Class: Phronomy::VectorStore::RedisSearch

Inherits:
Base
  • Object
show all
Defined in:
lib/phronomy/vector_store/redis_search.rb

Overview

Redis-backed vector store using the RediSearch module (FT.* commands).

Requires:

  • The +redis+ gem (add to your Gemfile)
  • A Redis server with the RediSearch (RedisSearch) module enabled (or Redis Stack which bundles RediSearch)

Vectors are stored as FLOAT32 binary blobs in Redis Hash fields and searched using the KNN approximate-nearest-neighbour algorithm.

Examples:

Usage

redis = Redis.new(url: "redis://localhost:6379")
store = Phronomy::VectorStore::RedisSearch.new(redis: redis, dimension: 1536)
store.add(id: "doc1", embedding: [0.1, 0.9], metadata: {text: "hello"})
results = store.search(query_embedding: [0.1, 0.8], k: 5)

Instance Method Summary collapse

Constructor Details

#initialize(redis:, index_name: "phronomy_vectors", dimension: nil) ⇒ RedisSearch

Returns a new instance of RedisSearch.

Parameters:

  • redis (Redis) —

    configured Redis client

  • index_name (String) (defaults to: "phronomy_vectors") —

    RediSearch index name

  • dimension (Integer, nil) (defaults to: nil) —

    vector dimension; auto-detected on first add. When connecting to an existing RediSearch index, you MUST pass dimension: explicitly. Without it, a freshly constructed instance treats the index as uninitialized until #add is called, and #search silently returns [] in the meantime.



33
34
35
36
37
38
39
40
41
42
43
44
45
46
 
# File 'lib/phronomy/vector_store/redis_search.rb', line 33

def initialize(redis:, index_name: "phronomy_vectors", dimension: nil)
  begin
    require "redis"
  rescue LoadError
    raise LoadError,
      "redis gem is required for Phronomy::VectorStore::RedisSearch. " \
      "Add `gem 'redis'` to your Gemfile."
  end
  @redis = redis
  @index_name = index_name
  @dimension = dimension
  @index_created = false
  @mutex = Mutex.new
end

Instance Method Details

#add(id:, embedding:, metadata: {}) ⇒ Object

Parameters:

  • id (String)
  • embedding (Array<Float>)
  • metadata (Hash) (defaults to: {}) 


51
52
53
54
55
56
57
58
59
60
61
62
63
 
# File 'lib/phronomy/vector_store/redis_search.rb', line 51

def add(id:, embedding:, metadata: {})
  # Establish expected dimension on first add (not race-free for concurrent
  # first adds), then validate, then create/reuse the index.
  @dimension ||= embedding.size
  validate_embedding_dimension!(embedding, @dimension)
  ensure_index!(@dimension)
  @redis.call(
    "HSET", "#{DOC_PREFIX}#{id}",
    "embedding", pack_vector(embedding),
    "metadata", metadata.to_json
  )
  self
end

#clear ⇒ Object 



95
96
97
98
99
100
101
102
103
104
105
 
# File 'lib/phronomy/vector_store/redis_search.rb', line 95

def clear
  @mutex.synchronize do
    begin
      @redis.call("FT.DROPINDEX", @index_name, "DD")
    rescue => e
      raise unless e.message.to_s.include?("Unknown Index name")
    end
    @index_created = false
  end
  self
end

#remove(id:) ⇒ Object 



90
91
92
93
 
# File 'lib/phronomy/vector_store/redis_search.rb', line 90

def remove(id:)
  @redis.call("DEL", "#{DOC_PREFIX}#{id}")
  self
end

#search(query_embedding:, k: 5) ⇒ Array<Hash>

Returns sorted by descending similarity score.

Parameters:

  • query_embedding (Array<Float>)
  • k (Integer) (defaults to: 5)

Returns:

  • (Array<Hash>) —

    sorted by descending similarity score



68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
 
# File 'lib/phronomy/vector_store/redis_search.rb', line 68

def search(query_embedding:, k: 5)
  # search never establishes dimension.  If dimension is unknown and the
  # index has not been created yet, there are no documents to return.
  return [] if @dimension.nil? && !@index_created

  validate_embedding_dimension!(query_embedding, @dimension)
  ensure_index!(@dimension)
  k_safe = Integer(k)
  blob = pack_vector(query_embedding)

  raw = @redis.call(
    "FT.SEARCH", @index_name,
    "*=>[KNN #{k_safe} @embedding $BLOB AS score]",
    "PARAMS", 2, "BLOB", blob,
    "SORTBY", "score",
    "RETURN", 2, "score", "metadata",
    "DIALECT", 2
  )

  parse_results(raw)
end