​

Knowledge Retrieval

​

How It Works

1. Understand Request - Xagent analyzes your question
2. Search Knowledge Base - Finds relevant documents
3. Retrieve Context - Extracts matching content
4. Generate Response - Combines knowledge with reasoning

​

Using Knowledge in Tasks

​

Natural Language Queries

"Search the product documentation for how to reset the device"
"What does the employee handbook say about remote work?"
"Check the knowledge base for troubleshooting steps for error 404"
"Look up the return policy in our documentation"

• Automatically search relevant knowledge bases
• Find matching documents
• Extract relevant information
• Provide answers based on retrieved content

​

When Knowledge is Used

• Your question references specific information
• You ask to search or look up documentation
• You mention knowledge base, documents, or manuals
• You ask for company policies, procedures, or guidelines

​

Using Knowledge in Agents

​

Configuring Agent Knowledge

1. Go to Build page
2. Create or edit an agent
3. In the Knowledge Bases section:
   • Select one or more knowledge bases
   • Agent will only search these knowledge bases
   • Useful for domain-specific agents

• Customer Support Agent - Attach product documentation and FAQ knowledge bases
• HR Assistant - Attach employee handbook and policy knowledge bases
• Technical Support Agent - Attach troubleshooting and API documentation

​

Agent Behavior

• Agent automatically searches when questions match the domain
• Only searches specified knowledge bases
• Provides answers based on retrieved knowledge
• Cites sources when available

​

Search Options

​

Search Types

• Combines dense (vector) and sparse (keyword) search
• Best balance of relevance and coverage
• Recommended for most use cases

• Pure vector similarity search
• Better for semantic understanding
• Good for conceptual queries

• Pure keyword matching
• Better for exact terms
• Good for specific phrases or names

​

Search Parameters

• Maximum number of results per knowledge base
• Higher values = more results, slower
• Adjust based on your needs

• Minimum relevance threshold (0.0 - 1.0)
• Filters out low-quality matches
• Higher = stricter filtering

​

Search Results

​

What You Get

• Relevant text passages from documents
• Document source and name
• Section or page reference

• Relevance score
• Knowledge base name
• Document information

• Source document name
• Page or section reference
• Link to original document (if available)

​

Interpreting Results

• Very relevant to your query
• Directly addresses the question
• Primary source for answer

• Somewhat relevant
• Contains related information
• May need additional context

• Loosely related
• General background information
• Use with caution

​

Tips for Better Retrieval

​

For Users

Good: "How do I configure API keys in the production environment?"
Less effective: "How do I configure it?"

Good: "What's the SLA for API rate limiting?"
Less effective: "What are the limits?"

Good: "Search the documentation for Docker setup instructions"
Less effective: "How do I set up Docker?"

​

For Agent Builders

• Only attach knowledge bases the agent needs
• Too many = slower, less accurate
• Group related documents together

• Separate knowledge bases by domain
• Product docs vs. policies vs. procedures
• Makes results more relevant

• Remove outdated documents
• Add new information regularly
• Re-upload when content changes significantly

​

For Knowledge Base Managers

• Well-formatted documents process better
• Clear structure and headings
• Remove duplicates and outdated content

• Smaller chunks = more precise results
• Larger chunks = more context
• Adjust overlap for your use case

• Monitor search quality
• Update content regularly
• Remove failed or duplicate documents

​

Troubleshooting

​

No Results Found

• Knowledge base doesn’t contain relevant information
• Search terms don’t match document content
• Min score threshold too high
• Wrong knowledge base selected

• Try different search terms
• Check knowledge base has relevant documents
• Lower min score threshold
• Search broader domain

​

Irrelevant Results

• Documents too generic
• Chunk size too large
• Knowledge base contains unrelated content
• Search query too vague

• Be more specific in your query
• Filter knowledge base content
• Adjust chunk size
• Use domain-specific terminology

​

Slow Search

• Too many knowledge bases
• Large knowledge base size
• High top_k value
• Slow embedding model

• Reduce number of knowledge bases
• Lower top_k value
• Use faster embedding model
• Clean up knowledge base

​

Agent Not Using Knowledge

• Knowledge bases are attached to agent
• Agent has knowledge tool category enabled
• Query matches knowledge base domain
• Knowledge base contains relevant content

• Test agent with knowledge-specific questions
• Check agent configuration in Build page
• Confirm knowledge bases are published

​

Next Steps

• Knowledge Base Overview - Learn about knowledge features
• Uploading Knowledge - Add documents to your knowledge base
• Embedding Models - Configure vector embeddings
• Building Agents - Create agents with knowledge