Analyze your stored data with Gemini assistance

This page describes how to use AI-powered assistance through Gemini to help you better understand your data storage landscape in Cloud Storage. When you use Gemini Cloud Assist, you can enter prompts about how Cloud Storage works in general, and when you enable a Storage Intelligence subscription, you can enter prompts about specific buckets and objects.

You can then use the information provided by Gemini Cloud Assist to do the following:

  • Analyze data usage patterns and trends

  • Identify opportunities to save on storage costs

  • Optimize your data for security and compliance

  • Make informed decisions on managing your data

Gemini doesn't use your prompts or its responses as data to train its models without your express permission. For more information about how Google uses your data, see How Gemini for Google Cloud uses your data.

This page is intended for developers, data analysts or data engineers, platform admins, finance operators, and compliance officers. It assumes that you know how to use Cloud Storage and query linked datasets.

What prompts can Gemini answer?

To understand your data storage, you can provide Gemini Cloud Assist with a prompt, which is a natural language statement or question.

When you use Gemini Cloud Assist alone without a Storage Intelligence subscription, you can ask prompts about how Cloud Storage generally works, such as the following:

  • "How do I transition objects from one storage class to another?"

  • "How do I enable soft delete on a bucket?"

When you enable a Storage Intelligence subscription, you can also ask prompts related to cost saving opportunities, security and compliance, and data discovery. Gemini Cloud Assist can use the metadata contained in Storage Insights datasets to generate insights, which are responses to prompts about your bucket and object metadata and usage. You can ask prompts such as the following:

  • Prompts related to usage and cost savings:

    • "5 largest buckets without Object Lifecycle Management"

    • "5 largest buckets without Autoclass enabled"

  • Prompts related to security and compliance:

    • "5 largest buckets without object versioning enabled"

    • "5 largest objects with a retention expiration date within the next 30 days"

  • Prompts related to data discovery:

    • "Buckets with a high volume of small files (each under 1 MB)"

    • "Objects in Standard storage class smaller than 50 MB"

Before you begin

In order to use Gemini Cloud Assist for general prompts related to Cloud Storage, you must first set up Gemini Cloud Assist, including getting required roles.

Set up Gemini Cloud Assist

Ensure that Gemini Cloud Assist is set up for your Google Cloud project.

If you want to enter prompts related to specific buckets and objects, you must also complete the following prerequisite steps:

  1. Enable Storage Intelligence, which gives you access to using Storage Insights datasets.

  2. Create a Storage Insights dataset, which Gemini Cloud Assist will analyze to provide information about specific buckets and objects.

    Alternatively, if there's an existing dataset you want to use, you can get the required IAM roles for accessing the existing dataset.

  3. Ensure that the Storage Insights service agent has access to the dataset Gemini Cloud Assist will analyze. This enables the dataset to be read and analyzed.

Enable Storage Intelligence

Ensure that Storage Intelligence is enabled on the project, folder, or organization that contains or will contain the datasets that Gemini Cloud Assist will use to answer prompts.

Create a dataset

Create a dataset and link the dataset to BigQuery.

Grant required roles for accessing datasets

When a user first creates a dataset configuration, an Storage Insights service agent is created. The service agent follows the naming format service-PROJECT_NUMBER@gcp-sa-storageinsights.iam.gserviceaccount.com and appears on the IAM page of the Google Cloud console when you select the Include Google-provided role grants checkbox.

In order to use Gemini Cloud Assist for prompts related to bucket or object metadata, you need to enable the Storage Insights service agent to read datasets. Ask your administrator to grant the service agent the BigQuery Data Viewer role (roles/bigquery.dataViewer) on the organization, folder, or project that contains the dataset you want to analyze.

For instructions on granting roles to service agents, see create and grant roles to service agents.

Get required roles for accessing datasets

To get the permissions that you need to get insights on bucket and object metadata, ask your administrator to grant you the following IAM roles on the project, folder, or organization that contains the datasets you want to analyze:

For more information about granting roles, see Manage access to projects, folders, and organizations.

You might also be able to get the required permissions through custom roles or other predefined roles.

Analyze your data storage by using natural language prompts

To enter prompts to Gemini Cloud Assist, follow these steps:

  1. In the Google Cloud console, go to the Cloud Storage Storage Insights page.

    Go to Storage Insights

  2. In the toolbar, click spark (Gemini) to open the Cloud Assist chat panel.

    The Cloud Assist chat panel appears.

  3. In the Cloud Assist chat panel, enter a natural language prompt about your data storage. For example, you might enter the following:

    Which is my largest bucket

  4. Click (Generate).

  5. If prompted to, enter the name of the dataset that Gemini will analyze to generate the response, then click (Generate).

    If successful, Gemini Cloud Assist generates a response similar to the following:

    Here's what I found by analyzing the data in EXAMPLE_DATASET:

    

    
Bucket name
Size

    

    
my-bucket
39.1 TB

    The underlying SQL query that Gemini Cloud Assist uses is also returned. The generated SQL query is similar to the following:

    SELECT bucket_id, bucket_size
FROM buckets
WHERE project_id = 'example-project'
ORDER BY bucket_size DESC
LIMIT 1;

Optionally, you can enter suggested prompts:

  1. In the Google Cloud console, go to the Cloud Storage Storage Insights page.

    Go to Storage Insights

  2. In the Suggested prompts section, select a suggested prompt. For example, a suggested prompt might say: Storage size broken down by object content type.

  3. If successful, Gemini Cloud Assist generates a response similar to the following:

    Here's what I found by analyzing the data in EXAMPLE_DATASET:

    

    
Content type
Size

    

    
MP4
483.2 GB

    

    
MOV
239.1 GB

    

    
MP3
125.8 GB

    The underlying SQL query that Gemini Cloud Assist uses is also returned. The generated SQL query is similar to the following:

    SELECT
oa.contentType, ROUND(sum(oa.size) / (1024 * 1024 * 1024), 2) AS total_size_gb
FROM object_attributes_latest AS oa
GROUP BY oa.contentType
ORDER BY sum(oa.size) DESC;

Limitations

  • When using prompts to analyze Cloud Storage resources, you can specify up to five resources in the prompt. For example: 5 largest buckets without Autoclass enabled. Even if you include more than five resources in the prompt, Gemini Cloud Assist can only return results for the top five resources that match the prompt's criteria.

  • Gemini Cloud Assist uses the bucket and object metadata contained in Storage Insights datasets to respond to prompts about your data storage.

    The following data is not available in Storage Insights datasets, meaning that Gemini Cloud Assist lacks the proper context to respond to prompts relating to that data:

    • Specific cost data, for example: "how much does my bucket cost per month".

    • Activity data, for example: "what is the last access time of my object".

    • Data about the configuration of certain features, such as soft delete. For example: "which buckets don't have soft delete enabled".

    Gemini Cloud Assist also lacks the proper context to respond to prompts related to time series information. For example: "how much did my bucket grow in the last 3 weeks". This is because Gemini Cloud Assist only reads a dataset's latest snapshot.

What's next