Skip to content

Release Notes

February 2026

User Experience

  • Support for Activation Key, App Key, and User License Expiry

    Introduced proactive early license expiry notifications along with activation key support for VM extension-based license management.

  • Support for MCP Server Deployment on SaaS

    Enables deployment and management of the MCP server in a SaaS environment

  • Value Dashboard for On-Prem Deployments

    Enables on-prem users to monitor and analyze key value metrics through a dedicated dashboard.

  • Support for Training Mode in On-Prem Deployments

    Allows on-prem deployments to use training mode for model learning, testing, and performance improvement.

  • MCP Configuration Export via Connection Dashboard

    Enables users to export MCP configuration details directly from the connection dashboard for easier management and setup.

Semantic Enhancements

  • Personalization Control for Edited Query Context

    Enable exclusion of data-model–modified saved queries from LLM context to prevent bias introduced by altered SQL.

  • Include all values for categorical columns

    Categorical column filtering now supports full column value inclusion to prevent hallucinated values.

  • Enhanced Ambiguity Detection and Surfacing

    Expand ambiguity alerts beyond tables and dimensions to include measures for comprehensive semantic conflict detection.

Platform Capabilities

  • Source Links for Saved Queries and Query References

    Attach source name and source url to each saved query to clearly indicate its origin and external reference.

  • Recent Questions Tab with Negative Question Alerts

    Added a dedicated Recent Questions tab with negative feedback alerts and actions to ignore or fix the data model.

  • Ambiguity Alert Management

    Enables users to explicitly mark ambiguity alerts (column, table, and measure conflicts) as Resolved (via rule enhancement) or Ignored (if non-impacting).

  • Value Page Enhancements with Parameter Editing

    The value page is now customizable, allowing users to adjust parameters to accurately reflect their specific needs.

  • Support for Table Description Rules in Visualizations

    Enable users to define rules that generate consistent table descriptions in visualizations.

  • Column Description Rules for Tabular Section

    Column description and rule details are now displayed in separate columns for dimensions, simple measures, and custom measures to improve clarity.

  • Annotation Rule Validation in Tabular Section

    Validate column description rules against saved queries before global application, and extend validation to tabular measures beyond visualization flows.

January 2026

User Experience

  • Form to add annotation rules

    Defining annotation rules is now easier and safer. A new guided form helps users configure rules with enforcement, priority, and hierarchy controls, reducing errors and improving governance.

  • Filter dropdown for charts

    Charts now support filter dropdowns, enabling more customizable and exploratory data analysis directly from visualizations.

  • Toast enhancement for the portal

    Toast messages across the portal have been improved to provide clearer, more actionable feedback, making system responses easier to understand.

  • BYOK cost estimation

    Users can now view cost estimates for Bring Your Own Key (BYOK) usage directly in the portal, improving transparency and cost planning.

  • Flight saved queries in bulk

    Run saved queries in bulk to quickly evaluate outcomes across newly trained models or configuration changes—ideal for regression testing and model comparison.

  • The data model is editable

    Users are now allowed to modify SQL queries according to their preferences to get more personalized and relevant results.

Semantic Enhancements

  • Support for join relationships over composite keys

    Semantic models now support join relationships across multiple key columns, enabling more accurate modeling of complex schemas.

  • Elevate User role to view semantic visualizations

    The User role can now view semantic visualizations without requiring elevated permissions.

  • Surface column ambiguities in the semantic graph

    Column ambiguities are now surfaced directly in the semantic graph, making potential conflicts easier to identify and resolve.

  • Validate annotation rules

    Column description rules are validated against a saved query before enforcement, allowing users to verify the impact and prevent unintended behavior.

Platform Capabilities

  • Added new BYOK model support

    Added support for o4-mini (on-prem and SaaS) and haiku (SaaS) to improve performance and deployment flexibility.

  • More efficient log rotation

    Preserves essential logs to ensure no user data is lost.

  • Saved queries marked as stale, with an option to recover

    Saved queries that become stale are now clearly flagged, with an option to review and restore them.

  • Used saved queries for query rewriting

    Saved queries can now be reused to automatically rewrite new queries, improving consistency and efficiency.

  • Support for Oracle views

    Added support for Oracle views, making it easier to work with existing Oracle database schemas.

  • API agent with backend flow

    API agent enables backend token generation, simplifying secure app-to-app authentication flows.

  • LangChain upgrade

    Dependencies have been upgraded to improve performance, stability, and long-term compatibility.

  • Query history feedback

    Query history now provides feedback on measures used, making it easier to understand and analyze past queries.

  • Column description in tabular format for measures

    Structured column rules (enforcers, priorities, overrides) guide accurate column interpretation and selection.

  • Support sub-keys and URLs in model level

    Users can now configure different keys and URL paths per model, enabling more flexible multi-model setups.

  • Upload the query context CSV for Cassandra

    For Cassandra dialects, Users can now upload a query-context CSV, improving query grounding and execution.

  • Multi-Join Support in Join Relationship feature

    Multi join feature now allows users to create multiple joins between two tables.

December 2025

User Experience

  • Feedback Signals

    Capture thumbs up / thumbs down feedback directly on insights.

  • Guided Tour Control

    Configuration to opt-out of the guided tour during login.

  • Improved Results Tables

    Cleaner tabular results with improved clarity and ease of use.

  • Analysis & Discovery

    Structured analysis suggestions, related questions, and chronological query breakdowns for better transparency.

  • UI Consistency

    Unified toast messaging and visible product version in the side panel.

  • Training Status Visibility

    Clear visibility into training states (stale, in-progress, completed) with intuitive color coding.

  • Product Version Visibility

    The active product version is now displayed in the side panel for quick reference.

Personalization

  • Time Zone Awareness and UTC Timestamp Standardization

    Time zones are now supported across Personalization and Training, with all timestamps normalized to UTC.

  • Casing Integrity Validation

    Added safeguards to ensure casing are preserved correctly across tables and charts.

Semantic Enhancements

  • Enhanced Semantic Layer

    Enhancements to dimensions, measures, and join handling for more accurate and expressive queries.

  • Column Description Rules

    Introduced annotation rules to standardize and improve column-level descriptions.

Search Quality

  • Smarter Filter Parsing

    Commas in filter predicates are now preserved, improving accuracy in complex conditions.

  • Improved Search Suggestions

    More relevant and context-aware suggestions to help users find the right answers faster.

Platform Capabilities

  • Expanded Cassandra Support

    Added support for multiple tables within Cassandra for broader query scenarios.

  • Oracle Thick Client Mode

    New support for Oracle database thick mode clients for enhanced compatibility.

  • Database Renaming

    Databases can now be renamed directly from the Settings page without reconfiguration.

  • Schema constraints based Join detection

    Join relationship population based on constraints defined in DB

November 2025

Core Capabilities

  • Cassandra Connection

    Connect to your NoSQL data for quick and easy search in Tursio — provide the graph schemas and start running queries in natural language.

  • Oracle Connection

    Connect to your transactional data in Oracle and search that in a single, unified place at Tursio.

  • Power BI DAX Integration

    Import your Vertipaq data model directly from Power BI to leverage existing DAX expressions for richer, context-aware searches in Tursio.

Search Experience

  • Smarter Search & Analysis

    • Split view for comparing results and suggestions side-by-side.
    • Explore related or follow-up queries effortlessly.
    • Quick query actions to run, copy, or export faster.

  • Better Organization & Collaboration

    • Separate Save and Like actions for personal vs. team use.
    • Centralized Saved Queries page for easy access and management.

  • Refined Interface & Experience

    • Cleaner dropdowns, tooltips, and contextual indicators.
    • Full mobile support with complete feature parity.

Personalization

  • Query modes

    Define default query behaviors tailored to each user type for more consistent outcomes.

  • Data controls

    Manage data size limits, formatting rules, and export permissions to ensure governance and compliance.

  • Layout preferences

    Customize how results appear — choose the visibility and order of a text, tables, and visuals.

  • Custom instructions

    Add notes or constraints to guide how your data is interpreted and presented.

Privacy & Security

  • Authorization agent

    Admins can now connect their Microsoft Entra ID app registration to enable secure authorization.

  • HTTPS Support

    Bring your own certificate for secure data traffic.

October 2025

Core Capabilities

  • Better Dimension Prediction

    Tursio predicts the dimensions to answer the questions from, and now it has significantly better predictions for even better accuracy.

  • Expressive Querying Mode

    Tursio queries structured data accurately, yet the responses need to be tailored to the user questions. We now perform additional rewriting steps to combine accuracy with expressivity.

Training Enhancements

  • Column, Table Descriptions

    Tursio now understands the domain vocabulary better. This typically means how people refer to things but also includes the meaning of various terminologies involved. Specifically, Tursio automatically generates descriptions of dimensions and measures as part of training. Users can modify and extend those descriptions anytime for better domain contextualization.

  • Better Join Inference

    Tursio detects join relationships as part of training, however, there could be both false positives and false negatives. We now extend the join inference algorithms with more semantic power to combine the deterministic inclusion dependencies with better semantic understanding.

Search Experience

  • Split View

    We have completely revamped Tursio's look and feel. First of all, the search results are displayed in a Perplexity-style rolling view with the search box moving to the bottom for follow-up questions. Secondly, users can analyze any given result further in a side panel, which acts like a side chat to dig deeper into the results. And finally, we have improved the navigation for better usability.

Deployment

  • On-premises

    Tursio supports on-premises deployment with several customizations, including the choice of stand-alone VM versus AKS cluster, option for certificate-based authentication, and support for private LLM keys and even custom LLM deployments.

  • Azure Marketplace (Preview)

    Tursio is now available as a VM extension on Azure Marketplace. This means users can spin up an Azure VM and try out Tursio in no time (only need to contact Tursio for an activation key).

September 2025

Core Capabilities

  • Cassandra Support

    Tursio already supports relational databases, and today we are extending support to a Cassandra NoSQL database. This allows users with application data in a Cassandra database to query them easily using Tursio. Note that we support both standalone Apache Cassandra and Azure Cosmos DB for Apache Cassandra.

  • Synapse Support

    Tursio now supports Synapse for querying in natural language. Microsoft's users managing Synapse clusters can now use Tursio's AI-powered query processing. This widens our support of the Microsoft stack with support for five backends, including SQL Server, Azure SQL, Azure Synapse, Azure Fabric, and Cosmos DB (Cassandra API).

  • Tailored Responses

    Tursio generates text response to user questions on structured data. Today we extend the response to tailor them to the specific tone and verbiage. This means that the same piece of a structured query result will have different responses for a "show" versus an "analyze" question. Behind the scenes, Tursio identifies the nuances in question formulation and tunes the prompts to generate tailored responses, including the type of description, the level of details, and the length of the response.

Model Training

  • Incremental vs Full

    Tursio trains a large number of small models, coined micromodels, on the underlying data sources. Users can further tune the inferred semantic graph and retrain anytime. Now we are making the training process modular, giving users the option to train "incrementally", which applies only the latest user feedback to the existing model, or "fully," which retrains the model from scratch using all available data.

  • Training Settings

    We have also made training settings more explicit, including the type of training by default (incremental or full), generating counts of IDs by default, and ignoring PII columns by default. Even though some settings are for informational purposes, users will be able to change all of these default settings in the future.

Visualization

  • Chart Customization

    Tursio generates visualization based on the data, and today we are giving users more options by generating multiple visualizations for each chart type (bar, pie, time series). Users can see the default visualization or change to any of the available alternatives.

  • Export Customization

    Tursio allows exporting the result into a PDF. We now further allow users to tune the EChart visualizations (e.g., pan, zoom) and export the final appearance to PDF.

Embedded Deployment

  • Minified Tursio

    Tursio provides a comprehensive querying interface for business users. We now allow enterprise customers to embed the Tursio experience into their existing applications. The goal is to customize the experience while speeding up the deployment process.

August 2025

Core Capabilities

  • Excel Support

    Tursio already supports all major data warehouses, and today we are adding support for Excel files. Business users frequently deal with Excel files exported from backend systems and even prepared manually. Tursio now allows training specific file formats (contact the team for this), and then upload files with those formats. Once uploaded, users can query them like any other database backend.

  • PostgresSQL

    PostgresSQL is one of the most popular open source databases, and we are now adding support for PostgresSQL as a data source. With this, users can ask questions on both their operational data in Postgres and their warehousing data (e.g., Fabric, Snowflake, Databricks, etc.) at the same place.

Semantic Usability

  • Node Search

    The Semantic graph visualizes the various elements in the underlying database. We have now enabled searching the various elements via a single search box. Users can find all relevant elements, e.g., "ids", and pan the visualization while going through them. Clicking on any search result zooms the visualization to that element.

  • Add/Update Visual Elements

    Users can now add or update elements in the semantic graph. This includes adding new simple measures, custom measures, and join relationships, and modifying properties of any element. The system detects any changes in the semantic graph and prompts users to retrain the model for the changes to be effective.

  • Dimension, Measure Aliases

    The final update in the semantic graph is the introduction of aliases for dimensions and measures. Users can add multiple aliases for each of the dimensions or measures and refer to them when asking questions in natural language. This is helpful when users have their own domain vocabulary that they want the AI to understand.

More Privacy

  • Bring-your-own-key (BYOK)

    Privacy is a key challenge for most organizations and the bar is higher with AI. To support enterprise-grade privacy requirements, we now support bring-your-own-key (BYOK) for Azure OpenAI. This means users can provision their own Azure OpenAI resource and use that with Tursio, i.e., all LLM prompts and responses are completely private and could be monitored/audited within the organization.

  • Personal Identifiable Information (PII)

    Exposing sensitive data to LLMs is risky and could lead to unintended consequences. Tursio now identifies potential PII columns upfront and excludes them from any training or query processing. Users can anytime validate the inferred PII columns or add new ones. The goal is to provide clear guardrails around what the LLMs are going to be consuming.

July 2025

Introducing Agents

  • Search Agent

    Tursio pulls data from structured databases and leverages reasoning models to generate the responses. However, pulling large datasets makes it hard to reason in a single prompt. Therefore, we are introducing a search agent to stage large results into a vector store, use search tool to drill down to the relevant information within the result, and finally generate the answer – all automatically and with a single user clicks.

  • Perplexity Agent

    Tursio excels at pulling enterprise data and generating answers on top of it. However, it lacked external data sources, e.g., market events, industry standards, weather patterns, and so on. We are now introducing a Perplexity agent that integrates with Perplexity to pull external context. Users can describe what context they want to pull and save it. Tursio will automatically pull both the enterprise data and the external context before generating a response based on that.

Better Search

  • More Flexible Verbs

    People ask questions in a variety of ways, e.g., “show total revenue” vs “describe total revenue”, that adds nuances to how the response should be crafted even though underlying data remains the same. We have incorporated 25+ verbs in the auto-mode to capture these variations and help users express their intent better.

  • Information Panel

    Tursio pulls data from structured databases and frequently limits the results that are too large to reduce the load on the underlying database. We have now added an information panel to inform users of how the result was truncated, allowing them to make informed decisions.

  • Apache ECharts

    Tursio has long relied on Plotly Dash to generate visualizations. While they simplify the Python stack, Dash also has its limitations. We are now moving to a more modern Apache EChart for enterprise grade visualizations, including richer charting and better customization.

  • Customizable Tables

    We have revamped our table layouts with better customizations, including table search, column projection, filtered export, as well as better responsive and formatting behavior.

Semantic Graph

  • Visualization

    Tursio generates the semantic model automatically as part of the training. We are now also introducing automatic visualization of the semantic graph. Users can understand how entities are connected, interpreted as dimensions and measures, and see the custom measures that get generated automatically. Users can also click on any part of the graph to inspect or modify the details.

  • Training Feedback

    The semantic graph is customizable with user feedback, and all of that is fed back to the training behind the scenes. This means as users improve their semantics, their ability to ask questions more simply or accurately gets better over time.

Management Capabilities

  • License Keys

    Tursio already supports both Microsoft and Google login. We are also adding the ability for tenant owners to generate license keys for specific users. This is helpful when Tursio is deployed in an air-gapped environment with no access to OAUTH servers. Tenant owners can manage all their licenses with complete control over who can see what.

  • Change Detection

    Tursio detects changes in the semantic model or in the querying scope and alerts the admins to trigger the training again. This is helpful to keep track when multiple people in a team are collaborating to improve their collective search experience.

  • Manage Saved Prompts

    Tursio allows saving prompts for giving better starting points to non-expert users. We now allow the tenant owners to manage the lifecycle of saved prompts and purge them when they are no longer needed.

  • Expanded Query History

    Tursio gives tenants admins the visibility into query history. We have now expanded the query history to also include follow-up reasoning questions. To make it self-explanatory, admins see both the parent and child questions together.

June 2025

New Capabilities

  • Multi-dimensional Clauses

    Auto-prompting mode is built for non-experts to construct queries in natural language while automatically translating them to valid SQLs behind the scenes. Based on the customer feedback, we are now expanding the “Auto” mode beyond single filter, group-by, and aggregate to support any number of clauses across multiple dimensions. This means users can apply multiple filters (conjunctions across columns or negations on the same column, and disjunctions of equality predicates on the same column), multiple group-by along different dimensions, and multiple aggregate functions.

  • Measure Feedback

    Users have domain metrics that they want to incorporate, and we are now introducing full flexibility in managing them. Users can see or edit the measure columns and default measures created by Tursio, as well as create new ones. They can also incorporate domain ontology associated with those metrics. Tursio takes care of processing future queries with the new feedback.

  • Saved Results

    Users want to discover results quickly and revisit them again. To make this simpler, we now surface the upvoted results on the homepage. Users can scroll through these results and see them again with a single click. Old or downvoted results get automatically purged over time.

  • Prompt History

    Prompts is a creative process, and we now show the recent prompt history with a single click on the history icon within the search box. This allows users to reuse older prompts or extend them over time without writing from scratch.

  • Databricks Support

    Finally, based on the requests from our customers, we have added support for Databricks. Once the connection is added, the rest of the Tursio experience remains identical, i.e., a single box for search and analyses over all data sources.

Better Usability

  • Recommended Questions

    Given that Search is an iterative process, we now recommend questions upvoted by other users that are relevant to the input question. The goal is to help in the exploration process and discover questions that are related and found useful by others.

  • Sharing “Auto” and “Research” Questions

    “Auto” and “Research” mode questions have different visual interfaces; one is built using well-formed query fragments while the other gets interpreted into well-formed query fragments. We now preserve their respective visual interface when the questions are saved (upvoted) or shared (via links).

  • Cues for “Analyze” Mode

    “Analyze” mode provides powerful reasoning capabilities on top of the retrieved data. We now display visual cues to hint users the various kinds of followups that are possible.

  • Guided Tour

    Finally, we are adding a guided tour for first users, helping them to navigate through login, dataset selection, and search functionalities. Users can also revisit the tour anytime.

Improved Performance

  • Streaming Mode

    Getting responses quickly is paramount, and we have now enabled “streaming” mode to reduce the time to first response fragment from 10 to 12 seconds down to under 3 seconds, a 4x speedup for better user experience. Tursio further parallelizes the rendering of other components as much as possible.

  • Tuning Prompt Size

    Large results from structured data lead to large prompt sizes that slow down further reasoning. We have tuned the prompt sizes with more compact data formats and better optimized instructions to reduce the overall prompt sizes by 30–60%.