Class: KairosMcp::VectorSearch::SemanticSearch

Inherits:
Base
  • Object
show all
Defined in:
lib/kairos_mcp/vector_search/semantic_search.rb

Overview

Semantic search implementation using hnswlib and informers

Requires optional gems:

- hnswlib (~> 0.9) - HNSW approximate nearest neighbor search
- informers (~> 1.0) - ONNX-based sentence embeddings

This class is only loaded when gems are available.

Constant Summary collapse

DEFAULT_MODEL = 
'sentence-transformers/all-MiniLM-L6-v2'
DEFAULT_DIMENSION = 
384
DEFAULT_SPACE = 
'cosine'

Instance Attribute Summary collapse

Instance Method Summary collapse

Constructor Details

#initialize(index_path:, dimension: DEFAULT_DIMENSION, model: DEFAULT_MODEL) ⇒ SemanticSearch

Returns a new instance of SemanticSearch.



22
23
24
25
26
27
28
29
30
31
32
33
34
35
 
# File 'lib/kairos_mcp/vector_search/semantic_search.rb', line 22

def initialize(index_path:, dimension: DEFAULT_DIMENSION, model: DEFAULT_MODEL)
  @index_path = index_path
  @dimension = dimension
  @model_name = model
  @id_map = {}        # Maps internal index -> document id
  @reverse_map = {}   # Maps document id -> internal index
  @metadata_store = {} # Stores metadata by document id
  @next_index = 0
  @ready = false
  @embedder = nil
  @index = nil

  ensure_index_directory
end

Instance Attribute Details

#dimension ⇒ Object (readonly)

Returns the value of attribute dimension.



20
21
22
 
# File 'lib/kairos_mcp/vector_search/semantic_search.rb', line 20

def dimension
  @dimension
end

#index_path ⇒ Object (readonly)

Returns the value of attribute index_path.



20
21
22
 
# File 'lib/kairos_mcp/vector_search/semantic_search.rb', line 20

def index_path
  @index_path
end

#model_name ⇒ Object (readonly)

Returns the value of attribute model_name.



20
21
22
 
# File 'lib/kairos_mcp/vector_search/semantic_search.rb', line 20

def model_name
  @model_name
end

Instance Method Details

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



37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
 
# File 'lib/kairos_mcp/vector_search/semantic_search.rb', line 37

def add(id, text, metadata: {})
  ensure_initialized

  id_str = id.to_s
  embedding = generate_embedding(text)
  
  # If document already exists, we need to handle update
  if @reverse_map.key?(id_str)
    internal_idx = @reverse_map[id_str]
    # hnswlib doesn't support true updates, so we mark as deleted and add new
    # For simplicity, we just overwrite the point
    @index.add_point(embedding, internal_idx)
  else
    internal_idx = @next_index
    @next_index += 1
    
    # Resize index if needed
    if internal_idx >= @index.max_elements
      # Create new larger index and copy data
      resize_index(@index.max_elements * 2)
    end
    
    @index.add_point(embedding, internal_idx)
    @id_map[internal_idx] = id_str
    @reverse_map[id_str] = internal_idx
  end

  @metadata_store[id_str] = metadata.merge(text: text)
  true
rescue StandardError => e
  warn "[SemanticSearch] Failed to add document #{id}: #{e.message}"
  false
end

#count ⇒ Object 



207
208
209
 
# File 'lib/kairos_mcp/vector_search/semantic_search.rb', line 207

def count
  @id_map.size
end

#load ⇒ Object 



172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
 
# File 'lib/kairos_mcp/vector_search/semantic_search.rb', line 172

def load
  return false unless File.exist?(index_file_path) && File.exist?(metadata_file_path)

  ensure_embedder

  # Load metadata first
  metadata = JSON.parse(File.read(metadata_file_path), symbolize_names: true)
  
  # Validate dimension and model match
  if metadata[:dimension] != @dimension
    warn "[SemanticSearch] Dimension mismatch: expected #{@dimension}, got #{metadata[:dimension]}"
    return false
  end

  @id_map = metadata[:id_map].transform_keys(&:to_i)
  @reverse_map = metadata[:reverse_map].transform_keys(&:to_s)
  @metadata_store = metadata[:metadata_store].transform_keys(&:to_s)
  @next_index = metadata[:next_index]

  # Load HNSW index
  max_elements = [@next_index * 2, 100].max
  @index = create_index(max_elements)
  @index.load_index(index_file_path)
  
  @ready = true
  true
rescue StandardError => e
  warn "[SemanticSearch] Load failed: #{e.message}"
  false
end

#ready? ⇒ Boolean

Returns:

  • (Boolean) 


203
204
205
 
# File 'lib/kairos_mcp/vector_search/semantic_search.rb', line 203

def ready?
  @ready
end

#rebuild(documents) ⇒ Object 



115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
 
# File 'lib/kairos_mcp/vector_search/semantic_search.rb', line 115

def rebuild(documents)
  ensure_embedder
  
  # Initialize fresh index
  max_elements = [documents.size * 2, 100].max
  @index = create_index(max_elements)
  @id_map.clear
  @reverse_map.clear
  @metadata_store.clear
  @next_index = 0
  @ready = true

  # Batch generate embeddings for efficiency
  texts = documents.map { |doc| doc[:text].to_s }
  embeddings = generate_embeddings_batch(texts)

  documents.each_with_index do |doc, i|
    id_str = doc[:id].to_s
    internal_idx = @next_index
    @next_index += 1

    @index.add_point(embeddings[i], internal_idx)
    @id_map[internal_idx] = id_str
    @reverse_map[id_str] = internal_idx
    @metadata_store[id_str] = (doc[:metadata] || {}).merge(text: doc[:text])
  end

  save
  true
rescue StandardError => e
  warn "[SemanticSearch] Rebuild failed: #{e.message}"
  false
end

#remove(id) ⇒ Object 



71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
 
# File 'lib/kairos_mcp/vector_search/semantic_search.rb', line 71

def remove(id)
  id_str = id.to_s
  return true unless @reverse_map.key?(id_str)

  internal_idx = @reverse_map[id_str]
  # hnswlib supports marking elements as deleted
  @index.mark_deleted(internal_idx) if @index.respond_to?(:mark_deleted)
  
  @id_map.delete(internal_idx)
  @reverse_map.delete(id_str)
  @metadata_store.delete(id_str)
  true
rescue StandardError => e
  warn "[SemanticSearch] Failed to remove document #{id}: #{e.message}"
  false
end

#save ⇒ Object 



149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
 
# File 'lib/kairos_mcp/vector_search/semantic_search.rb', line 149

def save
  return true unless @ready && @index

  # Save HNSW index
  @index.save_index(index_file_path)

  # Save metadata
  metadata = {
    id_map: @id_map,
    reverse_map: @reverse_map,
    metadata_store: @metadata_store,
    next_index: @next_index,
    dimension: @dimension,
    model: @model_name
  }
  File.write(metadata_file_path, JSON.pretty_generate(metadata))
  
  true
rescue StandardError => e
  warn "[SemanticSearch] Save failed: #{e.message}"
  false
end

#search(query, k: 5) ⇒ Object 



88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
 
# File 'lib/kairos_mcp/vector_search/semantic_search.rb', line 88

def search(query, k: 5)
  ensure_initialized
  return [] if @id_map.empty?

  embedding = generate_embedding(query)
  actual_k = [k, @id_map.size].min
  
  results = @index.search_knn(embedding, actual_k)
  
  # results format: [[indices], [distances]] or similar
  indices, distances = results
  
  indices.zip(distances).filter_map do |idx, dist|
    next unless @id_map.key?(idx)
    
    id = @id_map[idx]
    {
      id: id,
      score: 1.0 - dist, # Convert distance to similarity score
      metadata: @metadata_store[id] || {}
    }
  end
rescue StandardError => e
  warn "[SemanticSearch] Search failed: #{e.message}"
  []
end

#semantic? ⇒ Boolean

Returns:

  • (Boolean) 


211
212
213
 
# File 'lib/kairos_mcp/vector_search/semantic_search.rb', line 211

def semantic?
  true
end