Copilot Studio – Azure SQL Knowledge Source Connected but Agent Cannot Retrieve Data

Question

Dear reader,

I'm integrating an Azure SQL Database as a knowledge source in Copilot Studio, but despite the connection showing as successful and preview data being visible, the agent is unable to retrieve any data when queried. Below are the details and context of our setup.

Knowledge Source:

• Type: Azure SQL Database
• Connection status: Successful – preview data is visible inside Copilot Studio.

Although the data source is connected successfully and preview data displays correctly, when we query the agent with questions it doesn't seem to make a valid connection.

It is shown that it tries to read the table three times, but all time-out after almost exactly 60 seconds. Then the LLM simply responses with a generic answer without any connected knowledge.

After the response, we can see that it did know to search through the table, but it shows a red exclamation mark with text on hover saying 'Request time-out occured. Dataverse Knowledge Search API call timed out'

The following settings are disabled:

• Web Search
• Content Generation

I know there is a similar question on this forum (https://learn.microsoft.com/en-us/answers/questions/5643897/copilot-studio-azure-sql-knowledge-source-connecte), but that didn't yield any solution whatsoever.

What causes this problem? And how should I approach it?

Answer 1

Thanks for the responses Jerald and Pilladi, I appreciate the help!

To answer Pilladi first:

• Some tables have up to tens of thousands records. However, I also tested it on a very small table with only a couple of rows and columns, and the issue still persists. I can't check directly how the Azure SQL Database is configured, as I don't have an Azure subscription in my own tenant. I'm connecting to the Database purely as an app identity via clientID and clientSecret given to me. Since I suspect preview also does a live read against the table, I'd expect a firewall/VNet block to have affected both preview and runtime. Nevertheless, It's something I'll check with the administrator. Jerald:- Your first proposed solution to use semantic indexing is not fit for my case. I need exact filtering, not a semantic similarity search.- I also don't think the second proposition firts my case. My tables have up to 20 columns. The question I need to support are genuinely open-ended, rather than a fixed set I could enumerate as parameterized queries.

Thus these suggestions unfortunately don't seem to fit my case. If anyone has another idea I'd be glad to hear it!. Otherwise, if we do have a Power Platform support plan, I'll be sure to open a ticket.

Best regards,

Niek

Answer 2

Hi @Niek
Based on the information you've shared, the Azure SQL knowledge source appears to be configured correctly, as the connection is successful and the preview data is visible in Copilot Studio. Since the failure occurs only during runtime and the error displayed is "Dataverse Knowledge Search API call timed out," this suggests that the runtime knowledge retrieval request is timing out rather than a connection or authentication failure.

To help narrow down the cause, could you please let us know approximately how many rows are present in the Azure SQL table? Does the table contain any large text columns (for example, nvarchar(max) or varchar(max))? Does the issue persist if you create a small test table with only a few rows and simple text columns? Additionally, is the Azure SQL Database configured with a Private Endpoint, VNet integration, or firewall restrictions that could affect runtime connectivity? It would also be helpful to know whether this behavior occurs with all Azure SQL knowledge sources or only with this specific table.

Since the connection and preview are working as expected, testing with a small sample table can help determine whether the timeout is related to the dataset or to the runtime retrieval service. If the issue persists even with a minimal test table and there are no networking restrictions, it may indicate a service-side issue with the Dataverse Knowledge Search component and would require further investigation.

Answer 3

Hello Niek,

Greetings! Thanks for raising this question in Q&A forum.

The root cause of what you are seeing is a fundamental architectural limitation of how the Azure SQL knowledge source connector works at runtime. The preview in Copilot Studio only validates the connection, it does not simulate actual runtime queries. At query time, Copilot Studio routes the request through the Dataverse Knowledge Search API, which has a hard 60-second timeout. Azure SQL, being an external structured database, does not integrate natively into Dataverse's search indexing pipeline the way SharePoint or Dataverse tables do. So instead of running a fast indexed semantic search, it attempts live SQL queries through the connector at runtime, which consistently times out. The red exclamation mark with "Request time-out occurred. Dataverse Knowledge Search API call timed out" confirms this exactly.

This is a known platform-level limitation with the Azure SQL enterprise data connector in Copilot Studio, as seen in similar threads, and it currently has no direct fix within the SQL knowledge source approach itself. Here is how to work around it effectively:

1. The most reliable approach for your scenario is to use Azure AI Search as the knowledge source instead. Sync your Azure SQL data into an Azure AI Search index, either directly using an Azure AI Search SQL indexer, or via a scheduled Azure Data Factory pipeline. Then add the Azure AI Search index as your knowledge source in Copilot Studio. This approach uses pre-built semantic indexes and will not time out. Reference: https://learn.microsoft.com/en-us/microsoft-copilot-studio/knowledge-azure-ai-search
2. If you need real-time data from Azure SQL (for example, orders that change frequently), the correct pattern is to build a Power Automate flow or a custom connector action that executes a targeted parameterized SQL query, and wire it into Copilot Studio as a topic action rather than a knowledge source. This gives you full control over what query runs and avoids the timeout entirely.
3. If you want to keep the Azure SQL knowledge source while the platform matures, go to your agent settings, remove the Azure SQL knowledge source, and re-add it with the connection refreshed. This sometimes clears a stale connector state, though it will not resolve the underlying timeout if your table size is large or the connector latency is high.
4. While you are testing, reduce your table to the smallest possible subset of rows to determine if table size is a contributing factor. You mentioned you already have only 5 rows, which tells us the timeout is not a data volume problem but a connector routing issue, which further confirms this is a platform-side limitation rather than something in your configuration.
5. If you have a Power Platform support plan, open a support ticket and reference the error "Dataverse Knowledge Search API call timed out" with your Azure SQL knowledge source. The product team is aware of this class of issue and a ticket helps them track impact.

If this answer helps you kindly accept the answer which will help others who have similar questions.

Best Regards,

Jerald Felix.