What you will be able to do
- Explain the two checks a user must pass before a Genie Agent (Genie space) answers them: a grant on the space and Unity Catalog access to its data
- Pick the right Genie permission level (CAN VIEW, CAN RUN, CAN EDIT or CAN MANAGE) for a given person
- Share a Genie Agent from the Share dialog with users, groups or all account users, and tell agent sharing apart from conversation sharing
Key concept
Two gates for every Genie user — Opening a Genie Agent needs a permission on the agent itself. Seeing any data through it needs the user's own Unity Catalog privileges on the tables behind it. Compute is the exception: it runs on the author's embedded warehouse credentials, so users need no warehouse grant.
1.What a user needs before Genie will answer
A Genie Agent (formerly called a Genie space) is only useful once people outside its author can ask it questions. Giving them that ability touches three separate things: the agent, the data and the compute. Each one is handled differently, and most access problems come from treating one as if it covered another.
The agent gate is a permission on the Genie Agent object. Anyone who will ask questions needs at least CAN VIEW or CAN RUN on the agent, plus the consumer access or Databricks SQL workspace entitlement. Without a grant on the agent they cannot use it, even if they can read every table it covers. SELECT on the tables is a data privilege. It does not count as permission on the agent.
The data gate is Unity Catalog. Every user needs at least SELECT on all the Unity Catalog objects the agent uses, and Genie checks this against the user's own identity, never the author's. The privilege model itself, including the USE CATALOG and USE SCHEMA grants a reader needs on the way down to a table, is covered in Unity Catalog privileges, usage rules and inheritance. For Genie, the point is that a grant on the agent never gives anyone access to data.
Compute is the one thing users do not need. The SQL that Genie generates runs on the warehouse using credentials the author embedded when they configured it, so you never grant CAN USE on the warehouse to people who only ask questions. Because data access follows each user, row filters and column masks in Unity Catalog apply automatically. Two people asking the same question can see different rows, and every query shows up under the person who asked it in query history.
Checkpoint 1 of 4· Check yourself
Marco has SELECT on every table behind a finance Genie Agent, but nobody has added him in the Share dialog. He also has no permissions on the agent's SQL warehouse. What does he still need before he can ask the agent questions?
SELECT is a data privilege and never stands in for a permission on the agent, so Marco needs at least CAN VIEW or CAN RUN on it. He needs no warehouse grant because Genie's queries run on the compute credentials the author embedded.
“End users do not need direct warehouse permissions.”Source: docs.databricks.com
Checkpoint 2 of 4· Exam question
A data analyst on the Marketing team has been granted SELECT privileges on every Unity Catalog table and view curated inside a Genie space, but has not been added to the space's Share dialog. When the analyst tries to open the space and ask a question, what happens?
Correct answer: B — The analyst is blocked from querying, because Genie enforces both an explicit CAN VIEW or CAN RUN grant on the space itself and SELECT privileges on the underlying data.
- A. SELECT privileges on the curated tables are necessary but not sufficient. Genie also requires the user to hold an explicit permission grant on the space itself, so table-level access alone does not open the space.
- B. Correct. Genie access is layered: the user needs an explicit CAN VIEW or CAN RUN grant on the space through the Share dialog, and separately needs SELECT on the underlying Unity Catalog objects the space queries.
- C. Warehouse compute for a Genie space runs under the embedded credentials of the space's author, not the querying user's own warehouse permission, so this is not the reason access would be blocked.
- D. Unity Catalog SELECT privileges have no automatic relationship to Genie space permission tiers, and there is no mechanism that escalates a data grant into a CAN EDIT request.
Sources1
2.The four Genie permission levels
Genie Agents use four permission levels: CAN VIEW, CAN RUN, CAN EDIT and CAN MANAGE. The levels are cumulative, but they do not form four useful steps. For a Genie Agent, CAN VIEW and CAN RUN grant exactly the same abilities, so a reader-only agent effectively has three tiers: ask, curate, own.
| Ability | CAN VIEW / CAN RUN | CAN EDIT | CAN MANAGE |
|---|---|---|---|
| See the agent in the Genie Agent list | Yes | Yes | Yes |
| Ask questions, provide response feedback, upload files to a conversation | Yes | Yes | Yes |
| Add or edit instructions and sample questions; add or remove included tables | No | Yes | Yes |
| Monitor a space and view other users' conversations | No | No | Yes |
| Modify permissions and delete the space | No | No | Yes |
Match the level to the job. Business users who only ask questions get CAN VIEW or CAN RUN, since the two are equivalent here. Analysts who maintain the instructions, sample questions and table list get CAN EDIT. CAN MANAGE goes to the agent's owners: only they can change who has access, delete the agent, or open the monitoring view of everyone's conversations. You never assign CAN MANAGE to the person who builds an agent, because Genie Agent creators get it automatically on agents they create.
Checkpoint 3 of 4· Check yourself
A sales team's business users have CAN VIEW on a Genie Agent. One of them asks to be upgraded to CAN RUN so they can start asking the agent questions. What should you do?
On a Genie Agent, CAN VIEW and CAN RUN grant exactly the same abilities, and both include asking questions. CAN EDIT is for curating instructions and tables. Users never need a warehouse grant, because queries run on the author's embedded credentials.
“At least CAN VIEW permission on the Genie Agent.”Source: docs.databricks.com
3.Sharing from the Share dialog
A new Genie Agent starts private to its author, and folder permissions are the quiet way it gets shared. New agents are saved to the author's user folder and, like other workspace objects, inherit permissions from the folder they sit in. If you move an agent into a team folder, everyone with rights on that folder gets the same rights on the agent. That is convenient for a team space and a surprise in a broadly shared one.
The Share button at the top of the agent is the explicit route. You can share with specific people or with everyone:
- Click Share.
- Enter users or groups, click Add, and set each one's permission level. Individual users and members of small groups get an email notification.
- Alternatively, select All account users and give everyone in the account a single level.
- Use Copy link at the bottom of the dialog to get a URL. Only users who already have permission can open it in a new tab and ask questions.
The copied link is a pointer, not a grant. Pasting it into a team chat does nothing for someone without a permission on the agent and SELECT on its data. Grant first, then send the link.
Two other Share controls look similar but do different jobs. A conversation's own Share option, found in the chat history menu, controls who can view that conversation: Private, Reviewable by agent managers (the default), or All account users. It never changes permissions on the agent. The External tab of the agent's Share dialog shares the agent through OpenSharing with recipients outside your organization. That sends a fixed snapshot of the agent's data and instructions, not live access to it.
Checkpoint 4 of 4· Check yourself
You paste the agent's Copy link URL into a team chat. One colleague there has SELECT on every table the agent uses but has never been granted anything on the agent. What gets them asking questions?
The copied link only points to the agent and grants nothing. The colleague still needs a permission on the agent itself, and their SELECT grants cover only the data. A conversation's Share option affects only who can see that conversation. The External tab sends a fixed snapshot to recipients outside the organization, not live access.
“End users: Must have explicit access to the Genie Agent and the underlying data assets it uses.”Source: docs.databricks.com
Sources
(7)
Every claim above is drawn from one of these pages, quoted as it was written on the date shown.
- 1.
“Data access: At least SELECT privileges on all of the Unity Catalog data objects used in the agent.”
↩︎ What a user needs before Genie will answer“The consumer access or Databricks SQL workspace entitlement.”
↩︎ What a user needs before Genie will answer“Row filters and column masks defined in Unity Catalog are enforced per user automatically, regardless of how the Genie Agent is shared.”
↩︎ What a user needs before Genie will answer“Genie Agent creators automatically have CAN MANAGE permissions on agents they create.”
↩︎ The four Genie permission levels“Like other workspace objects, they inherit permissions from their enclosing folder.”
↩︎ Sharing from the Share dialog“Use the Copy link button at the bottom of the Share dialog to get a shareable link to the Genie Agent.”
↩︎ Sharing from the Share dialog“Genie Agent ACLs: At least CAN VIEW/CAN RUN permissions on the Genie Agent.”
↩︎ Exam trap 1 - 2.
“For a Genie Agent, CAN VIEW and CAN RUN grant the same abilities.”
↩︎ The four Genie permission levels“For a Genie Agent, CAN VIEW and CAN RUN grant the same abilities.”
↩︎ Exam trap 3 - 3.
“It is separate from the Share button at the top of the Genie Agent, which manages permissions on the agent itself.”
↩︎ Sharing from the Share dialog - 4.
“In the upper-right corner, click Share, then click the External tab.”
↩︎ Sharing from the Share dialog
Also cited
“Unity Catalog evaluates data access as the end user's own identity, not the author's.”
↩︎ Key concept“End users do not need direct warehouse permissions.”
↩︎ Exam trap 2“Any question about data a user cannot access returns an empty response.”
↩︎ Prediction“End users do not need direct warehouse permissions.”
↩︎ Checkpoint“At least CAN VIEW permission on the Genie Agent.”
↩︎ Checkpoint“End users: Must have explicit access to the Genie Agent and the underlying data assets it uses.”
↩︎ Checkpoint