Utilities & Configuration

Overview

Vector utilities and AI provider configuration for embedding generation.

Vector Utilities

Functions for working with vectors:

Get vector properties:

GQL
LET v = AI.VECTOR([3.0, 4.0])
RETURN
  AI.DIMENSION(v) AS dims,
  AI.MAGNITUDE(v) AS magnitude

Normalize embeddings:

GQL
MATCH (d:Document)
SET d.normalized_embedding = AI.NORMALIZE(d.embedding)

Combine and normalize vectors:

GQL
MATCH (p:Product)
LET combined = AI.ADD(p.text_embedding, p.image_embedding)
SET p.combined_embedding = AI.NORMALIZE(combined)

Convert vector to list:

GQL
MATCH (d:Document {id: 'DOC-1'})
RETURN d.title, AI.TOLIST(d.embedding) AS embedding_array

Provider Configuration

Configure AI providers for embedding generation.

Supported Providers:

• OpenAI (text-embedding-3-small, text-embedding-3-large)
• Anthropic (Claude)
• Cohere (embed-english-v3.0)
• Ollama (local models)
• Google (Gemini)
• Custom HTTP endpoints

Configure OpenAI API key:

GQL
CALL AI.SETAPIKEY('openai', $OPENAI_API_KEY)

Set default embedding provider:

GQL
CALL AI.SETPROVIDER('openai')

Get current provider:

GQL
RETURN AI.PROVIDER() AS current_provider

Configure local Ollama model:

GQL
CALL AI.SETPROVIDER('ollama', {
  endpoint: 'http://localhost:11434',
  model: 'nomic-embed-text'
})

Configure custom HTTP endpoint:

GQL
CALL AI.SETPROVIDER('custom', {
  endpoint: 'https://my-embeddings.example.com/embed',
  headers: {'Authorization': 'Bearer ' + $API_KEY},
  model: 'my-custom-model'
})

Embedding Models

Different models produce different dimension vectors:

Specify embedding model:

GQL
LET embedding = AI.embed('text to embed', 'text-embedding-3-large')
RETURN AI.DIMENSION(embedding) AS dimensions

Verify dimension compatibility:

GQL
LET embedding = AI.embed('test text')
LET dims = AI.DIMENSION(embedding)

// Ensure consistency with existing embeddings
MATCH (d:Document)
WHERE d.embedding IS NOT NULL
LET existing_dims = AI.DIMENSION(d.embedding)
RETURN dims = existing_dims AS compatible

NOTE

Warning: All vectors in the same index must have the same dimensions. Choose your model carefully before creating embeddings.