Weekly release notes

Fix

Non-admin users in RBAC clusters were not visible or shareable in the dbt connection share modal. This issue is now fixed.

Fix

Previously, Spotter would select a single data source at the start of a conversation and remain locked to it for all follow-up questions. Spotter can now answer questions that span multiple data sources within a single conversation, which was not possible earlier.

Fix

Previously, Spotter would give the option to download a CSV file which was not actually available. This has now been fixed and Spotter no longer mentions the file when it is not available to download.

Enhancement

Spotter agent is now more intelligent with better context propagation to the Answer generation tool.

100% relevant context (coaching/instructions) is considered by the agent resulting in better consistency.

Addresses an inconsistency observed between Spotter 3 and previous versions of the Spotter agent.

Fix

Liveboard export APIs did not correctly apply filter_rules embedded in tokens generated by /api/rest/2.0/auth/token/custom, and would return blank pdfs, or visualizations not showing any data. This has now been fixed and exports correctly honor filter_rules in all supported formats, ensuring secure data export for embedded applications that rely on mandatory token filters.

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

Fix

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

Enhancement

Enhancement

We’ve enhanced our LLM configuration options to give administrators more control while maintaining a seamless experience.

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.