custom processing to determine the index field value when adding to the Knowledge Base.
- Extended index field properties
-
You can create and configure custom index fields in the AI Knowledge Base integrations, enabling more advanced search and filtering. String-based fields support vectorization, search, retrieval, and filtering, while other types provide default filtering. The system automatically upgrades existing fields, and workflow activities support mapping variables to these new field types.
For details, refer to Integrate TotalAgility with AI Knowledge Base in the TotalAgility Advanced Studio Help.
- Enrichment process for the index field value
-
When adding a document to the Knowledge Base, the options for populating index fields are extended, enabling greater flexibility and control over how metadata is generated.
-
Variables: Extended variable type support, allowing index fields to align with the full range of field types.
-
String collections: Ability to populate collection fields using a comma-separated list or a single-column complex/dynamic complex variable.
-
Agent/custom service: Ability to specify a synchronous process that includes one initialization parameter (string) and returns a value matching the index field type. The system creates a job for each chunk, passes in the chunk text, and automatically populates the index field using the value returned by the process
-
- Extended filter operators
-
Filter operators available in the "Knowledge discovery agent" activity, Copilot Insights, and Generative AI chat control for Knowledge Base search are enhanced based on the filter field type, providing complex filtering options. An overall filter operator (AND/OR) is available.
For details, refer to Integrate TotalAgility with AI Knowledge Base in the TotalAgility Advanced Studio Help.
- Retrievable and vector fields
- The use of retrievable and vector fields to customize field selection in search queries enhances the flexibility and efficiency of search configurations, improves the relevance and precision of LLM responses, and optimizes the overall user experience.
- Override query type
- The
Enable query type override setting in the Knowledge Base configuration allows
designers to choose different query types at design time rather than restricting them to the type defined in the Knowledge Base
integration. At runtime, the system uses the overridden query type if specified; otherwise, it falls back to the query type
configured in the Knowledge Base integration.
The Knowledge Base feature is only available to Enterprise customers.