> [!WARNING]
>
> **Preview**
>
>
> This product or feature is
>
> subject to the "Pre-GA Offerings Terms" in the General Service Terms section
> of the [Service Specific
> Terms](https://docs.cloud.google.com/terms/service-terms#1).
>
> Pre-GA products and features are available "as is" and might have limited support.
>
> For more information, see the
> [launch stage descriptions](https://cloud.google.com/products/#product-launch-stages).

Conversational Analytics API has the following known limitations on the number of data sources, style of visualizations, and size of datasets.

## Data source limitations

This section describes the constraints and behaviors of the Conversational Analytics API when
you connect to and query BigQuery, Looker, and database sources---
AlloyDB for PostgreSQL, Cloud SQL for MySQL, Cloud SQL for PostgreSQL,
and Spanner.

### Looker data source limitations

When you connect to a Looker data source, note the following:

- You can query any included Explore in a conversation.
- An agent can only query one Explore at a time. It is not possible to perform queries across multiple Explores simultaneously.
- An agent can query multiple Explores in the same conversation.
- An agent can query multiple Explores in a conversation that includes questions with multiple parts, or in conversations that include follow-up questions.

  For example: A user connects two Explores, one called `cat-explore` and one called `dog-explore`. The user inputs the question "What's greater: the count of cats or the count of dogs?" This would create two queries: one to count the number of cats in `cat-explore` and one to count the number of dogs in `dog-explore`. The agent compares the number from both queries after completing both queries.
- The `QueryData` method does not support BigQuery or Looker data sources.

### Database data source limitations

When you connect to an AlloyDB, Cloud SQL for MySQL,
Cloud SQL for PostgreSQL, or Spanner data sources, consider the
following:

- Data agents access the data using the credential of the user who is interacting
  with the data agent. If a user accesses a shared data agent for which they don't
  have access to the agent's configured tables, then the data agent can't access
  those tables.

- The table selection for the data agent guides your agent about which
  tables to focus on. Table selection isn't a security setting. Even if you specify that the
  data source can only pull information from certain tables---like `table1` and
  `table2`---the system might still return data from an unintended table (`table3`)
  if the user running the query has general permissions to view the content of
  `table3` in the same database.

### BigQuery search scope limitations

When you use a search scope to let the Conversational Analytics API choose which
BigQuery tables to query, the following limitations apply:

- You can use a search scope only in stateless chat requests with inline context. Search scopes aren't supported in persistent data agents or in stateful chat requests that use a conversation reference.
- You can specify up to 100 datasets and up to 100 projects.

For more information, see
[Connect to BigQuery datasets and projects](https://docs.cloud.google.com/gemini/data-agents/conversational-analytics-api/connect-bigquery.md.txt#connect-to-bq-datasets-projects).

## Visualization limitations

The following visualization types are supported:

- Area
- Bar
- Geoshape
- Heatmap
- Line (Time series)
- Pie
- Scatter

> [!NOTE]
> **Note:** Charts with tooltips are partially supported and may exhibit unexpected behavior.

## Data processing limitations

- For Looker data sources, the Conversational Analytics API can return a maximum of **5,000 rows** per query.
- For BigQuery data sources, the Conversational Analytics API limits data queries to 500 GB of bytes processed.
- For AlloyDB, Cloud SQL for MySQL, Cloud SQL for PostgreSQL, and Spanner data sources, the Conversational Analytics API can return a maximum of 1,000 rows per query.
- The Conversational Analytics API's Python-based reasoning and content retrieval capabilities can accommodate time complexities of up to `O(100k)` rows.
- Querying large amounts of data can cause reduced reasoning accuracy in data agents.
- The Conversational Analytics API has a maximum token output length of **8,192 tokens** . Querying large amounts of data can return a `MAX_TOKENS` error.
- The data returned within the `DataResult` field of a system message is subject to a size limit. Data results are truncated to a maximum of 3,000,000 bytes. This truncation process keeps as many full rows as possible within this size constraint.

## Query limitations

- BigQuery's flexible column names feature isn't supported.
- Structs in BigQuery are supported but may sometimes fail.
- For Looker data sources, the API can't set the value of a filter-only field that is defined by using the LookML `parameter` parameter.
- Using the Conversational Analytics API to connect to a private IP Looker (Google Cloud core) instance using Data Studio Pro when that Looker (Google Cloud core) instance is inside a VPC Service Controls perimeter isn't a supported configuration and doesn't meet VPC Service Controls compliance requirements.
- For connections to [Looker (Google Cloud core) instances with private IP configurations](https://docs.cloud.google.com/looker/docs/looker-core-networking-options.md.txt#private_ip_connections), the Conversational Analytics API doesn't support Looker (Google Cloud core) instances that are [configured to use CMEK](https://docs.cloud.google.com/looker/docs/looker-core-cmek.md.txt) or VPC Service Controls.
- The Conversational Analytics API doesn't work well with Data Studio data sources that have [field editing in reports](https://docs.cloud.google.com/looker/docs/studio/edit-fields-in-your-reports) disabled because this setting prevents Conversational analytics from creating calculated fields.
- When a failure occurs during query validation or execution, the Conversational Analytics API may automatically retry the operation by generating a corrected query. This kind of retry will be attempted a maximum of three times per request.

  If a query fails because of permission or authentication issues, the Conversational Analytics API won't retry the query. Retries are non-deterministic; if the error message suggests that a query is unrecoverable, then the Conversational Analytics API won't try the query again, even if it's still below the limit of three errors per request.

## Quota limitations

- The Conversational Analytics API has the following limits for overall requests (including both chat and non-chat requests):
  - A maximum rate of 30 queries per second (QPS), which is 1,800 queries per minute (QPM) per project.
  - A maximum rate of 10 QPS, which is 600 QPM per user per project.
- As an exception, chat requests have more restrictive limits:
  - A maximum rate of 30 QPM per project.
  - A maximum rate of 30 QPM per user per project.
  - Each chat request counts as 8 quota units, rather than the standard of 1 unit. As a result, the Google Cloud console [Quotas](https://console.cloud.google.com/apis/api/geminidataanalytics.googleapis.com/quotas) page displays limits that are 8 times the actual request limits. For example, a limit of 240 units represents 30 chat requests.
- Agent-to-Agent (A2A) requests have a maximum rate of 100 QPM per project per agent. Built-in agents each receive a dedicated limit, while custom data agents share a single pooled 100 QPM limit per project.
- Conversational Analytics API for AlloyDB, Cloud SQL for MySQL, Cloud SQL for PostgreSQL, and Spanner has a 50 QPM limit per project. To increase these limits, contact [Google Cloud Customer Care](https://cloud.google.com/support?e=48754805).

## Question types limitations

- The Conversational Analytics API supports questions that can be answered by a single visualization, for example:

  - Metric trends over time
  - Breakdown or distribution of a metric by dimension
  - Unique values for one or more dimensions
  - Single metric values
  - The top dimension values by metric
- The Conversational Analytics API doesn't yet support questions that can only be answered with the following types of complicated visualizations:

  - Prediction and forecasting
  - Advanced statistical analysis, including correlation and anomaly detection