Copilot Studio

How to use CA agent built in Copilot studio around Fabric Agents...

(2) Share

Report

Posted on by P1999

143

So , I am have been working with fabric Agents added via the Agents tab.

Things that I have tried out

1. Fabric Agents cannot be called from within the topics.

2 .'Create Generative Answer' cannot be used to generate answers from connected agent.

3. Fabric agent is automatically invoked at the end of a session, even when I use 'send a message' node

How to use CA to control user's query before fabric agent is invoked ?

Categories:

General topics

Topic creation & management

I have the same question (0)

All responses (7)

Answers (0)

Sort by

P1999 143 on at

Like (1)

Report
Copy link

Link copied!

@Valantis have you used fabric Agents in Copilot Studio ? Help me out with this.

Was this reply helpful? Yes No
Suggested answer

11manish 1,366 on at

Like (0)

Report
Copy link

Link copied!

You cannot directly control or invoke Fabric Agent inside topics.
It is automatically triggered as a fallback when the bot cannot resolve the query.

Reference :

https://learn.microsoft.com/en-us/fabric/data-science/data-agent-microsoft-copilot-studio#how-to-add-a-fabric-data-agent-to-the-custom-ai-agent-in-copilot-studio

https://learn.microsoft.com/en-us/microsoft-copilot-studio/add-agent-fabric-data-agent

Was this reply helpful? Yes No
Suggested answer

Valantis 3,406 on at

Like (2)

Report
Copy link

Link copied!

Hi @P1999,

Yes I have worked a bit with Fabric Data Agents in Copilot Studio.

The limitations you found are confirmed in Microsoft docs:

1. Fabric Data agents cannot be redirected to using the Redirect node inside a topic. This is an explicit confirmed limitation.
2. Fabric Data agents cannot be explicitly referenced in agent instructions.
3. Fabric Data agents do not work when the main agent is deployed to Microsoft 365 Copilot.

You cannot directly invoke or control it from inside a topic.

The Fabric Data Agent is handled by generative orchestration, which decides when to call it based on its description and the user query.

The only way to influence when it gets invoked is to write a very specific and detailed description on the Fabric Data Agent connection in Copilot Studio. Generative orchestration uses that description to decide whether to call the Fabric agent or handle the query itself. The more precise the description, the more predictable the routing.

If you need to pre-process or validate the user query before it reaches the Fabric agent, Try the workaround if helps:

- Create a topic that handles the specific intent you want to control
- Collect or validate the input in that topic
- End the topic with a message or variable
- Let the orchestrator then decide to invoke the Fabric agent based on the remaining context

You cannot force the Fabric agent call from a topic but you can shape the conversation context before it reaches generative orchestration.

Ref: https://learn.microsoft.com/en-us/microsoft-copilot-studio/authoring-add-other-agents

Best regards,

Valantis

✅ If this helped solve your issue, please Accept as Solution so others can find it quickly.

❤️ If it didn’t fully solve it but was still useful, please click “Yes” on “Was this reply helpful?” or leave a Like :).

🏷️ For follow-ups @Valantis.

📝 https://valantisond365.com/

Was this reply helpful? Yes No
P1999 143 on at

Like (2)

Report
Copy link

Link copied!

@Valantis I tried using instructions in Copilot Studio , to guide the agent to use a variable containing values to trigger the Fabric Agent. But it did not use it. Instead it invoked on its own and answered the question user had raised.

Was this reply helpful? Yes No
Suggested answer

Valantis 3,406 on at

Like (2)

Report
Copy link

Link copied!

Hi @P1999,

That is expected behavior and confirms what the docs say instructions cannot control or gate the Fabric Data Agent. The orchestrator invokes it directly based on description matching, bypassing your instructions entirely.

The only lever you have is the description on the Fabric agent connection itself. If you want the orchestrator to be more selective about when it calls the Fabric agent, make the description very narrow and specific. For example instead of a broad description like "answers questions about data" use something like "use this only when the user explicitly asks for [specific topic] data" with precise phrasing that limits the match surface.

If you need hard control over when it gets invoked, the honest answer is that the current preview does not support it. Redirect from topics is confirmed as not supported. Instructions are confirmed as not able to reference it. The only available workaround is description tuning.

Best regards,

Valantis

✅ If this helped solve your issue, please Accept as Solution so others can find it quickly.

❤️ If it didn’t fully solve it but was still useful, please click “Yes” on “Was this reply helpful?” or leave a Like :).

🏷️ For follow-ups @Valantis.

📝 https://valantisond365.com/

💼 LinkedIn

▶️ YouTube

Was this reply helpful? Yes No
P1999 143 on at

Like (1)

Report
Copy link

Link copied!

@Valantis for my usecase I wanted to control what the fabric agent uses to generate answers. But the Fabric Agent description only helps in deciding when it should be invoked. So just to be sure that I am at a dead end here. Are there any other ways using Power automate , Power BI etc to achieve my scenario in copilot studio.

Was this reply helpful? Yes No
Suggested answer

Valantis 3,406 on at

Like (1)

Report
Copy link

Link copied!

Hi @P1999,

You are not at a dead end, but the control needs to happen inside Fabric, not from Copilot Studio.

From Copilot Studio you cannot control what the Fabric agent uses to generate answers. That boundary is confirmed. But the Fabric Data Agent itself has rich configuration options specifically designed for this:

1. Data agent instructions (up to 15,000 characters): you can specify exactly which data sources to use for which types of questions, define business terminology, add SQL/DAX/KQL hints, and constrain query logic. This is where you control what gets queried and how.

2. Example queries (few-shot learning): provide sample question + query pairs for each data source. The agent retrieves the top relevant examples when answering and uses them to generate more accurate and predictable queries.

3. Data source instructions: per-source guidance on tables, columns, relationships, and query patterns to use.

These configurations are confirmed from Microsoft docs as the primary way to control answer generation behavior.

If you need deeper programmatic control that goes beyond what the Fabric agent configuration allows, the Power Automate alternative is: replace the Fabric connected agent with a Power Automate flow that calls the Fabric data agent REST API endpoint directly, processes the result, and returns it to Copilot Studio. This gives you full control over what gets sent and what gets returned but loses the natural NL2SQL capability.

Ref: https://learn.microsoft.com/en-us/fabric/data-science/how-to-create-data-agent

https://learn.microsoft.com/en-us/fabric/data-science/data-agent-configurations

Best regards,

Valantis

✅ If this helped solve your issue, please Accept as Solution so others can find it quickly.

❤️ If it didn’t fully solve it but was still useful, please click “Yes” on “Was this reply helpful?” or leave a Like :).

🏷️ For follow-ups @Valantis.

📝 https://valantisond365.com/

💼 LinkedIn

▶️ YouTube

Was this reply helpful? Yes No