MindsDB Release Notes v25.6.4.0

MindsDB continues to evolve with every release—and this one is packed with enhancements designed to make working with AI and data even more intuitive. From improved integrations to smarter tools, here’s everything you need to know about what’s new in this update.

Knowledge Bases

The Knowledge Base documentation has seen a wave of improvements to make it more helpful and easier to navigate. New sections like EVALUATE,  EVALUATE KNOWLEDGEBASES and added links to AI integrations expand the platform’s usability and support. To streamline the experience, the Python SDK Knowledge Base has been updated, and the REST API endpoints for Knowledge Bases have been updated for better accessibility. Additionally, conversation history handling in both MindsDBAgent and AgentTaskManager has been enhanced, contributing to a more efficient and responsive system overall.

Recent updates to the KB tools in v25.6.4.0 bring smoother and more reliable performance. A fix was made to prevent improper handling of input in the kb_query_tool, reducing the chance of avoidable errors during queries. The base prompts from Langchain have also been overridden, improving how the Knowledge Base tools interpret and respond to user input. Additionally, an issue affecting KB evaluation for MySQL database has been resolved, making the process more efficient and dependable.

Integrations

This release brings several important backend updates aimed at improving performance, compatibility, and maintainability across the platform. The Ludwig handler now benefits from an upgraded Ray version, resulting in smoother integration and better performance. Key dependencies have been updated to keep the system aligned with the latest software standards—Protobuf has been upgraded for the Lindorm handler, setuptools and requests have been bumped in the core requirements, and the LlamaIndex handler now runs on its latest version with enhanced capabilities. 

Furthermore, Google BigQuery has been added as a supported source in the Data Catalog, expanding integration options for users. Improvements to the Agent for Salesforce have also been rolled out, further strengthening the platform’s overall functionality.

Documentation

Several key documentation underwent updates aimed at improving clarity and guidance for users. Updates include revised instructions on how to start MindsDB, the renaming of relevance_threshold to relevance, and the removal of outdated references to legacy minds endpoints. A new note has been added to help users understand authentication when using MindsDB Context Protocol (MCP), and the Knowledge Base documentation now includes updated information for the Snowflake provider.

Additionally, support for Ollama in Knowledge Bases has been officially removed, with documentation adjusted to reflect this change. These updates help ensure developers have accurate, up-to-date resources at their fingertips.

SQL Operations

We introduce several enhancements focused on performance, clarity, and broader model support. Token usage for generating schema information in agents has been optimized, helping reduce overhead and improve efficiency. A notable SQL update replaces locals() with connection.register for DuckDB table registration—resulting in cleaner, more maintainable code.

The MindsDB information schema now includes entries from the Data Catalog, making it easier for users to track and understand their data sources. Users can now retrieve catalogued metadata using SQL queries on the following new virtual tables:

INFORMATION_SCHEMA.META_HANDLER_INFO
INFORMATION_SCHEMA.META_KEY_COLUMN_USAGE
INFORMATION_SCHEMA.META_TABLE_CONSTRAINTS
INFORMATION_SCHEMA.META_COLUMN_STATISTICS
INFORMATION_SCHEMA.META_COLUMNS
INFORMATION_SCHEMA.META_TABLES

Each of these tables can be filtered using TABLE_SCHEMA = "integration_name" to return metadata for a specific integration.

Finally, the SQL model now supports Gemini 2.5, offering expanded features and greater stability for AI-powered querying.

Bug Fixes & Improvements

This release includes a wide range of improvements focused on reliability, developer experience, and system consistency. Default values and descriptions for Salesforce column metadata have been refined to deliver more insightful data. Error messages are now clearer, helping developers identify and resolve issues more efficiently. Updates to how default model configuration parameters are handled streamline the overall workflow, while table name checks in the API handler have been made case-insensitive to prevent string matching issues.

Data extraction from Knowledge Base tables is now more robust, and developers can use their Default LLM as the re-ranking model, offering greater flexibility. Issues affecting Knowledge Base insert operations—via both SQL queries and raw data—have been fixed for smoother functionality. 

The Evaluate output has been cleaned up by removing unnecessary fields, and compatibility issues with Azure OpenAI models have been resolved. Test setups now require successful completion, improving test reliability. Deletion issues in the ML engine when the Data Catalog is enabled have been addressed, and a series of additional fixes—including API SELECT queries, Snowflake special character handling, and MySQL SSL options—contribute to a more polished and dependable user experience.

Open Source GUI v5.4.22

Several usability and performance enhancements were introduced in this release. Padding was added to the chat interface when sample questions are displayed, improving visual clarity. The default model configuration process was cleaned up and corrected to ensure accurate parameters are passed during back-end API calls.

As part of the onboarding experience, the traditional data source form was replaced with a more intuitive code snippet for easier setup.

Additionally, performance and memory usage in the response system were optimized to provide a smoother and more efficient user experience.

We’re excited to see how these updates help you push the boundaries of what’s possible with AI and data. Your feedback helps us shape MindsDB—so keep it coming! Explore the new features, build something great, and reach out if you have questions or ideas. Check out the full Release Notes.