Weekly release notes

|Fix| Users could not view or update the security settings on certain clusters. Service calls to the security configuration service were being blocked by network restrictions, causing Error fetching security settings on this cluster messages and API timeouts. These calls have now been updated to work correctly thus restoring reliable access to the security settings.

|Fix| The buttons icon in Liveboard tiles appeared stretched or missing at screen widths around 1100px. This has been fixed to ensure the buttons remain visually consistent and easy to interact across all screen widths.

|Fix| Resolved an issue where TML imports failed to recognize variables, causing Row Level Security (RLS) rules to be dropped. The import process now successfully retains all RLS rules provided the associated variables are pre-created in the system.

|Fix| Query and usage activity from non‑primary Orgs was not correctly reflected in the TS: AI and BI Stats worksheet and related Liveboards. This is now fixed, and activity from all Orgs is now correctly reflected and recorded in our usage statistics, providing end‑to‑end visibility across Orgs.

|Fix| Previously, non-admin users could not see or use dbt connections that were shared with them in the dbt integration wizard. This has now been fixed to ensure that any user or group with shared access to a dbt connection can view and select it in the dbt integration wizard without requiring administrative privileges.

|Fix| Spotter would crash with an LLM service error when the MCP connector returned a large payload. A fix has been implemented that automatically writes tool responses exceeding a specific size threshold to a file. This optimization prevents exhausting Spotter’s context window by replacing the full content with a reference link, allowing the system to access the data as needed without overloading the session.

|Fix| The way in which Spotter handles geographic visualizations has been improved. By refining how the system recognizes the geo keyword, errors that previously caused charts to fail or go missing are eliminated. You can now expect more consistent and accurate geo chart generation.

|Fix| If a tool (such as the code execution tool or any other) failed, it would disrupt Spotter’s flow, making it impossible to get answers to follow-up questions. Spotter now recovers from the technical glitches, allowing you to keep the conversation going without needing to start over.

|Enhancement| Spotter 3 now features optimized data model processing and an expanded context window for more accurate, sophisticated answer generation. The agent is tuned to understand always applicable generic instructions better so important information is never omitted during complex tasks.

|Enhancement| Spotter can now process up to 10,000 rows per data query, up from the previous limit of 1,000. If a dataset exceeds this new limit, Spotter recognizes that it is working with limited data and can either refine the query for better accuracy or alert you that the insights are based on partial data.

|Enhancement| Upgraded the default LLM model for answer generation to Anthropic Claude Opus 4.5. It is served using Google Vertex AI in partnership with ThoughtSpot.||This improves the accuracy and consistency of Spotter-generated answers.

|Fix| Spotter icon text overlaps the user query when opening Spotter in a Liveboard.

|Fix|“Unknown failure” error in Spotter when asking questions in a Liveboard using specific filters. Chart axis configurations were not properly passed on to downstream services.||This issue is now fixed in 26.2.0.cl.

|Fix| Columns used in models are not indexed when Index Settings on the Admin > Search & SpotIQ page are set to Index only columns used in Models.| |Indexing now correctly triggers based on the indexing scope enabled.

|Fix| Multilingual AI Context generation: AI Context is now generated in the same language as the data model’s column names and descriptions.| |For example, if columns are defined in Japanese, the generated AI Context will also be in Japanese, regardless of the locale settings.

|Enhancement| We are expanding our Bring Your Own LLM (BYOLLM) framework to include support for Amazon Bedrock. This integration allows you to leverage Amazon’s secure, managed infrastructure to power your generative AI workflows using your own AWS credentials in your BYOLLM key.| |Anthropic Claude integration: This release provides direct access to the Anthropic Claude foundation model family. You can now deploy these industry-leading models for advanced reasoning within our platform.| |Please refer to the BYOLLM key documentation for a detailed breakdown of the LLMs required for different generative AI features in ThoughtSpot. This integration enables you to bring in Claude models using Amazon Bedrock. Individual support for AI features may be model-specific. For more information, refer to the BYOLLM documentation.

|Fix| Spotter does not use the versus (vs,versus) keyword in a comparison query to return a single visualization.

|Fix| In certain scenarios we have noticed that KPIs were not visible in Liveboard PDF exports. We have identified and fixed the underlying issue.

|Enhancement a| We have introduced new privileges for SQL views, which were not available earlier.

The required privileges are:

|Fix| Browser unresponsive error no longer appears when user tries to switch between pages in ThoughtSpot Cloud.

|Enhancement a| The HostEvent.UpdateParameters now support updating multiple parameters.

|Enhancement | We’ve enhanced our LLM configuration options to give administrators more control while maintaining a seamless experience. | a|What’s new

Impact

Action required: If you want to remain on Azure OpenAI or Google Gemini, please update your settings in the Admin panel to select your preferred provider. Otherwise, no action is needed.

Recommendation: We strongly recommend staying on ThoughtSpot Enabled: Auto Select for the best Spotter experience which will provide access to cutting-edge models as they’re released.

|Fix| We have transitioned the Spotter Agent back to GPT-4.1 as the default model following successful testing and validation of tool call behavior. This adjustment addresses the earlier degradation observed and restores the intended experience.

|Fix| “Something went wrong” error no longer appears while generating answers in Spotter with GPT-4o model. |Fix| "Why" questions are now answered correctly on ThoughtSpot Cloud instances deployed on Google Cloud Platform (GCP).

|Fix| Spotter Agent summary displayed after answers is now more concise.

|Fix| Microsoft XLSX files received in email from a scheduled Liveboard can now be opened on a computer running Microsoft Windows.

|Fix| approx_unique_count formula now works.

|Fix| GPT-4o is now the default model due to degradation in responses from GPT-4.1. Minor changes in communication style are expected, but no functional impact for users. Customers using the Google Gemini model family continue to remain on Gemini.

|Enhancement| Translates relative date phrases like "last Monday with dates" into explicit calendar dates.