Searching Vectors

Vector search enables you to find relevant information from your vector stores using semantic similarity. Unlike traditional keyword search, vector search understands meaning and context, making it ideal for finding related information even when exact keywords don't match.

What is Vector Search?

Vector search uses semantic similarity to find relevant information:

Semantic Understanding: Understands meaning and context, not just keywords
Similarity Matching: Finds similar content based on vector distance
Relevance Scoring: Returns results with similarity scores
Filtered Results: Filter results by data source (Objects, Files, Reports)

How Vector Search Works

Search Process

Query Input: Enter your search query as natural language
Query Embedding: Your query is converted to an embedding using the same model as the vector
Similarity Calculation: System calculates similarity between query embedding and stored embeddings
Result Retrieval: Returns top-K most similar results above the similarity threshold
Result Display: Shows results with content, similarity scores, and metadata

Similarity Scoring

Results are scored based on similarity:

Score Range: 0.0 to 1.0 (0% to 100%)
Higher Scores: Indicate more similar/relevant results
Threshold Filtering: Only results above the similarity threshold are returned
Cosine Similarity: Uses cosine distance to measure similarity

Searching a Vector

Access Vector Search

To search a vector:

Navigate to the Vectors page
Find the vector in the list
Click the "Search" icon next to the vector
The vector search interface opens

Configure Search Parameters

Before searching, configure search parameters:

Set Similarity Threshold

Set the minimum similarity score for results.

Field: Similarity Threshold

Range: 0.0 to 1.0

Default: 0.3 (30%)

How to set:

Use the slider to adjust threshold
Or enter a value directly in the input field
Lower threshold = more results (less strict)
Higher threshold = fewer results (more strict)

Recommendation: Start with 0.3 and adjust based on result quality.

Set Top K

Set the maximum number of results to return.

Field: Top K

Range: 1 to 100

Default: 5

How to set:

Enter the number of results you want
Higher values return more results but may include less relevant ones
Lower values return fewer but more relevant results

Recommendation: Start with 5-10 results and adjust based on needs.

Filter by Data Sources (Optional)

Filter results to specific data sources:

For Object Vectors:

Objects Filter: Select specific Objects to search within
How to select: Use multi-select dropdown to choose Objects
Use Case: Search within specific Objects only

For File Vectors:

Files Filter: Select specific files to search within
How to select: Use multi-select dropdown to choose files
Use Case: Search within specific documents only

For Report Vectors:

Reports Filter: Select specific reports to search within
How to select: Use multi-select dropdown to choose reports
Use Case: Search within specific report configurations only

Note: Filters are optional. If no filters are selected, all data sources in the vector are searched.

Enter Search Query

Enter your search query:

Field: Search Query

How to enter:

Type your query in natural language
Use descriptive phrases that capture what you're looking for
Examples:
- "customer orders from last month"
- "product pricing information"
- "workflow for order processing"

Query Processing:

Query is automatically debounced (waits 500ms after typing stops)
Query is converted to embedding using the vector's embedding model
Search executes automatically when query is ready

Execute Search

Click the "Search" button or press Enter to execute the search.

Search Execution:

Query embedding is generated
Similarity search is performed
Results are filtered by threshold and top-K
Results are displayed in a table

Understanding Search Results

Result Table

Search results are displayed in a table with three columns:

Content Column

Shows the actual content from the data source:

Text Content: The text that matched your query
Scrollable: Long content can be scrolled within the cell
Source Context: Content comes from the original data source

Score Column

Shows the similarity score:

Percentage: Displayed as percentage (0% to 100%)
Higher = Better: Higher scores indicate more relevant results
Threshold: Only results above threshold are shown

Metadata Column

Shows metadata about the result:

Source Information: Which data source the result came from
Source ID: ID of the source (Object ID, File ID, etc.)
Additional Metadata: Other relevant information
JSON Format: Metadata displayed as formatted JSON

Result Information

Above the results table:

Result Count: Number of results found
Filtered Results: Only results above similarity threshold are shown
Sorted by Score: Results sorted by similarity score (highest first)

Search Best Practices

Query Formulation

Natural Language: Use natural language queries, not keywords
Descriptive: Be descriptive about what you're looking for
Context-Aware: Include context in your query
Examples:
- Good: "customer orders with status pending"
- Less Effective: "orders pending"

Similarity Threshold

Start Low: Start with lower threshold (0.3) to see more results
Adjust Up: Increase threshold if too many irrelevant results
Adjust Down: Decrease threshold if missing relevant results
Domain-Specific: Threshold may vary by domain and data type

Top K Configuration

Start Moderate: Start with 5-10 results
Increase if Needed: Increase if you need more results
Decrease for Precision: Decrease for more focused results
Consider Use Case: Adjust based on your specific use case

Filtering

Use Filters: Use data source filters to narrow search scope
Specific Searches: Filter to specific Objects/Files/Reports when needed
Broad Searches: Remove filters for broader searches
Performance: Filtering can improve search performance

Use Cases

Search for data related to a specific topic:

Example: "customer contact information"
Use Case: Find all customer records and related data
Result: Returns relevant customer records with contact details

Semantic Discovery

Discover information using meaning:

Example: "order processing workflows"
Use Case: Find workflows related to order processing
Result: Returns workflows that handle orders, even if they don't contain exact keywords

Context Retrieval for AI

Retrieve context for AI agents:

Example: "product specifications"
Use Case: Provide context to AI agent about products
Result: Returns relevant product information for AI agent context

Vectors Introduction - Overview of vectors
Creating Vectors - Learn how to create vectors
Using Vectors with AI Agents - Understand how vectors integrate with AI agents

Searching Vectors

On this page