Complete list of model-related configurations for setting up the Cloudera Data Explorer (Hue) SQL AI Assistant

Review the list of service, model, and semantic search-related configurations used for custom configuring the AI services and models you want to use with the SQL AI Assistant and how to specify them in the Hue Advanced Configuration Snippet in the Cloudera Manager.

List of service and model-related configurations

You can configure the AI services and models you want to use by going to Cloudera Manager > Clusters > Data Explorer (Hue) service > Configurations > Hue Service Advanced Configuration Snippet (Safety Valve) for hue_safety_valve.ini and adding the following lines:

[desktop]
[[ai_interface]]
[***CONFIG-KEY1***]='[***VALUE***]'
[***CONFIG-KEY2***]='[***VALUE***]'
[[semantic_search]]
[***CONFIG-KEY1***]='[***VALUE***]'
[***CONFIG-KEY2***]='[***VALUE***]'

AI interface-related configurations

Here is the complete list of configurations under [[ai_interface]], which allows you to specify the service and model to be used:


AI interface config key	Description
service	API service to be used for AI tasks. AI is disabled when a service is not configured. For example, the Cloudera AI Workbench and Cloudera AI Inference service are API services.
service_version	API service version to be used for AI tasks.
trusted_service	Indicates whether the LLM is trusted or not. Turn on to disable the warning. The default value is `True`.
model	The AI model you want to use for AI tasks. For example, `gpt` and `llama`.
model_name	The fully qualified name of the model to be used. For example, `gpt-3.5-turbo-16k`.
model_ref	The `model_ref` is a placeholder for adding the access key of the specific model you want to use.
base_url	Service API base URL.
add_table_data	When enabled, sample rows from the table are added to the prompt. The default value is `True`.
table_data_cache_size	Size of the LRU cache used for storing table sample data.
auto_fetch_table_meta_limit	Number of tables to load from a database, initially.
token	Service API secret token.
token_script	Provides a secure way to get the service API secret token.
enabled_sql_tasks	A comma-separated list of SQL-related AI tasks available in the Editor.

User Input Validation for Data Explorer SQL AI

Following is the complete list of configurations under [[ai_interface]]. It helps to specify the input validation to enhance security and optimize performance.

AI interface config key	Description
user_input_max_length	Ensure the configured user input length is not exceeded. The default limit is 1000, but you can configure it to a higher value if needed.
user_input_remove_characters	Remove specific characters from user input, such as newlines (\n), tab spaces (\t), and others, to ensure clean and consistent formatting.
user_input_banned_keyphrases	Block user input if certain configured keyphrases are found.
user_input_banned_regex	Block user input if a configured regex pattern match is found.
user_input_block_html	Escape HTML tags to prevent malicious activities and ensure secure input handling. This config accepts a boolean value: `True` to escape HTML tags or `False` to allow raw HTML. The default value is set to `False`.

The following sample configuration sets the validations for user input:

[[ai_interface]]
          service='azure'
          model_name='[***DEPLOYMENT-NAME***]'
          base_url='https://[***RESOURCE***].cloudera.site/model'
          token='[***RESOURCE-KEY***]'
          user_input_max_length=1000
          user_input_remove_characters="&\n\r\t"
          user_input_banned_keyphrases=""
          user_input_banned_regex=""
          user_input_block_html="False"

Personally Identifiable Information PII entity detection configuration

The pii_entities_to_detect configuration property sets up Personally Identifiable Information (PII) filtering for SQL AI queries.

The following table describes the configuration property:

Table 1. SQL AI Assistant Properties
Property	Description	Type	Default
`pii_entities_to_detect`	Specifies a comma-separated list of Personally Identifiable Information (PII) entity types to detect in SQL AI user input. When one of the listed entity types is detected with sufficient confidence, the request is rejected to prevent sensitive data from being sent to the model.	Comma-separated list	Empty (PII detection disabled)

The following example shows how to configure the pii_entities_to_detect configuration property in the hue-safety-valve configuration file:


pii_entities_to_detect=CREDIT_CARD,EMAIL_ADDRESS,PHONE_NUMBER,US_SSN

Row cap limit for SQL AI queries

For queries produced by the SQL AI Assistant generate task, the SQL Editor user interface automatically appends a default LIMIT clause in the preview before running the query to match the standard row cap. The default setting is LIMIT 1000, which functions as a fallback only when the SQL Editor snippet does not define a default_limit configuration property.

The system appends the limit only when the generated query is a read query using a SELECT or WITH Common Table Expression clause that includes a FROM clause and does not contain an existing LIMIT clause. For multi-statement queries, the system applies the limit to the final statement to run.

Semantic search-related configurations

Specify the semantic search-related configurations used for RAG under the [[semantic_search]] section, as listed in the following table:


Semantic search config key	Description
relevancy	The technology you want to use for semantic search. Acceptable values are `vector_search` or `vector_db`.
embedding_model	The model you want to use for data-embedding. This must be compatible with SentenceTransformer.
top_k	Number of top-ranking items returned by semantic search.
cache_size	Size of the LRU cache used for storing embedding.