Databricks Integration (Deprecated)

0.1.2 · deprecated · verified Thu Apr 16

The `langchain-databricks` package was an integration library connecting LangChain with Databricks AI features, including LLMs, vector search, and MLflow. It has been superseded by `databricks-langchain` to consolidate Databricks-related LangChain components. The last published version of `langchain-databricks` is 0.1.2, and it is no longer actively maintained or receiving new features. Users should migrate to the `databricks-langchain` package for continued support and new functionality.

Common errors

```
ModuleNotFoundError: No module named 'langchain_databricks'
```
cause Attempting to import from the deprecated `langchain_databricks` package when it's not installed or when the new `databricks_langchain` package is expected.

fix Install the new package (`pip install databricks-langchain`) and update your import statements to use `from databricks_langchain import ...`.
```
MlflowException: Failed to save runnable sequence: ... ChatDatabricks -- 1 validation error for ChatDatabricks\nendpoint\n Field required
```
cause The `endpoint` parameter is missing when initializing `ChatDatabricks` or `DatabricksEmbeddings` during MLflow logging or direct use.

fix Ensure `endpoint` is provided, e.g., `ChatDatabricks(endpoint="your-model-serving-endpoint")`.
```
MlflowException: Failed to save runnable sequence: ... trying to import ChatDatabricks from the legacy module path “langchain_databricks”
```
cause MLflow is attempting to serialize a `ChatDatabricks` object that still resolves to the deprecated `langchain_databricks` module, even if your current code uses `databricks_langchain` imports. This often happens if the object wasn't fully re-instantiated after migration.

fix Rebuild your LangChain chain components to ensure all instances of Databricks classes (like `ChatDatabricks`) are explicitly re-imported and re-instantiated from `databricks_langchain`. You might also need to include `langchain-databricks` in `pip_requirements` during `mlflow.langchain.log_model` as a temporary shim if full migration isn't immediate.
```
str type expected (type=type_error.str)
```
cause The input to the Databricks LLM endpoint is not in the expected string format, or the `transform_input_fn` is misconfigured. Some endpoints expect a dictionary or specific JSON structure, not just a raw string.

fix Verify the expected input format for your specific Databricks Model Serving endpoint. Ensure your LangChain `invoke` call provides the prompt in the correct format (e.g., as a `HumanMessage` in a list for chat models). If custom transformation is used, check `transform_input_fn` parameters.

Warnings

breaking The `langchain-databricks` package is deprecated. All features have been consolidated into the new `databricks-langchain` package. Future updates and new features will be released exclusively in `databricks-langchain`.
Fix: Migrate your `pip install` command to `pip install databricks-langchain` and update all imports from `langchain_databricks` to `databricks_langchain`.
gotcha When running outside a Databricks workspace, you must explicitly set Databricks authentication credentials (hostname and personal access token).
Fix: Set `DATABRICKS_HOST` and `DATABRICKS_TOKEN` environment variables or pass `databricks_host` and `databricks_token` directly during class instantiation (e.g., `ChatDatabricks(endpoint=..., databricks_host=..., databricks_token=...)`).
gotcha Model serving endpoint name mismatches or incorrect API key/configuration are common causes for invocation failures.
Fix: Double-check that the `endpoint` name passed to `ChatDatabricks` or `DatabricksEmbeddings` exactly matches the configured endpoint name in your Databricks workspace. Verify associated API keys and resource configurations.
gotcha When logging LangChain models with MLflow, ensure the model name adheres to the three-level Unity Catalog path (`<catalog>.<schema>.<model>`) and that `chain_type` is passed in `model_config` for proper reconstruction.
Fix: Use `mlflow.langchain.log_model(lc_model=..., registered_model_name='your_catalog.your_schema.your_model', model_config={'chain_type': 'stuff'})`.

Install

pip install langchain-databricks Install Deprecated Package
pip install databricks-langchain Install Recommended Replacement

Imports

ChatDatabricks
wrong:
```
from langchain_databricks import ChatDatabricks
```
correct:
```
from databricks_langchain import ChatDatabricks
```
The original import path for ChatDatabricks is deprecated. Use the new `databricks_langchain` package.
DatabricksEmbeddings
wrong:
```
from langchain_databricks import DatabricksEmbeddings
```
correct:
```
from databricks_langchain import DatabricksEmbeddings
```
The original import path for DatabricksEmbeddings is deprecated. Use the new `databricks_langchain` package.
DatabricksVectorSearch
wrong:
```
from langchain_databricks import DatabricksVectorSearch
```
correct:
```
from databricks_langchain import DatabricksVectorSearch
```
The original import path for DatabricksVectorSearch is deprecated. Use the new `databricks_langchain` package.

Quickstart

This quickstart demonstrates how to instantiate and use `ChatDatabricks` from the *recommended* `databricks-langchain` package to interact with a Databricks Model Serving endpoint. If running outside a Databricks Workspace, ensure `DATABRICKS_HOST` and `DATABRICKS_TOKEN` environment variables are set for authentication. The code for the deprecated `langchain-databricks` is shown for historical context but should not be used in new projects.

import os
from databricks_langchain import ChatDatabricks
from langchain_core.messages import HumanMessage

# --- IMPORTANT: FOR RUNNING OUTSIDE DATABRICKS WORKSPACE ---
# Set these environment variables or pass them directly to the ChatDatabricks constructor.
# os.environ["DATABRICKS_HOST"] = os.environ.get("DATABRICKS_HOST", "https://your-databricks-workspace.cloud.databricks.com")
# os.environ["DATABRICKS_TOKEN"] = os.environ.get("DATABRICKS_TOKEN", "dapi********************************")

# Instantiate the chat model using a Databricks Model Serving endpoint
# Replace 'databricks-meta-llama-3-70b-instruct' with your actual endpoint name.
# Ensure DATABRICKS_HOST and DATABRICKS_TOKEN are set if running outside Databricks workspace.
chat_model = ChatDatabricks(
    endpoint="databricks-meta-llama-3-70b-instruct",
    temperature=0.1,
    max_tokens=256
)

# Invoke the model
response = chat_model.invoke([HumanMessage(content="What is MLflow?")])
print(response.content)

# --- DEPRECATED PACKAGE USAGE (for context, prefer databricks-langchain) ---
# from langchain_databricks import ChatDatabricks
# deprecated_chat_model = ChatDatabricks(endpoint="your-deprecated-endpoint")
# print("WARNING: Using deprecated langchain-databricks package.")

view raw JSON →