Knowledge Base Agent
Overview
The Knowledge Base Agent is an AI agent that leverages a knowledge base to provide informed responses. It uses RAG (Retrieval-Augmented Generation) to combine knowledge retrieval with AI generation for accurate, context-aware responses based on stored documents and information.
Configuration Details
name
The unique name of the knowledge base agent within the workflow.
String
Yes
-
start
Indicates if this is the starting agent of the workflow.
Boolean
No
false
transition
Defines the transition to the next flow node after this agent completes. See TransitionDef
Object
No
-
answerAfterFinish
Whether the agent should provide an answer message after finishing execution.
Boolean
No
false
answerMessage
Custom message to display when the agent finishes (if answerAfterFinish is true).
String
No
-
systemMessage
System message that defines the agent's role and behavior context for knowledge base interactions.
String
No
"You are a helpful AI Assistant."
userMessage
User message template that defines how user queries are processed against the knowledge base. Can include expressions.
String
Yes
-
userIdentityExpression
Expression to identify the user context for personalized knowledge base responses.
String
No
-
llmConfig
Advanced LLM configuration for the knowledge base agent including temperature and token limits. See LLMConfigDef
Object
No
-
memory
Memory configuration for maintaining conversation context across knowledge base interactions. See ChatMemoryDef
Object
No
-
knowledgeBase
Knowledge base configuration including vector store, embedding model, and retrieval settings. See KnowledgeBaseDef
Object
Yes
-
ragMode
RAG (Retrieval-Augmented Generation) mode for knowledge retrieval and response generation.
String
No
"NAIVE"
loginRequired
Whether user authentication is required to access the knowledge base.
Boolean
No
-
streaming
Whether responses should be streamed back to the user for real-time feedback.
Boolean
No
-
Common Configuration Objects
ChatMemoryDef
Configures how AI agents manage conversation memory including short-term message windows, memory optimization through summarization, and long-term memory for persistent context across sessions.
memoryId
Unique identifier for the memory session
String
No
-
memoryType
Type of memory management strategy
String
No
"message_window"
maxMessages
Maximum number of messages to retain in memory
Integer
No
10
maxTokens
Maximum number of tokens to retain in memory (alternative to maxMessages)
Integer
No
-
memoryOptimizer
Configuration for memory optimization strategies like summarization
Object
No
-
longTermMemory
Configuration for long-term memory that persists across sessions
Object
No
-
MemoryOptimizerDef
Memory Optimizer Definition. Configures strategies for optimizing memory usage through summarization and other compression techniques to maintain relevant context while reducing memory footprint.
summarization
Configuration for conversation summarization to compress older messages
Object
No
ConversationSummarizationDef
Conversation Summarization Definition. Configures how older conversation messages are summarized to compress memory while retaining important context and recent interactions.
retainedMessages
Number of recent messages to retain without summarization
Integer
No
2
promptMessage
Custom prompt for guiding the summarization process
String
No
-
LongTermMemoryDef
Long Term Memory Definition. Configures persistent memory that survives across conversation sessions. Supports different memory types (semantic, episodic, procedural) and scopes (global, local) for various use cases.
memoryId
Unique identifier for the long-term memory instance
String
No
-
scope
Scope of the long-term memory: 'global' (shared across agents) or 'local' (agent-specific)
String
No
-
memoryType
Type of long-term memory: semantic (facts/knowledge), episodic (events/experiences), or procedural (skills/processes)
String
No
"episodic"
LLMConfigDef
LLM Configuration Definition. Defines the configuration for AI Language Models used by AI agents. Includes provider settings, authentication, and parameter overrides for fine-tuning model behavior.
provider
The name of the LLM provider
String
No
apiKey
The API key to access the AI Language Model provider
String
No
overrideParams
Parameters to override for the provider. These settings will take precedence over default model parameters. See LLMParamsDef
Object
No
LLMParamsDef
LLM Parameters Definition. Defines configuration parameters for Language Model behavior including sampling settings, token limits, penalties, and other model-specific options that control AI response generation.
model
The specific model to use for generating responses
String
No
-
temperature
The sampling temperature (0.0 to 2.0). Higher values make output more random, lower values more focused
Double
No
-
topP
The top-p sampling parameter (0.0 to 1.0). Controls nucleus sampling for token selection
Double
No
-
n
The number of completions to generate for each prompt
Integer
No
-
logprobs
The number of log probabilities to return for each token
Integer
No
-
echo
Whether to echo back the prompt in the response
Boolean
No
-
stop
Sequences where the model should stop generating further tokens
List
No
[]
maxTokens
The maximum number of tokens to generate in the response
Integer
No
-
presencePenalty
Presence penalty parameter (-2.0 to 2.0). Positive values penalize new tokens based on whether they appear in the text so far
Double
No
-
frequencyPenalty
Frequency penalty parameter (-2.0 to 2.0). Positive values penalize new tokens based on their existing frequency in the text
Double
No
-
logitBias
Logit bias configuration. Maps token IDs to bias values (-100 to 100) to increase or decrease likelihood of specific tokens
Map<String, Integer>
No
-
required
Additional required parameters specific to the model or provider
Object
No
-
KnowledgeBaseDef
Configuration for AI agents that use knowledge bases for RAG (Retrieval-Augmented Generation). Defines how the agent retrieves information from knowledge bases, including search strategies, retrieval configurations, and reranking options to provide accurate, context-aware responses.
queryStrategy
Strategy for querying the knowledge base
String
No
-
knowledgeBaseCodes
List of knowledge base codes/identifiers to search against
List
No
[]
ragConfig
RAG (Retrieval-Augmented Generation) configuration including history and retrieval settings
Object
No
default RagConfig
multiModalSupport
Whether the knowledge base supports multi-modal content (text, images, etc.)
Boolean
No
-
RagConfig
RAG (Retrieval-Augmented Generation) Configuration. Defines how the knowledge base retrieval and generation process works, including conversation history management and retrieval parameters.
history
Configuration for managing conversation history in RAG context
Object
No
default RagHistoryConfig
retriever
Configuration for the document retrieval component of RAG
Object
No
default RagRetrieverConfig
RagHistoryConfig
RAG History Configuration. Controls how conversation history is managed and used in RAG processing.
messageLimit
Maximum number of previous messages to include in RAG context for better conversation continuity
Integer
No
20
RagRetrieverConfig
RAG Retriever Configuration. Defines how documents are retrieved from the knowledge base, including search parameters, scoring thresholds, and advanced features like reranking and query rewriting.
maxResults
Maximum number of documents to retrieve from the knowledge base
Integer
No
5
minScore
Minimum similarity score threshold for document retrieval (0.0 to 1.0)
Double
No
0.3
kbLocalRerank
Whether to use local knowledge base reranking (deprecated - use rerankConfig instead)
Boolean
No
false
rerankConfig
Configuration for document reranking to improve retrieval quality
Object
No
-
rewriteQueryPrompt
Custom prompt for query rewriting (deprecated - use rewriteQueryConfig instead)
String
No
-
rewriteQueryConfig
Configuration for automatic query rewriting to improve search results
Object
No
-
includeDocReference
Whether to include document references in the response for citation purposes
Boolean
No
false
prefetchDocs
Whether to prefetch documents for faster access during retrieval
Boolean
No
true
allowLLMSearch
Whether to allow LLM-based search in addition to traditional retrieval methods
Boolean
No
false
maxSearchToolCalls
Maximum number of search tool calls allowed during retrieval
Integer
No
2
RewriteQueryConfig
disabled
Whether query rewriting is disabled
Boolean
false
rewriteQueryPrompt
Prompt for query rewriting
String
No
minQueryWords
Minimum words for query rewriting
Integer
-1
RerankConfig
useExternalRerank
Use external rerank
Boolean
true
perKbRerank
Per-knowledge base rerank
Boolean
false
TransitionDef
Defines the flow control mechanism for moving between different states or agents in a workflow. Specifies either a target destination or workflow termination, enabling complex branching and routing logic in AI agent conversations.
end
Whether this transition marks the end of the workflow execution. When true, the workflow terminates. When false, execution continues to the target agent
Boolean
No
false
targetId
The unique identifier of the target agent or node to transition to. Must correspond to a valid agent name defined in the workflow. Ignored if 'end' is true
String
No
-
Example Configuration
{
"name": "documentQAAgent",
"type": "kbagent",
"start": false,
"systemMessage": "You are a document Q&A assistant. Use the knowledge base to provide accurate, well-sourced answers based on company documentation.",
"userMessage": "Based on the company documentation, please answer: {{userQuestion}}",
"userIdentityExpression": "{{user.department}} - {{user.role}}",
"llmConfig": {
"provider": "openai",
"apiKey": "sk-1234567890abcdef",
"overrideParams": {
"model": "gpt-4o",
"temperature": 0.7,
"maxTokens": 1000,
"presencePenalty": 0.1
}
},
"memory": {
"memoryId": "kb-agent-session-123",
"memoryType": "message_window",
"maxMessages": 20,
"maxTokens": 4000,
"memoryOptimizer": {
"summarization": {
"retainedMessages": 5,
"promptMessage": "Summarize the key points from this knowledge base conversation"
}
},
"longTermMemory": {
"memoryId": "user-long-term-kb-123",
"memoryType": "episodic",
"scope": "local"
}
},
"knowledgeBase": {
"queryStrategy": "semantic",
"knowledgeBaseCodes": ["product-docs", "faq-kb", "user-manual"],
"multiModalSupport": true,
"ragConfig": {
"history": {
"messageLimit": 10
},
"retriever": {
"maxResults": 5,
"minScore": 0.3,
"includeDocReference": true,
"rerankConfig": {
"useExternalRerank": true,
"perKbRerank": false
},
"rewriteQueryConfig": {
"disabled": false,
"rewriteQueryPrompt": "Rewrite this query to be more specific and searchable: {query}",
"minQueryWords": 3
}
}
}
},
"ragMode": "NAIVE",
"streaming": false,
"loginRequired": true,
"transition": {
"targetId": "followupAgent"
},
"answerAfterFinish": false,
"answerMessage": "Knowledge base search completed successfully"
}Last updated