Establishing and Getting Began with Cloudera’s New SQL AI Assistant

As described in our latest weblog put up, an SQL AI Assistant has been built-in into Hue with the potential to leverage the ability of huge language fashions (LLMs) for quite a lot of SQL duties. It might aid you to create, edit, optimize, repair, and succinctly summarize queries utilizing pure language. This can be a actual game-changer for information analysts on all ranges and can make SQL improvement sooner, simpler, and fewer error-prone. 

This weblog put up goals that will help you perceive what you are able to do to get began with generative AI assisted SQL utilizing Hue picture model ​​2023.0.16.0 or greater on the general public cloud. Each Hive and Impala dialects are supported. Please confer with the product documentation for extra details about particular releases.

Getting began with the SQL AI Assistant

Later on this weblog we’ll stroll you thru the steps of tips on how to configure your Cloudera surroundings to make use of the SQL AI Assistant along with your supported LLM of selection. However first, let’s discover what the SQL AI Assistant does, and the way individuals would use it inside the SQL editor.

Utilizing the SQL AI Assistant

To launch the SQL AI Assistant, begin the SQL editor in Hue and click on the blue dot as proven within the following picture. This may increase the SQL AI toolbar with buttons to generate, edit, clarify, optimize and repair SQL statements. The assistant will use the identical database because the editor, which within the picture under is about to a DB named tpcds_10_text. 

The toolbar is context conscious and completely different actions shall be enabled relying on what you’re doing within the editor. When the editor is empty, the one possibility accessible is to generate new SQL from pure language.

Click on “generate” and sort your question in pure language. Within the edit discipline, press the down arrow to see a historical past of question prompts. Click on “enter” to generate the SQL question.

The generated SQL is offered in a modal along with the assumptions made by the LLM. This could embody assumptions concerning the intent of the pure language used, just like the definition of “prime promoting merchandise,” values of wanted literals, and the way joins might be created. Now, you’ll be able to insert the SQL straight into the editor or copy it to the clipboard.

When there may be an lively SQL assertion within the editor the SQL AI Assistant will allow the “edit,” “clarify,” and “optimize” buttons. The “repair” button will solely be enabled when the editor finds an error, equivalent to a SQL syntax error or a misspelled identify.

Click on “edit” to change the lively SQL assertion. If the assertion is preceded by a NQL-comment then that immediate might be reused by urgent tab. You too can simply begin typing a brand new instruction.

After utilizing edit, optimize, or repair, a preview reveals the unique question and the modified question variations. If the unique question has a unique formatting or key phrase higher/decrease case than the generated question, you’ll be able to allow “Autoformat SQL” on the prime of the modal for a greater end result. 

Click on “insert” to switch the unique question with the modified one within the editor.

The optimize and the repair performance don’t want person enter. To make use of them merely choose a SQL assertion within the editor, and click on “optimize” or “repair”  to generate an improved model displayed as a diff of the unique question, as proven above. “Optimize” will attempt to enhance the construction and efficiency with out impacting the returned results of working the question. “Repair” will attempt to routinely repair syntactic errors and misspelling.   

When you need assistance making sense of advanced SQL then merely choose the assertion, and click on “clarify.” A abstract and rationalization of the SQL in pure language will seem. You possibly can select to insert the textual content as a remark above the SQL assertion within the editor as proven under.

The SQL AI Assistant will not be bundled with a particular LLM; as an alternative it helps numerous LLMs and internet hosting providers. The mannequin can run domestically, be hosted on CML infra or within the infrastructure of a trusted service supplier. Cloudera has been testing with GPT working in each Azure and OpenAI, however the next service-model combos are additionally supported:

Observe: Cloudera recommends utilizing the Hue AI assistant with the Azure OpenAI service.

The supported AI fashions are pre-trained on pure language and SQL however they don’t have any data of your group’s information. To beat this the SQL AI Assistant makes use of a Retrieval Augmented Technology (RAG)-based structure the place the suitable data is retrieved for every particular person SQL activity (immediate) and used to enhance the request to the LLM. Throughout the retrieval course of it makes use of the Python SentenceTransformers framework for semantic search, which by default makes use of the all-MiniLM-L6-v2 mannequin. The SQL AI Assistant might be configured with many pre-trained fashions for higher multi-lingual assist. Under are the fashions examined by Cloudera:

It is very important perceive that by utilizing the SQL AI Assistant you’re sending your individual prompts and in addition vital further data as enter to the LLM. The SQL AI Assistant will solely share information that the at present logged-in person is allowed to entry, however it’s of utmost significance that you simply use a service that you may belief along with your information. The RAG-based structure reduces the variety of tables despatched per request to a brief listing of the more than likely wanted, however there may be at present no solution to explicitly exclude sure tables; consequently, data about all tables that the logged-in person can entry within the database may very well be shared. The listing under particulars precisely what’s shared:


  • Every part {that a} person inputs within the SQL AI Assistant
  • The chosen SQL assertion (if any) within the Hue editor
  • SQL dialect in use (Hive, Impala for instance)
  • Desk particulars equivalent to desk identify, column names, column information sorts and associated keys, partitions and constraints
  • Three pattern rows from the tables (following one of the best practices laid out in Rajkumar et al, 2022)

The administrator should get hold of clearance out of your group’s infosec staff to ensure it’s secure to make use of the SQL AI Assistant as a result of a few of the desk metadata and information, as talked about within the earlier part, is shared with the LLM.

Getting began with the SQL AI Assistant is a simple course of. First prepare entry to one of many supported providers after which add the service particulars in Hue’s configuration.

Utilizing Microsoft Azure OpenAI service

Microsoft Azure gives the choice to have devoted deployments of OpenAI GPT fashions. Azure’s OpenAI service is way more safe than the publicly hosted OpenAI APIs as a result of the info might be processed in your digital non-public cloud (VPC). Contemplating the added safety, Azure’s OpenAI is the really helpful service to make use of for GPT fashions within the SQL AI Assistant. For extra data, see the Azure OpenAI fast begin information.

Step 1. Azure subscription

First, get Azure entry. Contact your IT division to get an Azure subscription. Subscriptions may very well be completely different based mostly in your staff and objective. For extra data, see subscription issues.


2. Azure Open AI entry

At the moment, entry to this service is granted solely by utility. You possibly can apply for entry to Azure OpenAI by finishing the shape at As soon as accredited, it’s best to obtain a welcome e mail. 

3. Create useful resource

Within the Azure portal, create your Azure OpenAI useful resource:

Within the useful resource particulars web page, underneath “Develop”, you may get your useful resource URL and keys. You simply want any one of many two supplied keys.

4. Deploy GPT

Go to Azure OpenAI Studio at and create your deployment underneath administration > Deployments. Choose gpt-35-turbo-16k or greater.

5. Configure SQL AI Assistant in Hue

Now that the service is up and working along with your mannequin, the final step is to allow and configure the SQL AI assistant in Hue.

  1. Log in to the Cloudera Knowledge Warehouse service as DWAdmin.
  2. Go to the digital warehouse tab, find the Digital Warehouse on which you need to allow this characteristic, and click on “edit.”
  3. Go to “configurations” > Hue and choose “hue-safety-valve” from the configuration information drop-down menu.

Edit the textual content underneath the desktop part by including a subsection referred to as ai_interface. Populate it as proven under by changing the angle bracket values with these from your individual service:

Utilizing OpenAI service

1. Open AI platform enroll

Request entry to the Open AI platform out of your IT division or go to and create an account if allowed by your organization’s insurance policies.

2. Get the API key

Within the left menu bar, navigate to AI keys. You need to have the ability to view present keys or create new ones. The API key’s the one factor you might want to combine with the SQL AI Assistant.

3. Configure SQL AI Assistant in Hue

Lastly, allow and configure the SQL AI assistant in Hue.

  1. Log in to the info warehouse service as DWAdmin.
  2. Go to the digital warehouse tab, find the Digital Warehouse on which you need to allow this characteristic, and click on “edit.”
  3. Go to “configurations” > Hue and choose “hue-safety-valve from the configuration information drop-down menu. 
  4. Edit the textual content underneath the desktop part by including a subsection referred to as ai_interface. Solely two key worth pairs are wanted as proven under. Exchange the <api-key> worth with the API key from Open AI.

Amazon Bedrock Service

Amazon Bedrock is a totally managed service that makes basis fashions from main AI startups and Amazon accessible through an API. It’s essential to have an AWS account with Bedrock entry earlier than following these steps.

  1. Get your entry key and secret

Get the entry key ID and the key entry key for utilizing Bedrock-hosted fashions in Hue Assistant:

  1. Go to IAM console: 
  2. Click on “customers” within the left menu
  3. Discover the person who wants entry
  4. Click on “safety credentials”
  5. Go to the “entry keys” part and discover your keys there.

2. Get Anthropic Claude entry

Claude from Anthropic is among the finest fashions accessible in Bedrock for SQL-related duties. Extra particulars can be found at Upon getting entry, it is possible for you to to strive Claude within the textual content playground underneath the Amazon Bedrock service.

3. Configure SQL AI Assistant in Hue

Lastly, allow and configure the SQL AI assistant in Hue.


  1. Log in to the info warehouse service as DWAdmin.
  2. Go to the digital warehouse tab, find the digital warehouse on which you need to allow this characteristic, and click on “edit.”
  3. Go to “configurations: > Hue and choose “hue-safety-valve” from the configuration information drop-down menu.
  4. Edit the textual content to ensure the next sections, subsections and key worth pairs are set. Exchange the <access_key> and the <secret_key> with the values out of your AWS account.

Service- and model-related configurations are underneath ai_interface, and semantic search associated configurations used for RAG are underneath the semantic_search part.

The configurable LLMs are excellent at producing and modifying SQL. The RAG structure gives the correct context. However there isn’t any assure strategies from LLMs, or from human consultants, are at all times correct. Please pay attention to the next:

  • Non-deterministic: LLMs are non-deterministic. You can not assure the very same output for a similar enter each time, and completely different responses for very comparable queries can happen.
  • Ambiguity: LLMs could battle to deal with ambiguous queries or contexts. SQL queries typically depend on particular and unambiguous language, however LLMs can misread or generate ambiguous SQL queries, resulting in incorrect outcomes.
  • Hallucination: Within the context of LLMs, hallucination refers to a phenomenon the place these fashions generate responses which might be incorrect, nonsensical, or fabricated. Sometimes you may see incorrect identifiers or literals, and even desk and column names, if the supplied context is incomplete or person enter merely doesn’t match any information. 
  • Partial context: The RAG structure gives context to every request however it has limitations and there’s no assure the context despatched to the LLM will at all times be full.

The SQL AI Assistant is now accessible in tech preview on Cloudera Knowledge Warehouse on Public Cloud. We encourage you to strive it out and expertise the advantages it could possibly present in relation to working with SQL. Moreover, take a look at the overview weblog on SQL AI Assistant to be taught the way it might help information and enterprise analysts in your group velocity up information analytics. Take a look at the SQL AI Assistant documentation Attain out to your Cloudera staff for extra particulars.

Recent Articles

Related Stories

Leave A Reply

Please enter your comment!
Please enter your name here