Create a Genie Clean Room

Beta

This feature is in Beta. Account admins can manage access to this feature from the account console Previews page. See Manage account-level previews.

A Genie Clean Room lets a data provider share one Genie Agent with a recipient in another organization. The recipient asks the agent questions in natural language and gets answers over the provider's data, without seeing the agent's instructions or the raw data.

To learn about Genie Agents, see Genie Agents. For an overview of clean rooms, see What is Databricks Clean Rooms?.

Why use a Genie Clean Room?

Share your curated Genie Agents securely with your customers, without allowing them to access the underlying data, context, or proprietary IP:

• Share agents, not data: Customers can use your agents without seeing the underlying data and context.
• Combine datasets: Customers can bring their own data and get insights from the combination of both datasets.
• Govern usage: Set a daily prompt limit. Customers can't bulk download the agent's outputs.

How Genie Clean Rooms work

When you create the clean room, Databricks shares your Genie Agent and all of its instructions into the central clean room, along with the tables, views, and volumes the agent references. The recipient chats with the agent from the Chat tab on the clean room details page. Queries run in the central clean room, not in either collaborator's workspace.

The following table compares what each collaborator can do in a Genie Clean Room:

Capability	Provider	Recipient
Attach and sync the Genie Agent	Yes	No
Set the daily prompt limit	Yes	No
View the agent's instructions and examples	Yes	No
Add data assets	No (data can't be added through the clean room, only from the Genie Agent)	Yes
Ask the agent questions	No	Yes

Capability	Provider	Recipient
Attach and sync the Genie Agent	Yes	No
Set the daily prompt limit	Yes	No
View the agent's instructions and examples	Yes	No
Add data assets	No (data can't be added through the clean room, only from the Genie Agent)	Yes
Ask the agent questions	No	Yes

The recipient can view the provider's table names and column metadata. The clean room UI doesn't show the recipient sample values or table contents, but it's possible the agent returns row-level values if the recipient asks for them. All downloads are disabled, so the recipient can't bulk export agent output. The provider can't see any data assets the recipient adds, including their names.

The provider views the agent's instructions and examples under Genie Agent Instructions on the clean room's Overview tab. The recipient sees a notice that the instructions are hidden, and can't add instructions of their own.

If the provider updates the Genie Agent in their workspace, the recipient doesn't see the changes until the provider syncs them. See Sync changes to the Genie Agent.

Unlike a packaged clean room, where a consumer runs the provider's notebooks or JARs, a Genie Clean Room runs no notebooks or JARs.

Requirements

• Your workspace meets the requirements for clean rooms. See Requirements.
• An account admin for both the provider and the recipient has turned on the Genie Clean Room preview from the account console Previews page. See Manage account-level previews.
• To create the clean room, the provider must have:
  • The CREATE CLEAN ROOM privilege on the metastore.
  • CAN EDIT permission on the Genie Agent they want to share.
  • SELECT on every table and metric view, and READ VOLUME on every volume, that the Genie Agent references.
  • USE CATALOG and USE SCHEMA on the catalogs and schemas that contain those assets.
  • The recipient's Clean Room sharing identifier. See Step 1. Request the collaborator's sharing identifier.
• To add their own data to the clean room, the recipient must be the owner of the clean room or have the MODIFY CLEAN ROOM privilege on it, and must have SELECT on any table or view and READ VOLUME on any volume that they add, plus USE CATALOG and USE SCHEMA on the catalogs and schemas that contain them.
• To ask the agent questions, the recipient must have the EXECUTE CLEAN ROOM TASK privilege on the clean room.

Best practices

Databricks recommends the following when you create a Genie Clean Room:

• Finish curating before you share: Review the agent's instructions and examples in your workspace before you create the clean room, because the recipient's experience depends on them.
• Keep the agent focused: Reference only the tables, views, and volumes the recipient needs, and keep the agent within the size limit. See Limitations.
• Start with a conservative prompt limit: For a trial or evaluation, set a low daily limit and raise it later.
• Sync after every change: Changes you make to the agent in your workspace don't reach the clean room until you sync.

Create a Genie Clean Room

Complete the following steps as the provider.

Get the recipient's sharing identifier

Ask the recipient for their Clean Room sharing identifier. They can find it using the instructions in Find your sharing identifier.

Create the clean room

1. In your Databricks workspace, click Catalog.

2. Click the Clean Rooms > button.

3. Click Create Clean Room.

4. Under Clean room type, select Genie Clean Room.

5. In Clean Room name, enter a name for the clean room.

   The name can't use spaces, periods, or forward slashes (/), and you can't change it after you create the clean room.

6. In Region, select the cloud and region for the central clean room. See Regional availability.

7. In Clean Room sharing identifier, enter the recipient's sharing identifier.

   You can't change collaborators after you create the clean room.

8. Under Genie Agent, search for and select the Genie Agent to share. You can't replace it with a different agent later.

9. Click Create clean room.

Set the daily prompt limit

1. On the clean room details page, click the Output Controls tab.
2. In Allowed prompts per day, enter the maximum number of prompts the recipient can send per day.
3. Click Save.

Each message the recipient sends and each response they regenerate counts as one prompt. The count resets at midnight UTC. The recipient can view the limit on their Output Controls tab, but can't change it.

Sync changes to the Genie Agent

When you change the Genie Agent in your workspace, sync the clean room to share the changes with the recipient. Syncing updates the agent's instructions, examples, and referenced tables, views, and volumes in the clean room. It doesn't change or remove the data assets the recipient added.

1. On the clean room details page, click the Overview tab.
2. In the Genie Agent section of the right pane, click Sync Genie Agent.

Use a Genie Clean Room

Complete the following steps as the recipient.

Open the clean room

1. In your Databricks workspace, click Catalog.
2. Click the Clean Rooms > button.
3. Click the name of the clean room the provider shared with you.

The clean room opens on the Chat tab, where you chat with the provider's Genie Agent. To view the names of the provider's tables, click the Overview tab.

Add your own data (optional)

To let the agent combine your data with the provider's, add your Unity Catalog tables, views, metric views, streaming tables, foreign tables, or volumes to the clean room. The agent can query your data alongside the provider's data. The provider can't see the assets you add, including their names.

1. On the clean room details page, click the Overview tab.
2. Click Add data assets.
3. Select the data assets to add, review their aliases, and click Save.

For details about selecting assets and aliases, see Step 3. Add data assets and notebooks to the clean room. For the permissions you need, see Requirements. You can't add notebooks, JARs, or models to a Genie Clean Room.

Ask questions

On the Chat tab, enter a question in natural language, or click one of the suggested questions. The agent answers using the provider's instructions and the data in the clean room. To view past conversations, click the history icon.

The upper-right corner of the chat shows how many prompts you have left today, out of the provider's daily limit. If you reach the provider's daily prompt limit, the chat shows the following message until midnight UTC:

Text

You've reached this Genie clean room's limit of {limit} prompts per day. Please wait until the next UTC day before asking another question.

Limitations

• A Genie Clean Room has exactly two collaborators: the provider and one recipient. The collaborator who creates the clean room is the provider.
• A clean room can have only one Genie Agent, and you can't replace it with a different agent.
• Genie Clean Rooms support only chat mode. Agent mode isn't supported.
• You can't add or sync a Genie Agent that references any of the following. The agent can use these sources in your workspace, but they can't be shared into a clean room:
  • Tables in the system or samples catalogs.
  • Tables in an information_schema schema.
  • Tables you received through OpenSharing, including tables from Databricks Marketplace.
• Tables that the Genie Agent references from different catalogs can't share the same schema.table name.
• The provider can't view the recipient's conversations.
• Recipients can't upload files in the clean room chat.
• Recipients can't add their own instructions to the Genie Agent.
• If the Genie Agent definition is larger than 262,000 bytes when compressed, the clean room can't share it. Reduce the number of instructions or example queries.
• A Genie Agent created from an integration, such as Catalog Explorer, can't be shared into a clean room.
• Natural-language answers can still reveal information about the provider's data. Databricks doesn't enforce aggregation thresholds or limit the size of answers in Genie Clean Rooms.
• The recipient's prompts and conversations aren't recorded in the clean room events system table or in audit logs. When the provider attaches or syncs the agent or changes the prompt limit, audit logs record createCleanRoomAsset or updateCleanRoomAsset events. See Clean Rooms events.

Additional resources

• • Genie Agents
  • Build and curate the Genie Agent you share, including its instructions and example queries.
• • Work with Databricks clean rooms as an invited collaborator
  • Find your sharing identifier and work in a clean room that another organization shared with you.