Using Vectors with AI Agents

Vectors are integrated with AI agents to enable RAG (Retrieval-Augmented Generation), allowing agents to retrieve relevant context from your data and provide more accurate, context-aware responses.

What is RAG?

RAG (Retrieval-Augmented Generation) combines:

Retrieval: Finding relevant information from vector stores
Augmentation: Adding retrieved context to AI agent prompts
Generation: AI agent generates responses using retrieved context

This enables AI agents to:

Access Your Data: Use your actual data as context
Provide Accurate Answers: Answer questions based on your data
Stay Up-to-Date: Use current data, not just training data
Domain-Specific: Understand your specific domain and terminology

How Vectors Work with AI Agents

Vector Configuration

When configuring an AI agent, you can associate vectors:

Select Vector: Choose a vector store to use with the agent
Configure Parameters: Set search parameters (top-K, similarity threshold)
Filter Data Sources: Optionally filter to specific Objects/Files/Reports
Enable RAG: Agent uses vector for context retrieval

Context Retrieval Process

When an AI agent receives a query:

Query Analysis: Agent analyzes the user's query
Vector Search: Agent searches the configured vector(s) for relevant context
Context Retrieval: Retrieves top-K most similar results above threshold
Context Augmentation: Adds retrieved context to the agent's prompt
Response Generation: Agent generates response using retrieved context

Multiple Vectors

Agents can be configured with multiple vectors:

Multiple Context Sources: Use different vectors for different types of context
Combined Context: Combine context from multiple vectors
Specialized Vectors: Use specialized vectors for specific domains

Configuring Vectors for AI Agents

Vector Selection

Select which vector(s) to use with the agent:

Field: Vector Configuration

How to configure:

Navigate to AI Agent configuration
Find the Vector Configuration section
Select one or more vectors from the dropdown
Each vector can have its own configuration

Considerations:

Relevance: Choose vectors relevant to the agent's purpose
Data Coverage: Ensure vectors contain data the agent needs
Performance: Consider performance impact of multiple vectors

Search Parameters

Configure search parameters for each vector:

Top K

Set the number of results to retrieve from the vector.

Field: Top K

Range: 1 to 100

Default: 3

Recommendation:

Start with 3-5 results
Increase if agent needs more context
Decrease if context is too verbose

Similarity Threshold

Set the minimum similarity score for retrieved results.

Field: Similarity Threshold

Range: 0.0 to 1.0

Default: 0.7

Recommendation:

Higher threshold (0.7-0.9) for precise, relevant context
Lower threshold (0.5-0.7) for broader context
Adjust based on agent response quality

Data Source Filtering

Filter vector search to specific data sources:

Object Filtering

For Object vectors, filter to specific Objects:

Field: Objects Filter

How to configure:

Select specific Objects from multi-select dropdown
Only those Objects will be searched
Useful when agent needs specific data subsets

Use Cases:

Agent focused on specific business areas
Limiting context to relevant Objects
Improving search performance

File Filtering

For file vectors, filter to specific files:

Field: Files Filter

How to configure:

Select specific files from multi-select dropdown
Only those files will be searched
Useful for document-specific agents

Use Cases:

Agent focused on specific documents
Limiting context to relevant files
Document-based knowledge agents

Report Filtering

For report vectors, filter to specific reports:

Field: Reports Filter

How to configure:

Select specific reports from multi-select dropdown
Only those reports will be searched
Useful for report-specific agents

Use Cases:

Agent focused on specific report types
Limiting context to relevant reports
Report analysis agents

Vector Configuration Best Practices

Vector Selection

Relevant Vectors: Choose vectors that contain relevant data
Quality over Quantity: Fewer, high-quality vectors are better than many low-quality ones
Domain Alignment: Ensure vectors align with agent's domain
Data Freshness: Use vectors with up-to-date data

Parameter Tuning

Start Conservative: Start with default parameters
Test and Adjust: Test agent responses and adjust parameters
Balance Context: Balance between too little and too much context
Monitor Performance: Monitor agent performance and adjust accordingly

Filtering Strategy

Use Filters When Needed: Filter when agent needs specific data subsets
Avoid Over-Filtering: Don't filter too aggressively, may miss relevant context
Test Without Filters: Test agent without filters first, then add if needed
Document Filters: Document why filters are used

Agent Response Quality

Good Context Retrieval

Signs of good context retrieval:

Relevant Responses: Agent provides relevant, accurate responses
Uses Your Data: Responses reference your actual data
Context-Aware: Agent understands context from your data
Accurate Information: Information is accurate and up-to-date

Poor Context Retrieval

Signs of poor context retrieval:

Irrelevant Responses: Agent provides irrelevant information
Generic Responses: Responses don't use your data
Outdated Information: Information is outdated or incorrect
Missing Context: Agent lacks necessary context

Improving Context Retrieval

To improve context retrieval:

Adjust Similarity Threshold: Lower threshold for more context, higher for more relevant
Increase Top K: Retrieve more results for broader context
Improve Vector Data: Ensure vectors contain relevant, high-quality data
Use Filters: Filter to relevant data sources
Test Different Vectors: Try different vectors or combinations

Use Cases

Customer Support Agent

Configure agent with customer data vector:

Vector: Customer data vector with Objects (Customers, Orders, Products)
Top K: 5-10 results
Threshold: 0.7
Use Case: Answer customer questions using actual customer data

Document Q&A Agent

Configure agent with document vector:

Vector: Document vector with uploaded files
Top K: 3-5 results
Threshold: 0.8
Use Case: Answer questions about documents

Workflow Assistant Agent

Configure agent with workflow vector:

Vector: Workflow vector with MFlows
Top K: 3-5 results
Threshold: 0.7
Use Case: Help users understand and use workflows

Vectors Introduction - Overview of vectors
Creating Vectors - Learn how to create vectors
Searching Vectors - Understand how to search vectors

Using Vectors with AI Agents

On this page