Copilot Studio

How to make conversational design for an agent having fabric data agent.

(1) Share

Report

Posted on by P1999

126

So I have a Copilot Agent ,which has a fabric data agent added through the agents tab.

I just learnt , that fabric data agent can't be used from within the topics.

So I need help understanding ,how can I control the output which is automatically generated by the FDA.

i.e-limit the response to the user's query,avoid delusional responses,decide the format

Categories:

Building Copilot Studio chatbots in Microsoft Teams

Calling actions from Copilot Studio

Copilot Studio pre-built agents/templates

Copilot Studio skills development

Model context protocol

Topic creation & management

I have the same question (0)

All responses (5)

Answers (0)

Sort by

Suggested answer

Haque 1,758 on at

Like (0)

Report
Copy link

Link copied!
Hi @PratikJadhav08,

FDA responses in Copilot Agents are basically automatically generated and cannot be directly controlled or customized from within Topics. But we can influence and limit the FDA output by applying best practices/design patterns around how we invoke and handle the FDA responses in our overall Copilot Agent architecture.

Here are key approaches to control and shape FDA-generated output:

Limit the scope of the user query sent to FDA : Preprocess or filter user input before sending it to the FDA to keep queries focused and concise. Use prompt engineering or query templates to constrain FDA responses to specific topics or formats.

Post-process FDA output before presenting to users: Use additional logic or flow steps after receiving FDA output to trim, filter, or reformat the response. For example, apply regex, keyword filters, or length limits to avoid overly verbose or off-topic answers.

Use system or assistant instructions in your Copilot Agent design: Add guiding instructions or constraints in your agent’s system prompt or configuration to steer FDA responses toward desired formats or content boundaries. Explicitly instruct the agent to avoid hallucinations or to respond only with factual information.

Combine FDA with other agents or logic: Use FDA to retrieve data but combine it with other agents or custom logic that validate, format, or restrict the output before sending it to the user. This layered approach helps reduce delusional responses and enforce output structure.

Control output format via templates or schemas: If your use case requires structured output (e.g., JSON, tables), design your agent prompts and post-processing to enforce that format. Validate and parse FDA output against expected schemas and handle deviations gracefully.

Monitor and iterate: Continuously monitor FDA responses in your scenarios to identify patterns of undesired output. Refine prompts, filters, and logic accordingly.

References:

1. Consume Fabric Data Agent in Microsoft 365 Copilot

I am sure some clues I tried to give. If these clues help to resolve the issue brought you by here, please don't forget to check the box Does this answer your question? At the same time, I am pretty sure you have liked the response!

Was this reply helpful? Yes No
P1999 126 on at

Like (1)

Report
Copy link

Link copied!

@Haque , Is using 'Create Generative Answer' better or using prompts would do ? or is there some other ways to do it. Can you guide me on those other ways.

Was this reply helpful? Yes No
Suggested answer

Haque 1,758 on at

Like (0)

Report
Copy link

Link copied!

Hi @PratikJadhav08,

To be particular - using 'Create Generative Answer' is a feel of bit boundary becuase we will have less control on what is being generated. On the other hand - prompt is bit flexible - you can change the direction until yoy are satified. But, what is the sitation - if you think you don't have time/effort to write prompt, "Create Generative Answer' is not that bad. Justification upon you'r needs.

Was this reply helpful? Yes No
P1999 126 on at

Like (0)

Report
Copy link

Link copied!

@Haque , I created the agent. But apparently for me 'Create Generative Answers' node is not fetching data from fabric using the agent to agent way. Can you ellaborate exactly how can we use Create Generative Answers node in Copilot Studio for a fabric agent ?

Was this reply helpful? Yes No
Suggested answer

Haque 1,758 on at

Like (0)

Report
Copy link

Link copied!
Hi @P1999,

Let's see if the following tips help to fetch data from agent-to-agent way:

As the Create Generative Answers node responses based on knowledge sources, including those from Microsoft Fabric. However, when using it in an agent-to-agent scenario (Master Agent calling a Fabric-based Connected Agent), there are some important considerations:

Direct Use in Fabric Agents: Fabric agents typically leverage data from Fabric Lakehouse or other Fabric data sources. The Create Generative Answers node expects a properly configured knowledge source connected to the agent.If the Fabric agent is not exposing its knowledge source or generative answer capability as a callable endpoint, the Master Agent’s Create Generative Answers node won’t fetch data directly.

Agent-to-Agent Communication: Currently, the Create Generative Answers node does not natively support calling another agent’s generative answer endpoint as an input. Instead, you need to design the Master Agent to invoke the Connected Agent as a separate step (e.g., via a Connected Agent action or Power Automate flow) and handle the response explicitly.

How to Use Create Generative Answers with Fabric Agent: In the Fabric agent itself, use the Create Generative Answers node to generate responses from Fabric data. In the Master Agent, call the Fabric agent as a Connected Agent topic or action. Pass inputs via shared context or parameters. Capture the Fabric agent’s response and use it in the Master Agent’s flow. Avoid trying to embed Create Generative Answers node in the Master Agent to directly fetch Fabric data through the Connected Agent.

Alternative Approach: Use Power Automate or custom connectors to orchestrate calls between agents and handle generative responses. The Master Agent triggers a flow that calls the Fabric agent or its API, retrieves the generative answer, and returns it to the user.

References:
https://learn.microsoft.com/en-us/fabric/data-science/concept-data-agent
https://learn.microsoft.com/en-us/fabric/data-science/data-agent-microsoft-copilot-studio

Was this reply helpful? Yes No