Skip to main content
MindsDB provides the LLM() function that lets users incorporate the LLM-generated output directly into the data queries.

​
Prerequisites

The LLM() function requires a large language model, which can be defined in the following ways:
  • By setting the default_llm parameter in the MindsDB configuration file.
  • By saving the default model in the MindsDB Editor under Settings.
  • By defining the environment variables as below, choosing one of the available model providers.
    Here are the environment variables for the OpenAI provider:
    LLM_FUNCTION_MODEL_NAME
LLM_FUNCTION_TEMPERATURE
LLM_FUNCTION_MAX_RETRIES
LLM_FUNCTION_MAX_TOKENS
LLM_FUNCTION_BASE_URL
OPENAI_API_KEY
LLM_FUNCTION_API_ORGANIZATION
LLM_FUNCTION_REQUEST_TIMEOUT
    Note that the values stored in the environment variables are specific for each provider.
    Here are the environment variables for the Anthropic provider:
    LLM_FUNCTION_MODEL_NAME
LLM_FUNCTION_TEMPERATURE
LLM_FUNCTION_MAX_TOKENS
LLM_FUNCTION_TOP_P
LLM_FUNCTION_TOP_K
LLM_FUNCTION_DEFAULT_REQUEST_TIMEOUT
LLM_FUNCTION_API_KEY
LLM_FUNCTION_BASE_URL
    Note that the values stored in the environment variables are specific for each provider.
    Here are the environment variables for the LiteLLM provider:
    LLM_FUNCTION_MODEL_NAME
LLM_FUNCTION_TEMPERATURE
LLM_FUNCTION_API_BASE
LLM_FUNCTION_MAX_RETRIES
LLM_FUNCTION_MAX_TOKENS
LLM_FUNCTION_TOP_P
LLM_FUNCTION_TOP_K
    Note that the values stored in the environment variables are specific for each provider.
    Here are the environment variables for the Ollama provider:
    LLM_FUNCTION_BASE_URL
LLM_FUNCTION_MODEL_NAME
LLM_FUNCTION_TEMPERATURE
LLM_FUNCTION_TOP_P
LLM_FUNCTION_TOP_K
LLM_FUNCTION_REQUEST_TIMEOUT
LLM_FUNCTION_FORMAT
LLM_FUNCTION_HEADERS
LLM_FUNCTION_NUM_PREDICT
LLM_FUNCTION_NUM_CTX
LLM_FUNCTION_NUM_GPU
LLM_FUNCTION_REPEAT_PENALTY
LLM_FUNCTION_STOP
LLM_FUNCTION_TEMPLATE
    Note that the values stored in the environment variables are specific for each provider.
    Here are the environment variables for the Nvidia NIMs provider:
    LLM_FUNCTION_BASE_URL
LLM_FUNCTION_MODEL_NAME
LLM_FUNCTION_TEMPERATURE
LLM_FUNCTION_TOP_P
LLM_FUNCTION_REQUEST_TIMEOUT
LLM_FUNCTION_FORMAT
LLM_FUNCTION_HEADERS
LLM_FUNCTION_NUM_PREDICT
LLM_FUNCTION_NUM_CTX
LLM_FUNCTION_NUM_GPU
LLM_FUNCTION_REPEAT_PENALTY
LLM_FUNCTION_STOP
LLM_FUNCTION_TEMPLATE
LLM_FUNCTION_NVIDIA_API_KEY
    Note that the values stored in the environment variables are specific for each provider.
OpenAI-compatible model providers can be used like OpenAI models.There is a number of OpenAI-compatible model providers including OpenRouter or vLLM. To use models via these providers, users need to define the base URL and the API key of the provider.Here is an example of using OpenRouter.
LLM_FUNCTION_MODEL_NAME = "mistralai/devstral-small-2505"
LLM_FUNCTION_BASE_URL = "https://openrouter.ai/api/v1"
OPENAI_API_KEY = "openrouter-api-key"

​
Usage

You can use the LLM() function to simply ask a question and get an answer. 
SELECT LLM('How many planets are there in the solar system?');
Here is the output: 
+------------------------------------------+
| llm                                      |
+------------------------------------------+
| There are 8 planets in the solar system. |
+------------------------------------------+
Moreover, you can use the LLM() function with your data to swiftly complete tasks such as text generation or summarization. 
SELECT
    comment,
    LLM('Describe the comment''s category in one word: ' || comment) AS category
FROM example_db.user_comments;
Here is the output: 
+--------------------------+----------+
| comment                  | category |
+--------------------------+----------+
| I hate tacos             | Dislike  |
| I want to dance          | Desire   |
| Baking is not a big deal | Opinion  |
+--------------------------+----------+
⌘I