Knowledge Agent
The Knowledge Agent enhances agent invocations by integrating information from a referenced Knowledge Base.
It enables the agent to respond using internal or domain-specific knowledge, especially useful when user-specific context needs to be considered.
Use the Knowledge Agent when you need the AI to respond based on curated organizational content, such as policies, product documentation, FAQs, or user-specific resources, rather than relying solely on the language model's general knowledge.
Key Features
- Executes a single operation per invocation
- References a defined Knowledge Base
- Can be triggered via:
- Supports both structured and unstructured input/output formats
- Multiple operations can be defined and reused
- Optional auditing of retrieved knowledge content
Operations
Operations are defined in the same way as for other agent types using the operations model.
Each operation specifies the input and expected output. The LLM can use the knowledge base content to enrich or validate its responses.
Only one operation is executed per invocation.
Knowledge Base Usage
The linked knowledge base is queried at runtime to provide additional context for the AI.
This can improve the relevance and accuracy of results, especially in domain-specific applications.
You can also define fallback instructions to control the agent's behavior in cases where no relevant knowledge base content is found as part of the system message.
Auditing
Knowledge base lookups and agent responses can be audited if auditing is enabled.
This allows inspection of what information was used to generate a given answer and improves transparency for compliance-sensitive use cases.
The audit information can be seen and tracked back to the knowledge base in Flowable Control.
Summary
- The Knowledge Agent integrates Flowable's Knowledge Base into agent invocations.
- It supports single-operation tasks and can be used via REST, BPMN, or CMMN.
- Knowledge-enhanced responses improve accuracy in specific domains.
- Supports fallback instructions and optional auditing of retrieved content.