embedding_ollama()

Use local Ollama models to generate embeddings

Use a local Ollama model to generate embeddings for your vectorizer. Ollama allows you to run open-source models locally for complete data privacy and control.

Purpose

• Define which Ollama model to use
• Specify the dimensionality of the embeddings
• Configure how the Ollama API is accessed
• Configure the model’s truncation behavior and keep alive settings
• Configure optional, model-specific parameters like temperature

Samples

Basic Ollama embedding

SELECT ai.create_vectorizer(
    'blog_posts'::regclass,
    loading => ai.loading_column('content'),
    embedding => ai.embedding_ollama('nomic-embed-text', 768),
    chunking => ai.chunking_character_text_splitter(512)
);

With custom Ollama server

SELECT ai.create_vectorizer(
    'documents'::regclass,
    loading => ai.loading_column('content'),
    embedding => ai.embedding_ollama(
        'nomic-embed-text',
        768,
        base_url => 'http://my.ollama.server:11434'
    ),
    chunking => ai.chunking_character_text_splitter(512)
);

With model options and keep alive

SELECT ai.create_vectorizer(
    'text_data'::regclass,
    loading => ai.loading_column('text'),
    embedding => ai.embedding_ollama(
        'nomic-embed-text',
        768,
        options => '{"num_ctx": 1024, "temperature": 0.5}'::jsonb,
        keep_alive => '10m'
    ),
    chunking => ai.chunking_character_text_splitter(512)
);

Arguments

Name	Type	Default	Required	Description
model	text	-	✔	Name of the Ollama model to use (e.g., nomic-embed-text). The model must already be pulled on your Ollama server
dimensions	int	-	✔	Number of dimensions for the embedding vectors
base_url	text	-	✖	Base URL of the Ollama API. If not provided, uses OLLAMA_HOST environment variable
options	jsonb	-	✖	Additional model parameters such as temperature or num_ctx
keep_alive	text	-	✖	How long the model stays loaded in memory after the request (e.g., 5m, 1h)

Returns

A JSON configuration object for use in create_vectorizer().

Related functions

• embedding_openai(): use OpenAI models
• embedding_litellm(): use any provider through LiteLLM
• embedding_voyageai(): use Voyage AI models