What you will be able to do
- Identify the Unity Catalog data objects a Genie space can use and the limits that apply
- Explain the role of the SQL warehouse and how data access is governed
- Distinguish the knowledge store, instructions, example SQL queries, trusted assets, benchmarks and Agent mode
- Recognise the settings and interface areas users see in a Genie space
1.Curated Unity Catalog data
A Genie space (the documentation now calls it a Genie Agent; Genie Agents were formerly known as Genie Spaces) is a natural-language chat interface over a curated set of data. Everything starts with that data, and it must be registered to Unity Catalog. Supported objects include managed, external and foreign tables, views, metric views and materialized views. A single space can hold up to 50 tables, views or metric views. Each space supports up to 200,000 conversations, and each conversation can hold up to 10,000 messages.
Genie reads the Unity Catalog metadata attached to those objects. Well-annotated datasets paired with specific instructions are what make the experience good for end users.
The attached sources define the outer boundary of what a space can reach, and Unity Catalog permissions narrow it further for each person. Access to attached sources is still governed by each user's own Unity Catalog permissions. Two users in the same space can therefore see different results if their grants differ.
Checkpoint 1 of 5· Check yourself
Which of these can be the data foundation of a Genie space?
Genie spaces are built on Unity Catalog objects, including metric views, and they query only the objects attached to the space.
“A Genie Agent is based on data registered to Unity Catalog, including managed tables, external tables, foreign tables, views, metric views, and materialized views.”Source: docs.databricks.com
2.The SQL warehouse behind the space
Every SQL statement Genie generates runs on the space's default warehouse. That warehouse must be a pro or serverless SQL warehouse, and Databricks recommends serverless for the best performance.
The author's compute credentials are embedded in the space so that every user can reach the warehouse. Those embedded credentials cover the warehouse only. Data access is always evaluated against each end user's own Unity Catalog permissions. If a different author later changes the warehouse, that author's credentials are embedded instead.
Checkpoint 2 of 5· Check yourself
An analyst embeds their credentials by selecting the default warehouse for a Genie space. A business user without SELECT on one of the attached tables asks about it. What governs whether they see the data?
Embedded credentials give access to the warehouse only. Data access is always evaluated per end user in Unity Catalog.
“Data access is always evaluated using each end user's own Unity Catalog permissions.”Source: docs.databricks.com
Sources1
3.The knowledge components authors curate
Data and compute let Genie run queries. The knowledge components teach it what your business means. Data analysts curate each space with example SQL queries, SQL expressions for business semantics, and text instructions written in the organization's own terminology. The documentation defines six concepts.
| Concept | What it is |
|---|---|
| Knowledge store | Curated semantic definitions scoped to the space: table and column descriptions, synonyms, join relationships, SQL expressions, prompt matching settings |
| Instructions | Plain-text notes on business rules, preferred terminology, date formats and other domain context |
| Example SQL queries | Reference answers for common questions; Genie selects relevant ones from the query library |
| Trusted assets | Parameterized example queries and SQL functions whose exact logic an author has verified |
| Benchmarks | Test questions used only to evaluate response accuracy |
| Agent mode | Multi-step reasoning that runs multiple SQL queries and returns a structured report |
Two distinctions in this table come up often. First, the knowledge store lives in the space. Its configurations are scoped to the space and do not affect Unity Catalog metadata, so you can add synonyms or descriptions without changing the shared catalog. Second, benchmarks are for evaluation only. Unlike instructions and example SQL, benchmark questions and SQL are never used to improve Genie's context.
Trusted assets stand out because of the guarantee they give. When Genie answers with a trusted asset, the answer comes from logic an author has verified, not from freshly generated SQL.
Checkpoint 3 of 5· Match them up
Match each Genie space component to what it does
Tap a term, then the definition that fits it.
Each concept has a distinct job. Trusted assets guarantee verified logic, benchmarks only evaluate, the knowledge store adds semantics without touching Unity Catalog, and instructions add plain-text guidance.
“Unlike instructions, benchmarks are for evaluation only.”Source: docs.databricks.com
Checkpoint 4 of 5· Exam question
During Genie space creation, a data analyst is choosing which Unity Catalog objects to include so that the space stays focused and produces reliable answers for the marketing team. What is the recommended approach to this step?
Correct answer: A — Curate a small, well-understood set of tables and views directly relevant to the marketing team's questions, rather than exposing every catalog the analyst has access to.
- A. This is correct because curating a focused, well-understood set of datasets relevant to the intended audience is the recommended practice, since it reduces ambiguity for Genie and keeps generated SQL accurate and relevant.
- B. Exposing every catalog does not improve accuracy; it increases ambiguity about which tables are relevant to a given question and makes it harder for Genie to generate correct SQL.
- C. Genie's SQL generator can join across multiple curated tables and views, so restricting a space to one denormalized table is unnecessary and limits the questions it can answer well.
- D. Genie does not apply its own cleaning transformations ahead of query generation, so pointing it at unrefined bronze-layer tables would surface raw data quality issues in the answers rather than avoiding them.
Sources2
4.Settings and the interface users see
Beyond data and knowledge, a space has settings that shape how users find it and get started. The title appears in the workspace browser alongside other workspace objects. The description supports Markdown and explains the space's purpose. Tags help organize spaces, and a thumbnail appears on the chat landing page. Common questions are optional examples shown on the landing page. Author-written ones appear first, and Genie fills any remaining slots with auto-generated questions.
The user interface has these areas: - New chat: ask questions and read responses. - Chat history: threaded conversations. - Context: three tabs, About, Data and Instructions. Users can see the owner, the tables and the guidance behind the space. - Share: get a shareable link. - Monitoring: individual interactions, which users can rate or send for review.
A response usually combines a natural-language answer with a results table. Genie adds a visualization when one would make the answer clearer, and it includes the generated SQL, which users can view with Show code. Users can download up to about 1GB of results as CSV. Each response asks Is this correct?, and users can answer Yes, Fix it or Request review.
Feedback is a signal for authors, not an automatic learning loop. A space's behavior does not change on user feedback alone. Authors use the feedback to decide what to improve.
Checkpoint 5 of 5· Check yourself
Many users click 'Fix it' on the same wrong answer. What happens to the Genie space as a result?
Feedback surfaces improvement opportunities for editors, but the space's behavior changes only when someone changes its configuration.
“Your Genie Agent's behavior does not change based on user feedback alone.”Source: docs.databricks.com
Exam traps
Each one states something that sounds right. Open it to see what is actually true.
1.Adding benchmark questions and SQL to a space improves Genie's answers.Why is that wrong?
Benchmarks only measure accuracy. Genie does not use them as context.
Covered in The knowledge components authors curate
2.Adding synonyms or column descriptions in a Genie space overwrites the Unity Catalog comments for everyone.Why is that wrong?
Knowledge store settings are scoped to the space and leave Unity Catalog metadata unchanged.
Covered in The knowledge components authors curate
3.If the instructions mention a table, Genie can query it even though it is not attached to the space.Why is that wrong?
Genie uses only the sources explicitly attached to the space.
Covered in Curated Unity Catalog data
4.Any SQL warehouse, including a classic one, can power a Genie space.Why is that wrong?
Genie spaces require a pro or serverless SQL warehouse.
Covered in The SQL warehouse behind the space
Sources
Every claim above is drawn from one of these pages, quoted as it was written on the date shown.
- 1.
“You can add up to 50 tables, views, or metric views to a Genie Agent.”
↩︎ Curated Unity Catalog data“Each Genie Agent can support up to 200,000 conversations, and each conversation can include up to 10,000 messages.”
↩︎ Curated Unity Catalog data“For optimal performance, Databricks recommends using a serverless SQL warehouse.”
↩︎ The SQL warehouse behind the space“These embedded credentials grant access to the warehouse only.”
↩︎ The SQL warehouse behind the space“Common questions are optional examples of the types of questions users can ask.”
↩︎ Settings and the interface users see“A Genie Agent uses only the data sources explicitly attached under its Sources.”
↩︎ Exam trap 3“Genie Agents require a pro or serverless SQL warehouse.”
↩︎ Exam trap 4“Genie does not query tables or Unity Catalog functions that are not attached, even if the agent's instructions or metadata reference them.”
↩︎ Prediction“Data access is always evaluated using each end user's own Unity Catalog permissions.”
↩︎ Checkpoint - 2.
“Genie uses the metadata attached to Unity Catalog objects, as well as an author-curated agent-level knowledge store, to generate responses.”
↩︎ Curated Unity Catalog data“Parameterized example queries and SQL functions whose exact logic has been verified by an agent author.”
↩︎ The knowledge components authors curate“When Genie uses a trusted asset to generate a response, the answer comes from this verified logic.”
↩︎ The knowledge components authors curate“Genie does not use benchmark questions or SQL to improve its context.”
↩︎ Exam trap 1“Knowledge store configurations are scoped to the agent and do not affect Unity Catalog metadata.”
↩︎ Exam trap 2“A Genie Agent is based on data registered to Unity Catalog, including managed tables, external tables, foreign tables, views, metric views, and materialized views.”
↩︎ Checkpoint“Unlike instructions, benchmarks are for evaluation only.”
↩︎ Checkpoint - 3.
“Share: Get a shareable link to a Genie Agent.”
↩︎ Settings and the interface users see - 4.
“When Genie detects that a visualization could improve response clarity, it also returns a visualization.”
↩︎ Settings and the interface users see“Your Genie Agent's behavior does not change based on user feedback alone.”
↩︎ Checkpoint