Skip to content

Ollama Embeddings (Local)

GrafitoDB can generate embeddings locally using the Ollama embeddings API. This is ideal for offline or privacy-sensitive workloads.

Requirements

  • Ollama installed and running on your machine
  • An embeddings-capable model pulled locally (for example: nomic-embed-text)

Installation

pip install httpx

Basic Usage

from grafito import GrafitoDatabase
from grafito.embedding_functions import OllamaEmbeddingFunction

embed_fn = OllamaEmbeddingFunction(
    model="nomic-embed-text"
)

db = GrafitoDatabase(":memory:")
db.create_vector_index(
    name="docs_vec",
    dim=768,
    embedding_function=embed_fn
)

node = db.create_node(labels=["Doc"], properties={"text": "Local embeddings"})
db.upsert_embedding(node_id=node.id, text="Local embeddings", index="docs_vec")

Custom Host / Remote Ollama

By default, Grafito uses http://localhost:11434 or the OLLAMA_HOST environment variable if set.

export OLLAMA_HOST="http://localhost:11434"

embed_fn = OllamaEmbeddingFunction(
    model="nomic-embed-text",
    base_url="http://localhost:11434"
)

Notes

  • The embedding dimension depends on the model you choose.
  • Ollama generates embeddings one input at a time; batch large datasets carefully.

Next Steps