What you will be able to do
- Create a Genie space and name the permissions you need to build one
- Pick a small, well-documented set of Unity Catalog tables, views and metric views for a space
- Choose a suitable SQL warehouse and explain what its embedded credentials do and do not grant
- Set up the title, description and common (sample) questions that users see when they open the space
Key concept
Genie space as a curated context — A Genie space is a natural-language chat interface over a small set of Unity Catalog data. You build it by giving Genie the context it is missing: which data to use, how your business talks about that data, and verified SQL it can rely on. Every choice you make while building the space either adds to that context or narrows it.
1.What a Genie space is and how you create one
A Genie space lets business users ask questions about their data in plain language. Genie turns each question into SQL, runs it, and returns the results. Business users only use the chat. A data analyst sets up everything behind it: the datasets, the SQL warehouse, the instructions, and the example queries. A note on names: the current Databricks documentation calls these "Genie Agents" and says they were formerly known as Genie Spaces. The exam guide uses "Genie space", so this lesson does too. Treat the two names as the same thing.
Before you can create a space, you need four things. You need the Databricks SQL workspace entitlement. You need CAN USE on at least one pro or serverless SQL warehouse. You need SELECT on the data the space will use. To edit an existing space, you need at least CAN EDIT on it. Whoever creates a space gets CAN MANAGE on it automatically. Granting warehouse and data access usually takes an administrator, so arrange those grants before you start building.
The creation flow itself is short. Click Genie in the sidebar, click New in the upper-right corner, choose your data sources, and click Create. Genie Code then starts automatically. It reads your data and suggests context, such as table descriptions and example queries, which you can accept or reject. You can also describe your business domain and let Genie Code find the right tables for you. Either way, Databricks recommends adding example SQL queries and instructions before you share the space.
Checkpoint 1 of 5· Put it in order
Put the steps for creating a new Genie space in order.
- 1.Choose the data sources to include, then click Create
- 2.Click Genie in the sidebar
- 3.Click New in the upper-right corner of the screen
- 4.Review the context Genie Code suggests and accept what you want to keep
You start the space from the sidebar, add the data, and click Create. Genie Code launches only after the space exists and suggests context based on the data you chose.
“Choose the data sources that you want to include in your Genie Agent. Then, click Create.”Source: docs.databricks.com
Sources1
2.Curating Unity Catalog datasets
All data in a Genie space must be registered in Unity Catalog. That includes managed, external and foreign tables, views, metric views, and materialized views. You can add up to 50 tables, views or metric views. The limit is not the goal, though. Databricks recommends a small, focused set of tables, built around one topic and one audience, with only the columns that matter. If a topic needs many tables, pre-join them into views or metric views. Metric views work especially well because they already define the metrics, dimensions and aggregations.
Genie writes its SQL using Unity Catalog names and descriptions, so clear column comments matter a lot. Check any AI-generated descriptions before you keep them. Inside a space, you can also add your own column descriptions and synonyms, and hide columns that might confuse Genie. These changes apply only to that space and do not overwrite the metadata in Unity Catalog. Databricks' best-practices guide also says to fix data problems in the data itself. Remove rows or columns that should never be queried during data engineering, rather than relying on instructions to work around them.
Two more points. First, Genie queries only the sources you have attached to the space. If an instruction mentions a table that is not attached, Genie will not query it. Second, attaching a table does not change who can see its data: each user's own Unity Catalog permissions still apply. To change the data in a space later, open Configure > Data and add or remove tables. Click any table to see its columns and sample rows.
Checkpoint 2 of 5· Check yourself
An author writes a text instruction: "For returns, use retail.ops.returns_v." That view is not one of the space's data sources. What happens when a user asks about returns?
Only attached sources are used. Naming a table in instructions or metadata does not attach it.
“Genie does not query tables or Unity Catalog functions that are not attached, even if the agent's instructions or metadata reference them.”Source: docs.databricks.com
3.Choosing the SQL warehouse
Every SQL statement Genie generates runs on the space's default warehouse. Generated queries are always read-only. Only pro and serverless SQL warehouses are supported, and Databricks recommends serverless for the best performance. You set the warehouse under Configure > Settings, and changing it requires at least CAN EDIT on the space.
When you save the warehouse choice, your compute credentials are embedded in the space. This is how every user can run queries on that warehouse without needing their own grant on it. These credentials cover compute only. Data access is still checked against each user's own Unity Catalog permissions. If a different author changes the warehouse later, that author's credentials replace yours.
| Question | Answer |
|---|---|
| Which warehouse types can a space use? | Pro or serverless SQL warehouse; serverless recommended |
| What permission does the author need on the warehouse? | At least CAN USE |
| What do the embedded credentials grant users? | Access to the warehouse only |
| Whose permissions control data access? | Each end user's own Unity Catalog permissions |
| What happens if another author changes the warehouse? | That author's compute credentials are embedded instead |
Checkpoint 3 of 5· Check yourself
You built a space on a serverless warehouse and shared it with an analyst. The analyst has no SELECT on one of the space's tables. What happens when they ask about that table?
The embedded credentials let the analyst use the warehouse. They do not give the analyst access to data the analyst cannot otherwise read.
“These embedded credentials grant access to the warehouse only.”Source: docs.databricks.com
4.Title, description and sample questions
Configure > Settings also controls what users see when they open the space. The title shows in the workspace browser, so choose one that helps people find the space. The description appears when the space opens and supports Markdown, so you can explain what the space is for and link to more information. Tags help you organize spaces, and a thumbnail image appears on the chat landing page.
Common questions are the sample questions shown on the chat landing page. They are optional, and you can add them when you create the space or later. They tell users what kinds of questions the space can answer. Your own questions always come first. If you provide enough to fill the landing page, only yours appear. If you provide fewer, Genie fills the rest with questions it generates. Good sample questions match the space's purpose and the data you actually attached.
Checkpoint 4 of 5· Check yourself
You add two common questions, and the landing page has room for more. What do users see?
Your questions take priority and appear first. Genie fills any empty slots with questions it generates.
“When you provide fewer, Genie fills the remaining slots with auto-generated questions.”Source: docs.databricks.com
Checkpoint 5 of 5· Exam question
A data platform team is setting up a new Genie space for the sales org. While configuring compute, they try to attach a Classic SQL warehouse from the dropdown, but it is not among the valid options. Which warehouse choice should they make instead, and why?
Correct answer: A — A Pro or serverless SQL warehouse, since Genie spaces only support those two types, and Databricks recommends serverless for the best query performance
- A. This is correct because Genie spaces can only be attached to a Pro or a serverless SQL warehouse, and Databricks recommends serverless specifically because it delivers the best query performance for the space.
- B. This is incorrect because Classic warehouses are not one of the supported warehouse types for a Genie space regardless of whether autoscaling is turned on; the restriction is on warehouse type, not on scaling configuration.
- C. This is incorrect because a Genie space still relies on an attached workspace SQL warehouse to execute the SQL it generates; there is no separate internal compute engine that bypasses warehouse selection.
- D. This is incorrect because Pro warehouses are also a supported option for a Genie space; serverless is merely the recommended choice for performance, not the only permitted one.
Sources1
Exam traps
Each one states something that sounds right. Open it to see what is actually true.
1.Because the author's credentials are embedded in the space, users can see any data the author can see.Why is that wrong?
The embedded credentials grant access to the warehouse only. Data access is still checked against each user's own Unity Catalog permissions.
Covered in Choosing the SQL warehouse
2.Editing a column description or adding synonyms in a Genie space updates the table's comments in Unity Catalog.Why is that wrong?
Column descriptions and synonyms you add in a space apply only to that space. The metadata in Unity Catalog stays unchanged.
Covered in Curating Unity Catalog datasets
Sources
Every claim above is drawn from one of these pages, quoted as it was written on the date shown.
- 1.
“Genie Agents were formerly known as Genie Spaces.”
↩︎ What a Genie space is and how you create one“Genie Agent creators automatically have CAN MANAGE permissions on agents they create.”
↩︎ What a Genie space is and how you create one“Click New in the upper-right corner of the screen.”
↩︎ What a Genie space is and how you create one“You can add up to 50 tables, views, or metric views to a Genie Agent.”
↩︎ Curating Unity Catalog datasets“For optimal performance, Databricks recommends using a serverless SQL warehouse.”
↩︎ Choosing the SQL warehouse“If a different author changes the warehouse later, their compute credentials are embedded instead.”
↩︎ Choosing the SQL warehouse“Author-defined common questions take priority and appear first.”
↩︎ Title, description and sample questions“Choose a title that will help end users discover your Genie Agent.”
↩︎ Title, description and sample questions“Data access is always evaluated using each end user's own Unity Catalog permissions.”
↩︎ Exam trap 1“Choose the data sources that you want to include in your Genie Agent. Then, click Create.”
↩︎ Checkpoint“Genie does not query tables or Unity Catalog functions that are not attached, even if the agent's instructions or metadata reference them.”
↩︎ Checkpoint“These embedded credentials grant access to the warehouse only.”
↩︎ Checkpoint“When you provide fewer, Genie fills the remaining slots with auto-generated questions.”
↩︎ Checkpoint - 2.
“including managed tables, external tables, foreign tables, views, metric views, and materialized views”
↩︎ Curating Unity Catalog datasets“Generated queries are always read-only.”
↩︎ Choosing the SQL warehouse - 3.
“Metric views are particularly effective for Genie Agents because they pre-define metrics, dimensions, and aggregations.”
↩︎ Curating Unity Catalog datasets“bridge the gap between Genie's general world knowledge and the specialized language used in a specific domain or company”
↩︎ Key concept“This metadata is scoped to your Genie Agent and does not overwrite metadata stored in Unity Catalog.”
↩︎ Exam trap 2“Aim for five or fewer tables.”
↩︎ Prediction - 4.https://www.databricks.com/blog/data-dialogue-best-practices-guide-building-high-performing-genie-spacesSecondary source
“Do not rely on instructions or prompts to compensate for poor modeling choices.”
↩︎ Curating Unity Catalog datasets