MindsDB Release Notes - July 2025

We’re thrilled to introduce the latest MindsDB release, bringing a host of enhancements across the platform to improve performance, usability, and integration. This update includes major improvements to Knowledge Bases, streamlined agent functionality, expanded model support, and deeper documentation—plus the much-requested addition of Dark Mode in the GUI. 

Whether you're building with large language models, integrating AI into your existing data stack, or deploying real-time insights, these updates are designed to simplify your workflow, boost efficiency, and accelerate innovation. Here's a look at what’s new.

Knowledge Bases

This month’s releases brought significant enhancements to the functionality and performance of Knowledge Bases, including improvements to search and evaluation features, metadata updates, support for algebraic operations, and better handling of unknown metadata keys through filter restrictions. These updates collectively enhance search accuracy, metadata management, and overall usability. Additionally, the creation agent now supports wildcards in table and KB names, offering greater flexibility and efficiency when defining database schemas.

Knowledge Base error messages have been overhauled to provide more informative and helpful prompts, improving user guidance during troubleshooting in v25.7.2.0. Documentation has also been updated to include detailed instructions on the from_sql feature, and unit tests have been introduced for the KB Evaluate function to ensure consistent and reliable performance.

Embedding providers are now whitelisted to ensure a more controlled and secure environment for Knowledge Base usage. Additionally, new documentation has been added, offering guidance on knowledge base models, SQL algebra, and practical examples within KB query documents.

A new preprocessing field was added to the Knowledge Base Controller, streamlining the handling of preprocessing tasks. Knowledge Base algebra was also improved with support for concatenating AND values for content, enabling more expressive queries. Updated documentation now details the models supported by Knowledge Bases and includes a comprehensive data ingestion diagram. To ensure ongoing stability, a Knowledge Base integration test has been implemented within the continuous integration pipeline.

Integrations

A bug related to subscriptions in PostgreSQL has been resolved, improving the reliability and performance of real-time data syncing and publishing. Additionally, key fixes to the BigQuery Data Catalog enhance integration with Google BigQuery, ensuring smoother and more consistent data access.

Support for the Google Gemini model has been added to the TO_MARKDOWN() function, enabling users to process a broader range of data in structured formats. The function now also supports XML and Nessus files, further expanding its versatility across various file types. A compatibility issue with the Salesforce Data Catalog has been resolved, restoring its performance and integration. Additionally, the configuration update behavior has been adjusted to allow overwriting default model parameters, giving users greater control and flexibility in model setup.

Enhancements to Salesforce metadata handling, resolution of Snowflake Data Catalog stats errors, and improved support for lowercase table name specifications contribute to more robust data connectivity. Slack integration now handles list pagination more effectively, while a new public_url parameter simplifies S3 integration. YouTube integration has also been upgraded with search functionality, and the UV dependency has been downgraded to optimize overall performance.

 A fix to the staging integration tests ensures more reliable and consistent integration performance across environments. The HubSpot API client has been updated, strengthening the MindsDB-HubSpot connection, while improvements to the BigQuery connection method offer a smoother setup process. Updates to agent syntax and the LittleLM handler enhance usability, and outdated commands have been replaced to ensure more robust and reliable agent interactions with API endpoints.

Bug Fixes and Improvements

This release brings key backend improvements for greater stability and performance. Agents now handle improperly quoted table names more reliably, while updates to the Google Gemini model examples improve usability. The huggingface_handler has been temporarily disabled to optimize system resources.

Security has been strengthened with safeguards against data exposure, and the MCP dependency has been upgraded to v1.10.1 for improved stability. Agent streaming methods and A2A server serialization have been refactored for better efficiency, along with a fix to the A2A configuration path.

The vectordb handler bug has been fixed, and updates to Docker and local installations simplify API deployment. The Llama Index handler was upgraded, and support for structured JSON responses from LLMs has been added for more consistent output.

Agents now support Data Catalog loading through refactored logic, while fixes to NumPy truth value evaluation and partitioning errors improve reliability. Additional enhancements include alpha blending for image processing, improved Uvicorn logging, deprecated method replacements, and more flexible job queue management via the GET /tree endpoint.

Documentation

This release brings a wide range of documentation improvements aimed at enhancing clarity and usability across the MindsDB platform. Key updates include expanded guidance on Docker installations, model types, logging configuration, and the Evaluate feature, along with reorganized Knowledge Base docs for easier navigation.

New instructions cover using url_file_upload in config.json, refined agent model parameters, and LiteLLM integration. Examples for adding data to agents have also been added to simplify implementation.

Additional updates include configuration steps for the A2A host, revised REST API guides, and enhanced documentation for to_markdown, the quickstart tutorial, and the Data Catalog. Deprecated references and endpoints have been removed for a cleaner, more streamlined experience.

OS GUI

MindsDB has introduced Dark Mode in its GUI, offering users a sleek, visually comfortable interface optimized for low-light environments and extended work sessions.MindsDB's configuration has been enhanced with a fix for a bug affecting Google embedding models, ensuring smoother GUI experience. The DESCRIBE MODEL syntax has been improved for greater clarity, and the redundant 'view' prefix in view nodes has been removed to eliminate related errors. Additionally, the Hello World script has been updated to align with MindsDB’s core workflow—Connect, Unify, and Respond—providing a clearer onboarding experience for new users. 

More improvements have been made to the GUI’s datasource connection forms and highlighted keywords. Improvements have been made to the Azure OpenAI integration for a smoother and more reliable experience. The “Test and Save” feature now allows users to validate LLM connections before finalizing setup, while query text handling has been enhanced to ensure cleaner, more accurate API requests. Additionally, specifying schema details in the connection settings has been improved for greater precision and ease of configuration.

You can check out the full changelog for our releases:

• Release v25.7.1.0

• Release v25.7.2.0

• Release v25.7.3.0

• Release v25.7.4.0

This release marks another step forward in making AI more accessible, efficient, and deeply integrated with your existing data infrastructure. From enhanced model capabilities and agent flexibility to expanded documentation and UI improvements, we’re committed to helping you build smarter, faster, and with confidence. As always, we welcome your feedback and look forward to seeing what you create with the latest from MindsDB. Stay tuned—there’s even more to come.