Vector Search

Redlite supports Redis 8-compatible vector sets for K-NN similarity search. Requires --features vectors when compiling.

Installation

cargo add redlite --features vectors
# or
cargo install redlite --features vectors

Commands

VADD

Add a vector element to a set.

VADD key VALUES n v1 v2 ... element [NOQUANT|Q8|BF16] [SETATTR json]

Parameters:

• key - Vector set key
• VALUES n v1 v2 ... - Vector dimensions (n floats)
• element - Element identifier (string)
• NOQUANT|Q8|BF16 - Optional quantization (default: NOQUANT)
• SETATTR json - Optional JSON attributes

Examples:

# Add a 3-dimensional vector
127.0.0.1:6379> VADD embeddings VALUES 3 0.1 0.2 0.3 item1
(integer) 1

# Add with attributes
127.0.0.1:6379> VADD embeddings VALUES 3 0.4 0.5 0.6 item2 SETATTR '{"category":"books"}'
(integer) 1

# Add with quantization (8-bit)
127.0.0.1:6379> VADD embeddings VALUES 3 0.7 0.8 0.9 item3 Q8
(integer) 1

VSIM

Find similar vectors using K-NN search.

VSIM key (ELE element | VALUES n v1 v2 ...) [COUNT n] [WITHSCORES] [WITHATTRIBS] [FILTER expr]

Parameters:

• key - Vector set key
• ELE element - Reference element in the set
• VALUES n v1 v2 ... - Query vector
• COUNT n - Maximum results to return (default: 10)
• WITHSCORES - Include distance scores
• WITHATTRIBS - Include element attributes
• FILTER expr - Filter by attributes

Examples:

# Find 5 nearest neighbors
127.0.0.1:6379> VSIM embeddings VALUES 3 0.15 0.25 0.35 COUNT 5
1) "item1"
2) "item2"

# With distance scores
127.0.0.1:6379> VSIM embeddings VALUES 3 0.15 0.25 0.35 COUNT 5 WITHSCORES
1) "item1"
2) "0.0087"
3) "item2"
4) "0.1875"

# Using existing element as query
127.0.0.1:6379> VSIM embeddings ELE item1 COUNT 3
1) "item1"
2) "item2"
3) "item3"

# With attribute filter
127.0.0.1:6379> VSIM embeddings VALUES 3 0.1 0.2 0.3 FILTER '@category == "books"'
1) "item2"

VREM

Remove an element from a vector set.

VREM key element

Example:

127.0.0.1:6379> VREM embeddings item1
(integer) 1

VCARD

Get the number of elements in a vector set.

VCARD key

Example:

127.0.0.1:6379> VCARD embeddings
(integer) 3

VDIM

Get the dimensions of vectors in a set.

VDIM key

Example:

127.0.0.1:6379> VDIM embeddings
(integer) 3

VEXISTS

Check if an element exists in a vector set.

VEXISTS key element

Example:

127.0.0.1:6379> VEXISTS embeddings item1
(integer) 1
127.0.0.1:6379> VEXISTS embeddings nonexistent
(integer) 0

VGET

Get the embedding vector for an element.

VGET key element

Example:

127.0.0.1:6379> VGET embeddings item1
1) "0.1"
2) "0.2"
3) "0.3"

VGETALL

Get all elements and their embeddings.

VGETALL key

Example:

127.0.0.1:6379> VGETALL embeddings
1) 1) "item1"
   2) 1) "0.1"
      2) "0.2"
      3) "0.3"
2) 1) "item2"
   2) 1) "0.4"
      2) "0.5"
      3) "0.6"

VGETATTRIBUTES

Get JSON attributes for elements.

VGETATTRIBUTES key element [element ...]

Example:

127.0.0.1:6379> VGETATTRIBUTES embeddings item2
1) "item2"
2) "{\"category\":\"books\"}"

VSETATTRIBUTES

Set JSON attributes for an element.

VSETATTRIBUTES key element json

Example:

127.0.0.1:6379> VSETATTRIBUTES embeddings item1 '{"tags":["new","featured"]}'
OK

VDELATTRIBUTES

Delete attributes from an element.

VDELATTRIBUTES key element

Example:

127.0.0.1:6379> VDELATTRIBUTES embeddings item1
OK

VSIMBATCH

Batch similarity search across multiple vector sets.

VSIMBATCH n key1 key2 ... VALUES m v1 v2 ... [COUNT n] [WITHSCORES]

Example:

127.0.0.1:6379> VSIMBATCH 2 embeddings1 embeddings2 VALUES 3 0.1 0.2 0.3 COUNT 5
1) 1) "embeddings1"
   2) 1) "item1"
      2) "item2"
2) 1) "embeddings2"
   2) 1) "doc1"

Use Cases

Semantic Search

# Index document embeddings
VADD docs VALUES 384 0.1 0.2 ... doc:1 SETATTR '{"title":"Introduction to AI"}'
VADD docs VALUES 384 0.15 0.25 ... doc:2 SETATTR '{"title":"Machine Learning Basics"}'

# Search by query embedding
VSIM docs VALUES 384 0.12 0.22 ... COUNT 10 WITHATTRIBS

Image Similarity

# Index image feature vectors
VADD images VALUES 512 ... img:1 SETATTR '{"path":"/uploads/cat.jpg"}'
VADD images VALUES 512 ... img:2 SETATTR '{"path":"/uploads/dog.jpg"}'

# Find similar images
VSIM images ELE img:1 COUNT 5 WITHSCORES WITHATTRIBS

Recommendation Engine

# Index user embeddings
VADD users VALUES 128 ... user:1
VADD users VALUES 128 ... user:2

# Find similar users for recommendations
VSIM users ELE user:1 COUNT 10

Quantization

Redlite supports three quantization modes to trade off precision for storage:

Mode	Description	Storage	Precision
NOQUANT	Full precision (f32)	4 bytes/dim	Highest
Q8	8-bit quantization	1 byte/dim	Good
BF16	BFloat16	2 bytes/dim	High

Use quantization for large-scale deployments to reduce storage while maintaining search quality.

Implementation Notes

• Backend: sqlite-vec extension for SIMD-accelerated distance calculations
• Storage: Vectors stored as BLOBs in SQLite
• Distance metric: L2 (Euclidean) distance
• Auto-dimension: Vector dimensions are auto-detected from the first element