How to Add Dataverse Tables as Knowledge in Copilot Studio

You can add Dataverse tables as a knowledge source so a Copilot Studio agent can answer questions from structured business data. A reliable setup requires Dataverse search, Microsoft authentication, carefully selected tables, and clear descriptions for technical column names and business terms.

Prerequisites

• Dataverse search must be enabled in the environment. If the Dataverse option is unavailable in Copilot Studio, ask an administrator to check this setting.
• The agent must use Microsoft authentication. No authentication and manually configured authentication are not supported for this knowledge-source type.
• The signed-in maker must have permission to access the selected tables and the records the agent needs.
• For a work or school deployment, confirm data-loss-prevention, security-role, and environment policies before publishing the agent.

Add one or more Dataverse tables

1. Open the agent in Copilot Studio.
2. On the Overview or Knowledge page, select Add knowledge.
3. Select Dataverse.
4. Search for and select the relevant tables. The interface supports multiple tables per knowledge source; use the current limit displayed in Copilot Studio rather than designing around an old documented maximum.
5. Review the knowledge-source name and description. Explain what the tables contain, which business questions they answer, and how related records should be interpreted.
6. Add synonyms and glossary terms where the schema uses internal names, abbreviations, or organization-specific language.
7. Select Add to agent.

After adding the source, test it with questions that reflect how real users speak. Include direct lookups, ambiguous wording, filters, dates, and questions that should return no result.

Why descriptions, synonyms, and glossary terms matter

Dataverse schemas often contain names that are clear to developers but unclear to an AI system or end user. A column such as cr_123_abc does not reveal whether it represents a flight number, departure city, or internal status.

A useful description identifies the business meaning and its relationship to other data. For example: “The cr_123_abc column stores the departure city associated with each flight code.”

Use synonyms for alternate wording

Add synonyms when users may refer to the same field in several ways. A “customer” field might also be called “client” or “account.” For numerical columns, describe the unit and business meaning instead of supplying only a short label.

Use glossary definitions for business rules

A glossary term should explain how a user phrase maps to the data, especially when it requires a condition across several columns.

These examples are patterns, not universal definitions. Use the exact table names, column names, values, time zone, and rules used by your organization. Glossary changes may take time to appear, so retest after the update has propagated.

Configure searchable multiline text and file columns

Support for unstructured content in Dataverse may be presented as a preview feature. Preview capabilities can change, may have limited regional or language support, and may not be appropriate for a production agent. Review the preview label and current Microsoft documentation in your environment before enabling it.

Creating or expanding a Dataverse search index can also increase storage consumption. Estimate the effect before indexing large file columns.

Configure the table in Power Apps

1. Open Power Apps and choose the same environment used by the agent.
2. Select Dataverse > Tables, then open the table used as knowledge.
3. Mark the required multiline-text and file columns as searchable where the option is available.
4. In Data experiences, open Views.
5. Open the table's Quick Find View.
6. Add the columns that should be returned and include the appropriate columns in the find criteria.
7. Select Save and Publish.

Configure search before adding the table to the agent when possible. If an existing knowledge source does not recognize newly searchable columns after indexing completes, remove and re-add that source only after confirming that doing so will not disrupt a production agent.

Known constraints to verify

• Index updates and repopulation can take time after a view or searchable-column change.
• Attached-file language, image, and table support may be limited.
• Only certain Dataverse virtual-table providers may be eligible as agent knowledge.
• The agent can retrieve only data the authenticated user and configured connection are allowed to access.

Test before publishing

Create a small test set with known answers. Check exact matches, synonyms, business-rule terms, permission boundaries, and unsupported questions. Inspect the cited records or source references rather than judging only whether the response sounds plausible. If a question is answered incorrectly, refine the source description, schema terminology, view configuration, or test query before adding more tables.