LanceDB Has a Free API: The Serverless Vector Database Built on Lance Format

#ai #vectordb #database #embeddings

Pinecone costs $70/month for 1M vectors. LanceDB stores them on disk for free. Embedded, serverless, and built for AI.

What Is LanceDB?

LanceDB is an open-source vector database built on the Lance columnar format. It's embedded (like SQLite) — no server, no Docker, no infrastructure.

import lancedb
import numpy as np

# Open database (creates directory if needed)
db = lancedb.connect("./my-lancedb")

# Create a table with vectors
data = [
    {"text": "The cat sat on the mat", "vector": np.random.rand(384).tolist()},
    {"text": "Dogs are great pets", "vector": np.random.rand(384).tolist()},
    {"text": "Machine learning is fun", "vector": np.random.rand(384).tolist()},
]
table = db.create_table("documents", data)

# Search
query_vector = np.random.rand(384).tolist()
results = table.search(query_vector).limit(5).to_pandas()
print(results[["text", "_distance"]])

With Embeddings (Auto-Generate Vectors)

from lancedb.embeddings import get_registry
from lancedb.pydantic import LanceModel, Vector

# Use sentence-transformers for embeddings
embedder = get_registry().get("sentence-transformers").create()

class Document(LanceModel):
    text: str = embedder.SourceField()
    vector: Vector(embedder.ndims()) = embedder.VectorField()

# Vectors are generated automatically!
table = db.create_table("docs", schema=Document)
table.add([
    {"text": "The cat sat on the mat"},
    {"text": "Machine learning is transforming industries"},
    {"text": "Python is a great programming language"},
])

# Search with text (auto-embedded)
results = table.search("artificial intelligence").limit(3).to_pandas()

Full-Text Search + Vector Search

# Hybrid search
results = (table
    .search("machine learning", query_type="hybrid")
    .limit(10)
    .to_pandas()
)

Why LanceDB

Feature	Pinecone	LanceDB
Hosting	Managed ($70+/mo)	Embedded (free)
Setup	Account, API key	`pip install lancedb`
Storage	Cloud only	Local disk, S3, GCS
Full-text search	No	Yes
SQL queries	No	Yes
Versioning	No	Built-in (Lance format)