Vector Commands Documentation

Features

VECTOR.* Commands

Basic Vector Operations

VECTOR.ADD

Add a vector to an index with optional metadata.

Syntax:

VECTOR.ADD <index> <id> <vector> [key value ...]

Parameters:

Examples:


# Add a simple vector
VECTOR.ADD my-index doc1 "0.1,0.2,0.3,0.4"

# Add vector with metadata
VECTOR.ADD my-index doc2 "0.5,0.6,0.7,0.8" title "Document 2" category "AI" author "John Doe"

Returns: OK on success

VECTOR.GET

Retrieve a vector and its metadata by ID.

Syntax:

VECTOR.GET <index> <id>

Examples:

VECTOR.GET my-index doc1

Returns: Array containing [id, vector_data, key1, value1, key2, value2, ...] or null if not found

VECTOR.DEL

Delete a vector from an index.

Syntax:

VECTOR.DEL <index> <id>

Examples:

VECTOR.DEL my-index doc1

Returns: 1 if vector was deleted, 0 if not found

VECTOR.COUNT

Get the number of vectors in an index.

Syntax:

VECTOR.COUNT <index>

Examples:

VECTOR.COUNT my-index

Returns: Integer count of vectors

VECTOR.LIST

List all vector IDs in an index.

Syntax:

VECTOR.LIST <index>

Examples:

VECTOR.LIST my-index

Returns: Array of vector IDs

VECTOR.STATS

Get statistics about a vector index.

Syntax:

VECTOR.STATS <index>

Examples:

VECTOR.STATS my-index

Returns: Array with statistics:

[
  "vector_count", 150,
  "index_count", 1,
  "avg_dimension", "384.00",
  "min_dimension", 384,
  "max_dimension", 384,
  "avg_metadata_keys", "2.50"
]

Vector Search Operations

VECTOR.SEARCH

Perform similarity search with advanced filtering options.

Syntax:

VECTOR.SEARCH <index> <vector> <limit> [METRIC <metric>] [THRESHOLD <threshold>] [key value ...]

Parameters:

Examples:


# Basic similarity search
VECTOR.SEARCH my-index "0.1,0.2,0.3,0.4" 5

# Search with cosine similarity
VECTOR.SEARCH my-index "0.1,0.2,0.3,0.4" 10 METRIC COSINE

# Search with threshold
VECTOR.SEARCH my-index "0.1,0.2,0.3,0.4" 5 METRIC COSINE THRESHOLD 0.8

# Search with metadata filters
VECTOR.SEARCH my-index "0.1,0.2,0.3,0.4" 5 METRIC COSINE category AI author "John Doe"

Returns: Array of results, each containing [id, score, vector_data, metadata...]

VECTOR.KNN

Perform k-nearest neighbors search.

Syntax:

VECTOR.KNN <index> <vector> <k> [METRIC <metric>]

Parameters:

Examples:


# Find 5 nearest neighbors
VECTOR.KNN my-index "0.1,0.2,0.3,0.4" 5

# KNN with specific metric
VECTOR.KNN my-index "0.1,0.2,0.3,0.4" 3 EUCLIDEAN

Returns: Array of results, each containing [id, score]

FT.* Commands (RedisSearch Compatible)

Index Management

FT.CREATE

Create a vector index with specified dimension and distance metric.

Syntax:

FT.CREATE <index> DIM <dimension> [DISTANCE_METRIC <metric>]

Parameters:

Examples:


# Create index with default cosine similarity
FT.CREATE my-ft-index DIM 384

# Create index with Euclidean distance
FT.CREATE my-ft-index DIM 384 DISTANCE_METRIC EUCLIDEAN

Returns: OK on success

FT.INFO

Get information about a vector index.

Syntax:

FT.INFO <index>

Examples:

FT.INFO my-ft-index

Returns: Array with index information including vector count, dimensions, and index configurations

Vector Operations

FT.ADD

Add a vector to an index (equivalent to VECTOR.ADD).

Syntax:

FT.ADD <index> <id> <vector> [key value ...]

Examples:

FT.ADD my-ft-index doc1 "0.1,0.2,0.3" title "Document 1"

Returns: OK on success

FT.DEL

Delete a vector from an index (equivalent to VECTOR.DEL).

Syntax:

FT.DEL <index> <id>

Examples:

FT.DEL my-ft-index doc1

Returns: 1 if deleted, 0 if not found

FT.SEARCH

Search for similar vectors (equivalent to VECTOR.SEARCH with DISTANCE_METRIC parameter).

Syntax:

FT.SEARCH <index> <vector> <limit> [DISTANCE_METRIC <metric>] [key value ...]

Examples:

FT.SEARCH my-ft-index "0.1,0.2,0.3" 5 DISTANCE_METRIC COSINE

GraphRAG Hybrid Retrieval

Combine vector similarity with graph traversal to retrieve context-rich results.

# Build graph of documents and concepts
GRAPH.QUERY rag_graph 'CREATE (:Document {id:"doc1", title:"GraphRAG overview"})-[:MENTIONS]->(:Concept {name:"Knowledge Graph"})'
GRAPH.QUERY rag_graph 'CREATE (:Document {id:"doc2", title:"RAG with Redis"})-[:MENTIONS]->(:Concept {name:"Vector Search"})'

# Create and populate vector index
FT.CREATE rag_vectors DIM 384 DISTANCE_METRIC COSINE
VECTOR.ADD rag_vectors doc1 "0.10,0.20,0.30,0.40" title "GraphRAG overview" concept "Knowledge Graph"
VECTOR.ADD rag_vectors doc2 "0.25,0.10,0.05,0.60" title "RAG with Redis"    concept "Vector Search"

# Retrieve candidates by similarity
VECTOR.SEARCH rag_vectors "0.12,0.21,0.29,0.39" 5 METRIC COSINE

# Expand candidates via graph neighbors
GRAPH.RO_QUERY rag_graph 'MATCH (d:Document {id:"doc1"})-[:MENTIONS]->(c:Concept) RETURN d.title, c.name'

# Alternative route via FT.SEARCH
FT.SEARCH rag_vectors "0.12,0.21,0.29,0.39" 10 DISTANCE_METRIC COSINE

See docs/content/graph/graph_commands.md for more graph commands and profiling tools.

Returns: Array of search results

Similarity Metrics

Metric Aliases Description Use Case
COSINE COS Cosine similarity (normalized dot product) Text embeddings, semantic similarity
EUCLIDEAN L2, EUCL Euclidean distance (L2 norm) Spatial data, image features
DOT DOTPRODUCT, IP, INNER Dot product similarity When magnitude matters
MANHATTAN L1, MAN Manhattan distance (L1 norm) High-dimensional sparse data

Vector Data Formats

Vectors can be provided in multiple formats:

VECTOR.ADD my-index doc1 "0.1,0.2,0.3,0.4,0.5"

Array Format

VECTOR.ADD my-index doc1 ["0.1", "0.2", "0.3", "0.4", "0.5"]

Performance Tips

  1. Use appropriate dimensions: Higher dimensions provide more precision but slower search
  2. Choose the right metric:
    • Use COSINE for normalized vectors (embeddings)
    • Use EUCLIDEAN for coordinate data
    • Use DOT when magnitude is important
  3. Leverage metadata filtering: Pre-filter by metadata before vector similarity
  4. Set reasonable thresholds: Avoid returning too many low-similarity results

Usage Examples

Complete Workflow Example


# 1. Create vectors with metadata
VECTOR.ADD ml-embeddings doc1 "0.1,0.2,0.3,0.4" title "Machine Learning Basics" category "AI" difficulty "beginner"
VECTOR.ADD ml-embeddings doc2 "0.2,0.3,0.4,0.5" title "Deep Learning Advanced" category "AI" difficulty "advanced"  
VECTOR.ADD ml-embeddings doc3 "0.8,0.7,0.6,0.5" title "Statistics Fundamentals" category "Math" difficulty "intermediate"

# 2. Get vector count and statistics
VECTOR.COUNT ml-embeddings
VECTOR.STATS ml-embeddings

# 3. Search for similar vectors
VECTOR.SEARCH ml-embeddings "0.15,0.25,0.35,0.45" 5 METRIC COSINE THRESHOLD 0.7

# 4. Search with metadata filtering (only AI category)
VECTOR.SEARCH ml-embeddings "0.15,0.25,0.35,0.45" 3 METRIC COSINE category AI

# 5. Find nearest neighbors
VECTOR.KNN ml-embeddings "0.15,0.25,0.35,0.45" 3

# 6. Get specific vector
VECTOR.GET ml-embeddings doc1

# 7. List all vectors
VECTOR.LIST ml-embeddings

# 8. Delete vector
VECTOR.DEL ml-embeddings doc3

FT Commands Example


# 1. Create FT index
FT.CREATE semantic-search DIM 384 DISTANCE_METRIC COSINE

# 2. Add vectors
FT.ADD semantic-search embedding1 "0.1,0.2,..." title "Document Title"

# 3. Search 
FT.SEARCH semantic-search "0.1,0.2,..." 10 DISTANCE_METRIC COSINE

# 4. Get index info
FT.INFO semantic-search

Integration with Applications

Python Example

import redis

# Connect to Orbit vector store
r = redis.Redis(host='127.0.0.1', port=6381)

# Add vector with metadata
r.execute_command('VECTOR.ADD', 'my-index', 'doc1', 
                  '0.1,0.2,0.3,0.4', 
                  'title', 'My Document',
                  'category', 'AI')

# Search for similar vectors
results = r.execute_command('VECTOR.SEARCH', 'my-index', 
                           '0.1,0.2,0.3,0.4', 5, 
                           'METRIC', 'COSINE',
                           'THRESHOLD', '0.8')

Node.js Example

const redis = require('redis');
const client = redis.createClient({host: '127.0.0.1', port: 6381});

// Add vector
await client.sendCommand(['VECTOR.ADD', 'my-index', 'doc1', 
                         '0.1,0.2,0.3,0.4', 
                         'title', 'My Document']);

// Search vectors
const results = await client.sendCommand([
  'VECTOR.SEARCH', 'my-index', '0.1,0.2,0.3,0.4', '5'
]);

Error Handling

Common error responses:

Advanced Use Cases


# Store document embeddings
VECTOR.ADD docs doc1 "..." title "Introduction to AI" tags "ai,ml,intro"
VECTOR.ADD docs doc2 "..." title "Deep Learning" tags "ai,dl,advanced"

# Find semantically similar documents
VECTOR.SEARCH docs "query_embedding..." 10 METRIC COSINE THRESHOLD 0.7

Image Similarity


# Store image feature vectors
VECTOR.ADD images img1 "..." filename "photo1.jpg" category "landscape"
VECTOR.ADD images img2 "..." filename "photo2.jpg" category "portrait"

# Find similar images
VECTOR.SEARCH images "query_features..." 5 METRIC EUCLIDEAN

Recommendation System


# Store user preference vectors
VECTOR.ADD users user123 "..." age "25" location "NYC"

# Find similar users
VECTOR.KNN users "user_profile..." 10 COSINE

This implementation provides production-ready vector storage and search capabilities with full Redis protocol compatibility, making it easy to integrate into existing applications and workflows.