Built-in Embedders

vectlite.embedders provides ready-to-use embedding factories for the most common providers. Each factory returns a Callable[[str], list[float]] suitable for upsert_text() and search_text().

Every provider lazy-imports its SDK, so none of them are runtime dependencies of vectlite. Install only what you use.

Available providers

Factory	Install	Source	Cost
embedders.openai(model)	pip install openai	API	Paid
embedders.cohere(model)	pip install cohere	API	Paid
embedders.voyage(model)	pip install voyageai	API	Paid
embedders.fastembed(model)	pip install fastembed	Local ONNX	Free
embedders.sentence_transformer(model)	pip install sentence-transformers	Local PyTorch	Free
embedders.ollama(model)	(Ollama running locally)	Local HTTP	Free

Usage

import vectlite
from vectlite import embedders

db = vectlite.open("knowledge.vdb", dimension=1536)
embed = embedders.openai("text-embedding-3-small")

vectlite.upsert_text(db, "doc1", "Auth setup guide for SSO", embed, {"source": "docs"})
results = vectlite.search_text(db, "how to authenticate", embed, k=5)

The dimension you pass to open() must match the model's output dimension.

Provider notes

OpenAI

embed = embedders.openai("text-embedding-3-small")  # 1536 dims
embed = embedders.openai("text-embedding-3-large")  # 3072 dims

Reads OPENAI_API_KEY from the environment.

Cohere

embed = embedders.cohere("embed-english-v3.0")  # 1024 dims
embed = embedders.cohere("embed-multilingual-v3.0")

Reads COHERE_API_KEY from the environment.

Voyage AI

embed = embedders.voyage("voyage-3")
embed = embedders.voyage("voyage-3-large")

Reads VOYAGE_API_KEY from the environment.

FastEmbed (local ONNX, no GPU needed)

embed = embedders.fastembed("BAAI/bge-small-en-v1.5")  # 384 dims, fast
embed = embedders.fastembed("BAAI/bge-large-en-v1.5")  # 1024 dims, better quality

First call downloads the ONNX model from HuggingFace Hub. Lightweight, no PyTorch dependency.

SentenceTransformers (local PyTorch)

embed = embedders.sentence_transformer("sentence-transformers/all-MiniLM-L6-v2")  # 384 dims
embed = embedders.sentence_transformer("BAAI/bge-large-en-v1.5", device="cuda")

Heavier than FastEmbed but unlocks the full sentence-transformers ecosystem (cross-encoders, multilingual, instruction-tuned models).

Ollama (local server)

embed = embedders.ollama("nomic-embed-text")
embed = embedders.ollama("mxbai-embed-large")

Requires a running Ollama server at http://localhost:11434. Useful when you already have Ollama for LLMs and want a single backend.

Picking a provider

Constraint	Pick
Easiest start, low scale	openai
Best multilingual	cohere (embed-multilingual-v3.0)
Best quality / context	voyage
Fully offline, lightweight	fastembed
Already on Ollama	ollama
Custom or fine-tuned model	sentence_transformer

Async embedders (Node.js)

Node text helpers accept embed functions that return either a value or a Promise. There is no built-in embedders module on Node — wire up the provider SDK yourself:

const OpenAI = require('openai')
const openai = new OpenAI()

async function embed(text) {
  const res = await openai.embeddings.create({
    model: 'text-embedding-3-small',
    input: text,
  })
  return res.data[0].embedding
}

await vectlite.upsertText(db, 'doc1', 'Hello world', embed)
const results = await vectlite.searchText(db, 'greeting', embed, { k: 5 })