Own database · PostgreSQL and SQL Server
Connect your own read-only database to AI
Connect PostgreSQL or Microsoft SQL Server directly to DigiData AI. You choose the schemes, use an account with read-only rights and let DigiData convert foreign keys into a manageable semantic model.
Only for AI and semantic models · no write access
What is a native database connection for AI?
This feature allows you to use an existing PostgreSQL or SQL Server database as a live data source for DigiData AI agents. You don't have to have a new software connector built first or synchronize all source data to a separate reporting database. DigiData reads the allowed database structure, builds a semantic model from it and only executes controlled read queries. The connection is intended for organizations that already have their own data warehouse, reporting database or carefully managed application database.
Why this connection
Use existing data without building a new connector
A proprietary database often contains exactly the compiled and cleaned data that teams want to analyze. The direct AI connection makes that source usable with a limited technical and functional scope.
Your data is already correct
Use the existing data warehouse or reporting database in which your organization has already brought together definitions, history and source systems.
Relationships have already been established
Foreign keys and composite keys give DigiData reliable join paths. As a result, an AI model does not have to guess how orders, customers and invoice lines are related.
AI needs context
Technical tables alone are not enough. The semantic model adds names, descriptions, grain, relationships and controlled calculations to the available structure.
Read-only design
Multiple layers protect your database
Read-only is not a single check mark. DigiData combines database permissions, connection properties, query validation and explicit access per agent.
Dedicated database account
You create a separate login with CONNECT and SELECT on only the necessary schemas. Do not use an owner, administrator, or application account for the AI connection.
Provider-specific protection
PostgreSQL queries run in a read-only transaction. For SQL Server, DigiData checks effective permissions and denies accounts that can write into the allowed schemas.
Schema and table access
First you choose allowed schemas. You then determine which tables are available per AI agent. A query outside that combination will not be executed.
Checked SQL
Only one SELECT or WITH query is allowed. Writes, locks, system schemas, cross-database references and known side effects are blocked.
Set
From database to AI agent in four steps
A database administrator controls the minimum rights. A DigiData administrator then chooses the source and tables for each agent.
Create a read-only login
Give a dedicated user only connection and read rights to the schemas that are needed. Verify that the account cannot create objects or modify data.
Make the database accessible securely
Use TLS and apply firewall rules when your database only accepts known outgoing addresses. DigiData tests host, port, database and certificate settings before saving the source.
Generate the semantic model
DigiData reads tables, columns and foreign keys from the allowlist. Review the generated relationships and add business language or fixed metrics where necessary.
Give an agent access
When creating an AI agent, connect the database and select the tables that this agent can use. The data source of an existing agent cannot be switched unnoticed.
Automatic semantic model
From database schema to understandable business context
A database contains technical names and keys. DigiData converts that structure into a controlled context that an AI agent can use for queries and calculations.
Tables and columns
DigiData only reads metadata from the schemas you have allowed. Names and data types form the basis of the model. You can then fine-tune display names, descriptions, and visibility yourself.
Foreign keys first
Database foreign keys are the most reliable source for joins. DigiData also supports composite foreign keys and saves alternative relationships as inactive when multiple paths between the same tables cause confusion.
Careful inference
If a foreign key is missing, DigiData looks for clear patterns such as CustomerId, BillingCustomerId or CustomerKey. If there are multiple possible target tables, no relationship is guessed.
What else is distracted?
Columns with numeric values receive an appropriate default aggregation, while ID, code, and number columns are not automatically summed. When a foreign key points to a unique key, DigiData can use that key to describe the grain of the target table. The result remains editable: an administrator can adjust company names, definitions, relationships, hierarchies, and monitored metrics. Regeneration preserves valid manual choices.
Conscious demarcation
What the connection is and is not used for
The connection is not a general data integration. The limited scope makes it clear which components are allowed to work directly on your database.
Available
- AI agents that you explicitly link to this database.
- Retrieve metadata from allowed schemas.
- Generate and manage a semantic model.
- Run controlled, read-only queries and metrics.
Not available
- Add, change or delete records.
- Publish the database as an OData feed or Power BI source.
- Share the connection via MCP or regular integrations.
- Read schemas that are not on the allowlist.
Do you need automatic synchronization or reporting? Then check out the software integrations or read how Connects Power BI via OData.
Frequently asked questions
Frequently asked questions about AI and your own database
Please call: +31 6 15534337
Which databases can I link directly to DigiData AI?
The direct database connection supports PostgreSQL and Microsoft SQL Server. Per connection you choose one database and the schemas that DigiData AI can read.
Can DigiData AI change data in my database?
No. PostgreSQL queries run in a read-only transaction. For SQL Server, DigiData uses a read-intent connection and an account with effective write permissions is denied. In addition, the query layer only accepts bounded SELECT and WITH queries.
Do I have to copy my database to DigiData first?
No. This function reads the allowed tables directly when an AI agent executes an approved query. DigiData does store encrypted connection data and technical metadata for the semantic model.
How does DigiData create a semantic model of my database?
DigiData reads tables, columns and foreign keys from the allowed schemas. Real relationships, including composite keys, are leading. Missing relationships are only added when the table name and key name together produce one clear match.
Can I also use this connection for Power BI, OData or MCP?
No. A proprietary database connection is deliberately limited to DigiData AI agents and semantic models. For Power BI, OData, MCP and automatic source synchronization, use the appropriate DigiData solutions.
PostgreSQL and SQL Server
Do you want to know whether your database is suitable?
We'll discuss network access, minimum database permissions, and desired AI questions before you activate the connection.
