Providers

Embex supports a wide range of vector databases. Switching providers is as simple as changing the provider and URL parameters.

The Recommended Path

Stage	Provider	Why?
Development	LanceDB	Zero Setup. Runs locally (embedded). No Docker or API keys needed.
Production	Qdrant / Pinecone / Milvus	Scale. Managed services, high throughput, distributed.

---

LanceDB (Recommended for Dev)

Start here. LanceDB runs in-process and stores data in the local filesystem.

Python

# Persist to local folder (creates ./data directory)
client = await EmbexClient.new_async(provider="lancedb", url="./data")

Node.js

// Persist to local folder (creates ./data directory)
const client = await EmbexClient.newAsync("lancedb", "./data");

Cloud Providers (Production)

When you’re ready to scale, switch to a cloud provider.

Pinecone

Managed serverless vector database.

Python

client = await EmbexClient.new_async(
    provider="pinecone",
    url="https://index-name.svc.pinecone.io",
    api_key="your-api-key"
)

Node.js

const client = await EmbexClient.newAsync(
  "pinecone",
  "https://index-name.svc.pinecone.io",
  "your-api-key"
);

Qdrant

High-performance Rust vector database (Cloud or Self-Hosted).

Python

# Qdrant Cloud
client = await EmbexClient.new_async(
    provider="qdrant",
    url="https://xyz-example.eu-central.aws.cloud.qdrant.io:6333",
    api_key="your-api-key"
)

# Local Docker
client = await EmbexClient.new_async(
    provider="qdrant",
    url="http://localhost:6333"
)

Node.js

// Qdrant Cloud
const client = await EmbexClient.newAsync(
  "qdrant",
  "https://xyz-example.eu-central.aws.cloud.qdrant.io:6333",
  "your-api-key"
);

// Local Docker
const client = await EmbexClient.newAsync("qdrant", "http://localhost:6333");

Other Providers

Elasticsearch

The Elasticsearch adapter is available in the source tree for the next release. Use the elasticsearch provider with the cluster’s HTTP(S) endpoint. Pass the encoded Elasticsearch API key through api_key in Python or as the third argument to newAsync in Node.js. Omit it for a local cluster with security disabled.

Python

import os

client = await EmbexClient.new_async(
    provider="elasticsearch",
    url=os.environ["ELASTICSEARCH_URL"],
    api_key=os.environ.get("ELASTICSEARCH_API_KEY"),
)

Node.js

const client = await EmbexClient.newAsync(
  "elasticsearch",
  process.env.ELASTICSEARCH_URL ?? "http://localhost:9200",
  process.env.ELASTICSEARCH_API_KEY,
);

Create collections through Embex so the index has the vector and metadata mappings. Collection names use lowercase ASCII letters, digits, hyphens, underscores and periods, starting with a letter or digit. Elasticsearch 8.19 and 9.x provide the kNN and point-in-time APIs used by the adapter. The adapter uses Elasticsearch relevance scores, exact string filters, and numeric range filters. Give each metadata key a consistent type across points. Numeric metadata is indexed as double precision for filtering. Bulk writes wait for a refresh, which requires the maintenance or manage index privilege on secured clusters. A migration cursor stays valid for one minute between pages.

Run the optional lifecycle integration test against a dedicated Elasticsearch instance with ELASTICSEARCH_URL and, when needed, ELASTICSEARCH_API_KEY set.

cargo test -p bridge-embex-elasticsearch --test elasticsearch real_elasticsearch_lifecycle -- --ignored

Milvus

Python

client = await EmbexClient.new_async(provider="milvus", url="http://localhost:19530")

Node.js

const client = await EmbexClient.newAsync("milvus", "http://localhost:19530");

Weaviate

Python

client = await EmbexClient.new_async(provider="weaviate", url="http://localhost:8080")

Node.js

const client = await EmbexClient.newAsync("weaviate", "http://localhost:8080");

Chroma

Python

client = await EmbexClient.new_async(provider="chroma", url="http://localhost:8000")

Node.js

const client = await EmbexClient.newAsync("chroma", "http://localhost:8000");

PgVector

Python

client = await EmbexClient.new_async(
    provider="pgvector",
    url="postgresql://user:pass@localhost:5432/db"
)

Node.js

const client = await EmbexClient.newAsync(
  "pgvector",
  "postgresql://user:pass@localhost:5432/db"
);

Supported Providers

Provider	Provider String	Example URL
Pinecone	"pinecone"	"https://index-name.svc.pinecone.io"
Qdrant	"qdrant"	"http://localhost:6333"
Milvus	"milvus"	"http://localhost:19530"
Weaviate	"weaviate"	"http://localhost:8080"
Chroma	"chroma"	"http://localhost:8000"
PgVector	"pgvector"	"postgresql://user:pass@localhost:5432/db"
LanceDB	"lancedb"	"./data"
Elasticsearch	"elasticsearch"	"http://localhost:9200"

The opensearch and redis adapters are experimental placeholders. Their vector operations currently return NotImplemented.