JetBrains DataGrip Introduces AI Agents for Enhanced Database Management and Analysis

AI agents can be granted specific permissions that allow them to read database schemas without accessing sensitive row data. This ensures a secure operational environment by separating metadata access from actual data access.

These permissions specifically cover metadata such as schema names, object names, and the overall structure of the database. The host emphasized, "This specific permission doesn't allow access to actual table data. Instead, it allows the agent to retrieve metadata such as schema names, object names, and the overall structure."

The AI agent can effectively distinguish between core database tables and extraneous tables, such as those added from a sandbox environment. This capability helps maintain database integrity and organization by highlighting non-standard elements.

The process involves the agent analyzing the entire schema, identifying any tables that are outliers, and then providing a clear reasoning for why those specific tables are classified as such. The host noted, "The agent quickly recognizes that the core schema is the Tequila DVD rental database and then it analyzes the rest to find the outliers."

Before executing any 'drop' commands, the AI agent performs crucial safety checks to verify dependencies. This includes checking for views or foreign keys that might rely on the tables slated for deletion, preventing accidental data loss or integrity issues.

Once the agent has presented its generated SQL statements for the cleanup, final execution requires explicit user confirmation. This two-step process ensures that potentially destructive operations are reviewed and approved by a human operator. As the host explained, "Notice how the agent performs safety checks first. It replies before dropping let me check whether anything else like views or foreign keys depends on these tables."

Users have the ability to prompt the AI agent to summarize all past queries that have been executed within a specified timeframe. This feature provides an audit trail and helps users understand the agent's actions and decisions. The host mentioned, "After a moment of thought, the agent provides a comprehensive list of those queries and even politely asks for feedback."

By typing the '@' symbol, developers can directly link specific database objects within the chat interface, triggering code completion. This method provides an intuitive way to explicitly tell the AI agent which database object the user is inquiring about or wants to interact with. The host stated, "I select the at DB object identifier and code completion kicks in. This is the easiest way to explicitly tell the agent exactly which database object you're asking about."

For the agent to execute routines for analysis, particularly those with potential internal side effects, it requires 'modify data' permission. This precaution ensures that the agent has the necessary authorization for deeper interactions while maintaining security protocols.

The AI agent successfully identified MySQL-specific syntax that was erroneously left within a PostgreSQL procedure. This capability highlights the agent's ability to recognize and flag database-specific syntax errors across different SQL dialects.

The agent explained that these logic issues likely stemmed from an improper conversion during the migration process from the original Tequila database to PostgreSQL, where the MySQL-specific `CASE` statement was not correctly adapted. The host explained, "It detected MySQL specific syntax within this PostgreSQL procedure. It explains that this logic was likely inherited from the original Tequila database and was never properly converted to a case when statement during the migration to Postgress."

DataGrip now allows users to attach local folders, enabling the AI agent to analyze raw datasets such as CSV files directly within the integrated development environment (IDE). This expands the agent's utility beyond live database interactions to include local data preparation and quality checks.

The AI agent is capable of identifying common data quality issues within these local files, including hostile header names, missing primary keys, and various logical semantic errors. The host affirmed, "It confirms that there are indeed data problems in housing 2.csv. ... The first issue it highlights relates to physical lines and line endings."