AI Assistant settings

Use this page to configure the general behavior of AI in DBeaver.

AI Assistant setup

Note

AI Assistant is turned on by default. If it was disabled earlier, you can re-enable it in Window -> Preferences -> AI by selecting Enable AI integration.

To activate AI Assistant, open Window -> Preferences -> AI and select your provider from the Service dropdown menu. Next, set up your provider.

Chat Settings

For configuring AI Chat settings, navigate to Window -> Preferences -> AI -> Chat Settings.

Setting	Description	Default
Use stream mode	Enables data to load and display progressively as it’s received from the server, instead of waiting for the entire response. This improves responsiveness and reduces memory usage when working with large result sets. Available only if supported by the selected AI provider.	on
Days to keep chats	Sets how many days chat history is stored locally. Older chats will be deleted automatically.	7
Silence threshold	Defines the minimum sound level that’s considered speech. Lower values make recognition more sensitive to quiet voices, but may include background noise.	0.3%
Silence timeout	Sets how long to wait (in seconds) after silence before ending the transcription.	20 sec
Audio max duration	Limits the maximum length of recorded audio for transcription. Longer recordings are automatically trimmed.	60 sec

Important

If you add filters in the Database Navigator, only the filtered tables are sent to the AI. The same applies in the Data Editor — if you filter tables, they won’t appear in completion or be validated correctly in queries, because if they’re not in the navigator, there’s no metadata about them.

AI Configuration

For configuring specific settings, navigate to Window -> Preferences -> AI -> Configuration.

General

Setting	Description	Default
Language	Choose the language for AI responses. This affects the language used in answers, explanations, and code comments.	English

Info

Besides the listed options, you can also type in a language that's not included in the dropdown menu.

SQL Completion

Setting	Description	Default
Include source in query comment	Shows your original request above the AI-generated query in the SQL Editor.	Off
Execute SQL immediately	Runs the translated SQL query immediately after generation.	Off

AI query execution

These settings control how AI-generated SQL queries are executed. You can choose to run them immediately, show a confirmation prompt, or disable autocommit so changes must be committed manually.

Setting	Description	Options	Default
Select	Controls execution of SELECT queries.	- Execute immediately - run SELECT queries without confirmation. - Show confirmation - ask for confirmation before executing.	Execute immediately
Modify	Controls execution of data modification queries (INSERT, DELETE, UPDATE).	- Execute immediately - run modification queries without confirmation. - Show confirmation - ask for confirmation before executing. - Disable autocommit - turn off autocommit so changes require manual commit.	Show confirmation
Schema	Controls execution of schema modification queries (CREATE, DROP, ALTER, etc.).	- Execute immediately - run schema modification queries without confirmation. - Show confirmation - ask for confirmation before executing.	Show confirmation
Other	Controls execution of queries DBeaver cannot classify as valid SQL (including NoSQL queries).	- Execute immediately - run unknown or unclassified queries without confirmation. - Show confirmation - ask before executing queries not recognized as SQL.	Show confirmation

Tip

For the AI command, using Show confirmation is strongly recommended. It lets you preview the generated SQL in a dialog before execution, so you can review or cancel risky changes.

Send database structure

Setting	Description	Default
Send column data type information	Sends column data types (like INT, VARCHAR, etc.) to the AI provider. Improves completion quality, but consumes more tokens.	On
Send object description	Sends object description to the AI provider. Improves completion, but may consume significant amount of tokens.	Off

Automatic prompt reduction

When sending database metadata and sample data to the AI provider, the amount of information is limited by the selected context size. If the content exceeds this limit, DBeaver automatically shortens the prompt in two steps:

1. Removes extra details (data types, keys, constraints, sample data) if all are enabled and the size is too large.
2. If it’s still too large, trims the snapshot until it fits the context limit.

Tools and MCP

DBeaver supports two types of AI tools: built-in internal tools and external tools from connected MCP servers.

Info

For information on using tools, see AI Tools & MCP.

Prompts

Available since 26.0

Use this page to customize the built-in instructions that DBeaver sends to the AI provider for specific features. These instructions are applied automatically when you use the related AI feature.

Open it in Window -> Preferences -> AI -> Prompts.

Note

User-defined instructions are added on top of default instructions. Default instructions can’t be changed.

Prompt	What it affects
Analyze Data Compare results	Explains differences found in Data Compare results.
Describe database objects	Returns detailed database object description and its role in the database schema.
Explain SQL queries	Explains the structure and business value of the SQL query, including involved entities and relationships.
Explain SQL query execution plan	Explains the SQL query execution plan and helps understand how the database runs the query.
Fix SQL query errors	Fixes SQL query errors using the query text, the error message, and environment information.
SQL query completion	Performs SQL query completion based on the query, comments, and surrounding queries.
SQL query generation	Generates SQL queries from a natural language description.
Smart mapping	Finds the closest matching data types when converting columns between source and target databases.

Troubleshooting

AI chat shows N/A on Windows Server

If AI chat displays N/A and the error log contains unhandled event loop exceptions, the cause is likely a missing WebView2 component. For information on how to fix it, see Troubleshooting WebView2 issues.