August 2026 - Dashboard Release Notes

We now support percentage-based memory configuration for the Pi application, in addition to the traditional static memory configuration. This option is supported only for containerised deployments and is ideal when deploying to instances of different sizes, as the memory allocation is set automatically to an appropriate level.

You can set the maximum memory allocation using the PI_TOMCAT_MAX_MEMORY_PERCENTAGE environment variable. The value must be between 20 and 60, representing the percentage of the instance’s total memory. If the value is not supplied or falls outside the supported range, the application will revert to the default static memory configuration, with a maximum memory allocation of 1,024 MB.

Previously, AI Chart Generation responded only to direct chart creation prompts, such as: “Create a bar chart in the Bar category, using Department as the dimension and Number as the measure, from the TEST_AWS connection.”

This capability has now been enhanced to support a broader range of chart-related prompts. For example, users can ask which data connections are available and view the dimensions, measures and other data associated with a particular connection. They can then use the returned information to create a suitable chart.

The user’s conversation history is stored in the database and can be recalled unless they intentionally delete it by selecting Clear messages. The history remains available while the conversation is active. Messages that have not been updated for more than two hours are automatically deleted by a batch process that runs every hour.

Open

Regular tables currently run additional SQL queries to calculate accurate averages when an object’s select statement contains either a division operator (/) or the AVG() aggregate function, and the table column’s total type is set to Average.

This behaviour has now been introduced for crosstab tables, but currently applies only to row grand totals. Row subtotals and column totals continue to calculate averages using the existing method, without running additional SQL queries. Depending on the structure of the underlying data, this may produce inaccurate results.

The following example shows how a crosstab table calculated averages from the displayed table data before this change:

Open

Following this change, row grand total averages are calculated from the underlying data:

Open

The additional queries are shown in the chart debug panel. A warning is also displayed to explain that, for crosstab tables, this method currently applies only to row grand total averages:

Open

Guardrails have been added to AI Chart Search, AI Chart Generation and AI Report Summary. These guardrails detect inappropriate language, such as swear words, and ensure that the token limits for each AI feature are not exceeded.

Open

The Pi-AI MCP server provides several tools used by the AI Chart Builder. While these tools are running, status messages are now displayed beneath the loading icon in the Chart Builder chat, giving users greater visibility of progress.

We now provide a native ARM version of the Pirana Docker image, removing the need for emulation on ARM-based systems. The Pirana image is also now multi-architecture. From this release onwards, pulling the image will automatically retrieve the appropriate version for your system’s architecture.

Please consider this change if you pull, retag and host our images within your own infrastructure. Once pulled, the resulting local image is specific to that architecture and is no longer multi-architecture.

Several improvements have been made to the display and timing of loading icons while charts are being rendered:

• Resolved an issue where the Category Objects panel could cause a chart’s loading icon to be removed prematurely.

• Updated the loading icon logic so that the icon remains visible until immediately before the chart has finished rendering.

This means that visually complex charts containing large amounts of data, such as map charts, will continue to display a loading icon until they are ready to appear in the container.

A new Date/Time Filter category object has been added, enabling users to filter dashboard data using both date and time values. This provides more precise control when working with time-sensitive data and allows results to be narrowed to a specific period.

Open

Added a loading message and icon while an AI report summary is being generated. The AI Report Summary prompt is also no longer displayed.

Open

Recently removed database tables were incorrectly reported as missing in the logs during dashboard start-up. These obsolete tables have now been removed from the validation checks, preventing unnecessary warnings from being logged.

Previously, when creating and saving a model using the Analytics chart type, the generated SQL could use lists of individual values for comparisons, depending on the underlying data. The generated SQL now uses value ranges with the BETWEEN operator, resulting in more concise and efficient queries.

Example of generated SQL with previous implementation:

Open

Example of generated SQL with new implementation:

Open

AI Chart Generation responses now use the language associated with the user’s locale, regardless of the language used in the prompt. The example below shows AI-generated responses for a user whose locale is set to Japanese:

Open

A bug prevented charts from being displayed when Chart Viewer users performed an AI Chart Search, even when the search completed successfully. This issue has now been resolved, and matching charts are displayed as expected.

Open

The CORS security improvements introduced in the 2026_07 release caused issues when using JWT authentication through an initial POST request. This has now been resolved, with the /jwtLogin endpoint correctly inheriting the required CORS configuration.

Attempting to render a chart containing objects from a broken data connection could cause the AI panel to become unresponsive, requiring it to be closed before it could be used again. This issue has now been resolved.

Previously for Child Data Connections, any Data objects defined in the child itself would be returned via the V3 API and usable within the Chart Builder but those defined in the Parent would not. This has now been resolved.

When hovering over a sparkline value, speedometer charts incorrectly displayed two dials. In addition, neither speedometer nor power ring charts updated their central data value to reflect the selected sparkline point. Both issues were caused by structural changes to the chart configurations in a previous release and have now been resolved.

Last month, we removed code associated with the long-deprecated Flex application. We subsequently identified that charts created using certain Flex application configurations no longer rendered correctly. To restore compatibility, these changes have been reverted while a more comprehensive approach to addressing this technical debt is considered.

The health check used to monitor SSH tunnel connections has been updated to improve reliability. Port mappings are also now closed individually before the SSH tunnel is shut down, helping to ensure that associated resources are released cleanly.

Following recent licensing changes, callbacks to the ALM licensing service were occurring less frequently than intended. The callback and licence-caching behaviour has now been corrected. Existing XML-based licences are unaffected by this change.

Fixed an issue where the needle could obscure the data value in an exported Speedometer chart. This was most likely to occur when targets were configured and the short end of the needle overlapped the value.

• pi#3766 - Upgrade Logback to v1.5.35

  • CVE-2026-9828

  • CVE-2026-10532

• pi#3765 - Upgrade Grails to v7.2.1, Jackson Core to v2.22.1

  • CVE-2026-54515

• pi#3786 - Upgrade Netty to v4.1.136

  • CVE-2026-59901

  • GHSA-mfg7-5gfp-c4w3

  • CVE-2026-55831

  • CVE-2026-55833

  • CVE-2026-56745

  • CVE-2026-56746

  • CVE-2026-59898

  • CVE-2026-59899

  • CVE-2026-59921

  • CVE-2026-59900

• pi-ai#30 - Upgrade Spotless to v8.9.0

  • CVE-2026-39822

Added mis_ai_chat_memory table

migration#691