Knowledge Base
Description
Searches the current Agent's ingested knowledge for content relevant to a query. Use it directly in a workflow or as a retrieval tool for AI Agent or Conversational Agent.
Prerequisites
Configure the Knowledge Ingestion Service and ingest the knowledge sources required by the Agent. Retrieval uses the current execution's knowledge scope; it does not search arbitrary files supplied by the caller.
Inputs
| Property | Description |
|---|---|
| Query | Search request. For direct use, bind the question from Process Data. As an AI tool, allow the model to supply the query. |
| Chunk Limit | Maximum results. Default: 3. The activity constrains the value to 1–10. |
Outputs
| Output | Description |
|---|---|
| Results | Collection of relevant KnowledgeBaseSearchResult objects. |
| Count | Number of returned results. |
Each result provides:
| Field | Description |
|---|---|
| Text | Retrieved content chunk. |
| SourceName | Optional source document name. |
| SourceLink | Optional source URL. |
| ImageAssets | Related images with Id, Url, and AssociatedText. |
Count = 0 and an empty Results collection mean no content was returned. This can also occur for a blank query or an unavailable knowledge-search service. Do not treat an empty result as proof that the underlying source contains no relevant information.
Example: Retrieve a policy
- Ingest the Agent's policy documents.
- Set Query to
What is the refund policy?and Chunk Limit to3. - Run the activity and check Count.
- Iterate over Results to read Text and preserve any SourceName/SourceLink needed for attribution.
To use it as a tool, connect Knowledge Base through the AI activity's Tools connector and instruct the model to retrieve policy content before answering. Retrieval returns context, not a final user-facing answer.