Knowledge Base
Description
The Knowledge Base activity enables workflows and AI-powered activities to retrieve information from knowledge sources ingested through the IB-X Knowledge Ingestion Service.
The activity performs semantic retrieval against the knowledge associated with the current Agent and returns the most relevant content chunks matching the supplied query.
The Knowledge Base activity is commonly used as a tool by Conversational Agent and AI Agent activities, allowing Large Language Models (LLMs) to retrieve organization-specific knowledge before generating a response.
Although the activity can be used directly within a workflow, it is typically most effective when used in conjunction with an AI Agent or Conversational Agent, as the retrieved content can be interpreted, summarized, and presented in a more natural and contextual manner.
Agent-Specific Knowledge Retrieval
Knowledge retrieval is performed against the knowledge sources configured for the current Agent.
Only knowledge ingested for the current Agent is considered during retrieval.
This ensures:
- Knowledge isolation
- Domain-specific retrieval
- Improved relevance
- Better response quality
- Controlled access to Agent knowledge
For more information about Agent-specific knowledge ingestion, refer to the Knowledge Ingestion Service documentation.
Inputs
Query
Specify the search query used to retrieve relevant information from the Agent's knowledge base.
The query is matched semantically against the embeddings generated during the ingestion process.
Examples:
- What are the supported authentication methods?
- How do I configure an email trigger?
- Explain the Agent Health Model.
- What is the refund policy?
The retrieval engine searches for content that is semantically related to the supplied query rather than relying solely on keyword matching.
Chunk Limit
Specify the maximum number of knowledge chunks to retrieve.
The retrieval engine ranks available content by semantic relevance and returns the highest-ranked chunks up to the configured limit.
Higher values may provide more contextual information but can also increase the amount of retrieved content.
Typical values depend on the use case and the downstream activity consuming the results.
Examples:
| Chunk Limit | Description |
|---|---|
| 3 | Retrieve the top 3 matching chunks |
| 5 | Retrieve the top 5 matching chunks |
| 10 | Retrieve the top 10 matching chunks |
Retrieval Process
The Knowledge Base activity performs the following operations:
- Receive the query
- Generate a semantic search request
- Search the Agent-specific knowledge store
- Rank matching content by relevance
- Retrieve the top matching chunks
- Return the retrieved content
The activity relies on semantic embeddings generated by the Knowledge Ingestion Service.
Tool Usage
The Knowledge Base activity is commonly exposed as a tool to AI-powered activities.
Typical usage patterns include:
- Conversational Agent
- AI Agent
When used as a tool, the AI model can determine when knowledge retrieval is required and automatically invoke the Knowledge Base activity to obtain relevant information.
The retrieved content is then used as context for generating a grounded response.
Conversational Agent Integration
The Knowledge Base activity is frequently used as a tool by Conversational Agents.
Typical flow:
- User asks a question
- Conversational Agent determines knowledge retrieval is required
- Knowledge Base tool is invoked
- Relevant chunks are retrieved
- Retrieved content is supplied to the LLM
- Conversational Agent generates a grounded response
This approach helps improve response quality and reduces unsupported or hallucinated responses.
AI Agent Integration
The Knowledge Base activity is also commonly used as a tool by AI Agents.
Typical flow:
- AI Agent receives a prompt
- AI Agent determines additional knowledge is required
- Knowledge Base tool is invoked
- Relevant chunks are retrieved
- Retrieved content is supplied to the model
- AI Agent generates the final answer
This allows AI Agents to generate responses using organization-specific knowledge rather than relying solely on model knowledge.
Outputs
Retrieved Chunks
Represents the collection of knowledge chunks retrieved from the Agent's knowledge store.
Each chunk contains semantically relevant content associated with the supplied query.
The returned chunks can be:
- Processed directly within a workflow
- Passed to downstream activities
- Used as context for AI-powered activities
Best Practices
- Use the Knowledge Base activity as a tool for Conversational Agent and AI Agent activities whenever possible.
- Select an appropriate Chunk Limit to balance retrieval quality and context size.
- Ensure knowledge sources are properly ingested before performing retrieval operations.
- Use clear and specific queries to improve retrieval relevance.
- Periodically refresh ingested knowledge to ensure retrieval accuracy.
Notes
- Knowledge retrieval is Agent-specific.
- Only knowledge ingested for the current Agent is considered during retrieval.
- Retrieval is based on semantic similarity rather than keyword matching alone.
- The quality of retrieval depends on the quality of the ingested content and generated embeddings.
- The activity can be used directly within workflows or exposed as a tool to AI-powered activities.