Data Governance Agents
DIG is the product name for Data Governance Agents. The module in DAM is called Intelligent Data Governance. DIG adds AI assistance to DAM governance workflows and organizes governed asset semantics and knowledge into context for AI applications. Existing governance modules, approvals, and data permissions continue to apply. Without the appropriate license, the AI conversation icon and related pages are unavailable.
How DIG supports governance
DAM stores metadata, standards, quality rules, security classifications and levels, and catalog assets. DIG uses configured capabilities to propose changes against a defined objective. Confirmed results enter the relevant DAM module. Prepare accessible objects, business rules, and permissions before starting.
Open AI from a data source, table, or asset detail page for a specific task. For ongoing governance across a business domain, create a data intelligence space to organize assets, documents, and workflows.
Choose a starting point
| Goal | Entry point | Guide |
|---|---|---|
| Complete a table's business names and descriptions | DAM AI conversation or object detail page | Quick start |
| Govern a business domain over time | Data intelligence space | Create a space |
| Assess governance, run workflows, and publish context | Space governance lifecycle | End-to-end governance |
| Make governed knowledge available to AI queries | Context Management and Publish | Validate and publish context |
| Review standards or classification proposals | Approval requests | Approvals and collaboration |
| Reuse governance actions and workflows | Configuration Management > Governance Sprites | Sprites and Routines |
| Configure models and tools | Configuration Management | Initial setup, model services, administration |
Two phases of a space
DG by AI uses AI to assist governance. Add assets and documents, run the space agent or Routines to enrich metadata and organize standards, quality, security, and assets, then review the results.
DG for AI makes governed knowledge available to AI applications. Organize semantics, standards, lineage, and documents into trusted context. Validate it before enabling queries through a space Agent or MCP. Publishing context does not itself deliver underlying business data.
The workspace presents four steps: Start > Data Governance > Context Management > Publish. Determine completion from actual results rather than conversation length or execution counts.
Core concepts
| Object | Purpose | Relationship |
|---|---|---|
| Agent | Understands goals and coordinates execution | Uses models, tools, Skills, Sprites, and Routines |
| Agent template | Stores a configuration snapshot that cannot execute directly | A published template creates an independent space instance; later updates do not overwrite it |
| Skill | Packages reusable instructions and execution methods | Guides a task; it is neither a data source nor an approver |
| Governance Sprite | Performs one bounded, atomic task | Defines inputs, tools, and outputs; can be tested independently |
| Routine | Combines Sprites into a defined workflow | Controls node order, parameter mappings, and failure handling |
| Data intelligence space | Organizes scope, documents, and agents for each phase | Supports governance and subsequent context use |
MCP (Model Context Protocol) connects agents with external capabilities through defined tools. Configuring an MCP service lets DIG call external tools. Publishing a space MCP lets external AI applications use that space's context tools. Configure their addresses, authorization, and scope separately.
Read by responsibility
Data stewards can start with the quick start and governance capabilities, then follow end-to-end governance. Approvers should read approvals. Space owners should read space operations and context publishing. Administrators should complete setup, templates, and administration. Actual roles and permissions are configured by your organization.
Sprite designers should read Sprite design before workflow designers proceed to Routine orchestration.
Validate the outcome
Check model output against enterprise policies and actual data. Confidence scores, successful task status, and an agent's completion message do not prove that DAM contains the correct result. Available agents, templates, and tools depend on deployment configuration; example names are not guaranteed built-in resources.