What is DatabricksEmbeddings?
DatabricksEmbeddings is the LangChain wrapper for embedding models served by Databricks Model Serving and the Mosaic AI Foundation Model APIs. It lets you call a model you have deployed as a serving endpoint inside your own Databricks workspace — for example bge-large-en or a fine-tuned embedding model — rather than sending text to an external provider. For teams whose data already lives in a Databricks Lakehouse, this keeps embeddings inside the same governance, billing, and network boundary.
You point the class at a serving endpoint by name through the endpoint argument, and authenticate with your workspace host and a personal access token, usually supplied through the DATABRICKS_HOST and DATABRICKS_TOKEN environment variables. On a Databricks cluster or notebook those are often injected automatically. The class exposes the standard embed_query and embed_documents methods, so it drops into any LangChain retriever or vector store unchanged — it pairs naturally with Databricks Vector Search for an end-to-end in-platform RAG stack.
The package moved over time: prefer the databricks-langchain package, which supersedes the older langchain-databricks and the even older DatabricksEmbeddings shim that lived in the community package. Because it targets a serving endpoint, throughput and latency depend on how that endpoint is provisioned — scale-to-zero endpoints add cold-start latency on the first call. As with every embeddings model, index and query with the same endpoint; switching the underlying model invalidates existing vectors.
When to Use
You're on Databricks platform. Use for unified data+AI infrastructure.
Use Cases
- • Databricks ecosystem
- • Data + AI integration
- • Enterprise ML
- • Lakehouse
- • Unified data
- • Cloud ML
Key Features
- ✓ Databricks-native
- ✓ Unified ecosystem
- ✓ Enterprise
- ✓ Multi-workspace
- ✓ Integrated
- ✓ Optimized
When NOT to Use
Unless you're using Databricks.
Notes
Use the databricks-langchain package
Install and import from databricks-langchain. It supersedes the older langchain-databricks package and the community DatabricksEmbeddings shim, which are deprecated. Mixing import paths across a project is a common source of confusing version-skew errors.
Endpoint, not model name
DatabricksEmbeddings targets a serving endpoint you deployed, addressed by the endpoint argument. The model lives behind that endpoint. A 404 or RESOURCE_DOES_NOT_EXIST error almost always means the endpoint name is wrong or lives in a different workspace than your host points to.
Cold starts on scale-to-zero endpoints
If the serving endpoint is configured to scale to zero, the first request after idle pays a cold-start delay while the model spins up. For latency-sensitive apps, provision a minimum of one running instance or warm the endpoint before serving traffic.
Auth is host plus token
Set DATABRICKS_HOST and DATABRICKS_TOKEN, or pass them explicitly. Inside a Databricks notebook or job these are often injected automatically; outside the platform you must supply a valid personal access token with permission to query the endpoint.
Import
from langchain_databricks import DatabricksEmbeddingsConfiguration
| Parameter | Type | Default | Purpose |
|---|---|---|---|
| host | str | None | Databricks workspace URL |
Usage Examples
Call a serving endpoint
from databricks_langchain import DatabricksEmbeddings
embeddings = DatabricksEmbeddings(endpoint='databricks-bge-large-en')
vector = embeddings.embed_query('What is a lakehouse?')
print(len(vector))Authenticate with host and token
import os
from databricks_langchain import DatabricksEmbeddings
os.environ['DATABRICKS_HOST'] = 'https://my-workspace.cloud.databricks.com'
os.environ['DATABRICKS_TOKEN'] = 'dapi...'
embeddings = DatabricksEmbeddings(endpoint='my-embedding-endpoint')
vectors = embeddings.embed_documents(['doc one', 'doc two'])In-platform RAG with Databricks Vector Search
from databricks_langchain import DatabricksEmbeddings, DatabricksVectorSearch
embeddings = DatabricksEmbeddings(endpoint='databricks-bge-large-en')
store = DatabricksVectorSearch(
index_name='catalog.schema.docs_index',
embedding=embeddings,
)
hits = store.similarity_search('reset my password', k=4)Common Pitfalls
❌ Use without Databricks workspace
✅ Requires Databricks setup
Alternative Embedding Models
| Model | When to Use |
|---|---|
| OpenAIEmbeddings | For standalone use |
Related LangChain References
Browse the full LangChain API reference index to explore more classes, methods, and decorators, or start with the LangChain introduction tutorial for end-to-end context on building with DatabricksEmbeddings and the wider framework.
DatabricksEmbeddings FAQ
What is DatabricksEmbeddings in LangChain?
Use Databricks-hosted embeddings. DatabricksEmbeddings is the LangChain wrapper for embedding models served by Databricks Model Serving and the Mosaic AI Foundation Model APIs. It lets you call a model you have deployed as a serving endpoint inside your own Databricks workspace — for example bge-large-en or a fine-tuned embedding model — rather than sending text to an external provider. For teams whose data already lives in a Databricks Lakehouse, this keeps embeddings inside the same governance, billing, and…
Which package provides DatabricksEmbeddings?
DevShelfHub documents DatabricksEmbeddings from the langchain-databricks package. Pin your installed LangChain version and match imports to the snippet on this page.
When should I use DatabricksEmbeddings?
You're on Databricks platform. Use for unified data+AI infrastructure.
When should I avoid using DatabricksEmbeddings?
Unless you're using Databricks.
How do I import DatabricksEmbeddings in Python?
from langchain_databricks import DatabricksEmbeddings
Where can I explore more LangChain API reference pages?
Open the LangChain API reference index on DevShelfHub to browse classes, methods, and decorators, each with runnable examples, parameters, common mistakes, and cross-links.