Skip to main content
This documentation describes the integration of MindsDB with LangChain, a framework for developing applications powered by language models. The integration allows for the deployment of LangChain models within MindsDB, providing the models with access to data from various data sources.

​
Prerequisites

Before proceeding, ensure the following prerequisites are met:
  1. Install MindsDB locally via Docker or Docker Desktop.
  2. To use LangChain within MindsDB, install the required dependencies following this instruction.
  3. Obtain the API key for a selected model provider that you want to use through LangChain.
Available models include the following:

​
Setup

Create an AI engine from the LangChain handler. 
CREATE ML_ENGINE langchain_engine
FROM langchain
USING
      serper_api_key = 'your-serper-api-key'; -- it is an optional parameter (if provided, the model will use serper.dev search to enhance the output)
Create a model using langchain_engine as an engine and a selected model provider. 
CREATE MODEL langchain_model
PREDICT target_column
USING
      engine = 'langchain_engine',           -- engine name as created via CREATE ML_ENGINE
      <provider>_api_key = 'api-key-value',  -- replace <provider> with one of the available values (openai, anthropic, google, litellm)
      model_name = 'model-name',             -- optional, model to be used (for example, 'gpt-4' if 'openai_api_key' provided)
      prompt_template = 'message to the model that may include some {{input}} columns as variables',
      max_tokens = 4096; -- defines the maximum number of tokens
This handler supports tracing features for LangChain via LangFuse. To use it, provide the following parameters in the USING clause:
  • langfuse_host,
  • langfuse_public_key,
  • langfuse_secret_key.
There are three different tools utilized by this agent:
  • MindsDB is the internal MindsDB executor.
  • Metadata fetches the metadata information for the available tables.
  • Write is able to write agent responses into a MindsDB data source.
Each tool exposes the internal MindsDB executor in a different way to perform its tasks, effectively enabling the agent model to read from (and potentially write to) data sources or models available in the active MindsDB project.
Create a conversational model using langchain_engine as an engine and a selected model provider. 
CREATE MODEL langchain_openai_model
PREDICT answer
USING
     engine = 'langchain_engine',       -- engine name as created via CREATE ML_ENGINE
     provider = 'openai',               -- one of the available providers
     openai_api_key = 'api-key-value',
     model_name = 'gpt-3.5-turbo',      -- choose one of the available OpenAI models
     mode = 'conversational',           -- conversational mode
     user_column = 'question',          -- column name that stores input from the user
     assistant_column = 'answer',       -- column name that stores output of the model (see PREDICT column)
     base_url = 'base-url-value',       -- optional, default is https://api.openai.com/v1/
     verbose = True,
     prompt_template = 'Answer the users input in a helpful way: {{question}}',
     max_tokens = 4096;

CREATE MODEL langchain_anthropic_model
PREDICT answer
USING
     engine = 'langchain_engine',          -- engine name as created via CREATE ML_ENGINE
     provider = 'anthropic',               -- one of the available providers
     anthropic_api_key = 'api-key-value',
     model_name = 'claude-2.1',            -- choose one of the available OpenAI models
     mode = 'conversational',              -- conversational mode
     user_column = 'question',             -- column name that stores input from the user
     assistant_column = 'answer',          -- column name that stores output of the model (see PREDICT column)
     verbose = True,
     prompt_template = 'Answer the users input in a helpful way: {{question}}',
     max_tokens = 4096;

CREATE MODEL langchain_ollama_model
PREDICT answer 
USING
     engine = 'langchain_engine',       -- engine name as created via CREATE ML_ENGINE
     provider = 'ollama',               -- one of the available providers
     model_name = 'llama2',             -- choose one of the models available from Ollama
     mode = 'conversational',           -- conversational mode
     user_column = 'question',          -- column name that stores input from the user
     assistant_column = 'answer',       -- column name that stores output of the model (see PREDICT column) 
     verbose = True,
     prompt_template = 'Answer the users input in a helpful way: {{question}}',
     max_tokens = 4096;
Ensure to have Ollama set up locally by following this guide on how to download Ollama.
CREATE MODEL langchain_litellm_model
PREDICT answer 
USING
     engine = 'langchain_engine',            -- engine name as created via CREATE ML_ENGINE
     provider = 'litellm',                   -- one of the available providers
     litellm_api_key = 'api-key-value',
     model_name = 'gpt-4',                   -- choose one of the models available from LiteLLM
     mode = 'conversational',                -- conversational mode
     user_column = 'question',               -- column name that stores input from the user
     assistant_column = 'answer',            -- column name that stores output of the model (see PREDICT column)
     base_url = 'https://ai.dev.mindsdb.com',
     verbose = True,
     prompt_template = 'Answer the users input in a helpful way: {{question}}',
     max_tokens = 4096;

CREATE MODEL langchain_google_model
PREDICT answer 
USING
     engine = 'langchain_engine',            -- engine name as created via CREATE ML_ENGINE
     provider = 'google',                    -- one of the available providers
     google_api_key = 'api-key-value',
     model_name = 'gemini-1.5-flash',        -- choose one of the models available from Google
     mode = 'conversational',                -- conversational mode
     user_column = 'question',               -- column name that stores input from the user
     assistant_column = 'answer',            -- column name that stores output of the model (see PREDICT column)
     verbose = True,
     prompt_template = 'Answer the users input in a helpful way: {{question}}',
     max_tokens = 4096;

CREATE MODEL langchain_mindsdb_model
PREDICT answer
USING
     engine = 'langchain_engine',       -- engine name as created via CREATE ML_ENGINE
     provider = 'mindsdb',              -- one of the available providers
     model_name = 'mindsdb_model',      -- any model created within MindsDB
     mode = 'conversational',           -- conversational mode
     user_column = 'question',          -- column name that stores input from the user
     assistant_column = 'answer',       -- column name that stores output of the model (see PREDICT column)
     verbose = True,
     prompt_template = 'Answer the users input in a helpful way: {{question}}',
     max_tokens = 4096;

​
Usage

The following usage examples utilize langchain_engine to create a model with the CREATE MODEL statement. Create a model that will be used to ask questions. 
CREATE ML_ENGINE langchain_engine_google
FROM langchain;

CREATE MODEL langchain_google_model
PREDICT answer 
USING
     engine = 'langchain_engine_google',
     provider = 'google',
     google_api_key = 'api-key-value',
     model_name = 'gemini-1.5-flash',
     mode = 'conversational',
     user_column = 'question',
     assistant_column = 'answer',
     verbose = True,
     prompt_template = 'Answer the users input in a helpful way: {{question}}',
     max_tokens = 4096;
Ask questions. 
SELECT question, answer
FROM langchain_google_model
WHERE question = 'How many planets are in the solar system?';
Next StepsGo to the Use Cases section to see more examples.