How to control when knowledge sources are queried in Copilot Studio with generative orchestration?

(1) Share

Report

Posted on by MD-09061112-0

Hi everyone,

I’m working with Copilot Studio using generative orchestration, and I’m trying to control when specific knowledge sources are queried

Here’s the situation:

I have knowledge sources attached inside Generative answers nodes within specific topics. My goal is to improve the quality of answers by always asking the user for additional context information upfront (such as role or department). This lets me filter the query and select the correct knowledge base via the flow before passing the user’s pre-processed query to the knowledge source. This approach improves answer relevance and precision.

However, I’ve noticed that:

Even though the knowledge source is attached inside a Generative answers node, it still appears under the global Knowledge tab. During testing, generative orchestration sometimes queries the knowledge source directly, without first running the topic that collects the necessary context.

I’m wondering how to best achieve this when generative orchestration is enabled. I know I could use classic orchestration to manually control the flow, but I’d like to use generative orchestration for benefits like slot-filling and more natural conversation flow.

My questions are:

Are knowledge sources attached inside Generative answers nodes always considered part of the global knowledge pool?
How can I configure Copilot Studio to ensure that a knowledge source is queried only after a specific topic has been selected and context has been gathered?
How can this filtering and flow control be designed effectively when generative orchestration is used, to combine pre-processing and the benefits of slot-filling?

Any insights or best practices would be greatly appreciated.

Thanks in advance!

Categories:

Topic creation & management

I have the same question (0)

All responses (2)

Answers (0)

Sort by

CA1105 545 Moderator on at

Like (1)

Report
Copy link

Link copied!

Following since I have same type of query for my Agent.

Was this reply helpful? Yes No
Suggested answer

CU26051310-2 32 on at

Like (1)

Report
Copy link

Link copied!

The order of execution

0 : user prompt (some time after 1, depend of the prompt)

1 : agent global prompt
2 : orchestration using topics / action
3 : Orchestration trying to find knowledge
4 : conversationnal boosting

If you want full control just disable the orchestration and global AI mode. Source will be use only in conversationnal boosting. Use topic to direct exactly. (but yes quality drop)

If you still want orchestration, you must manage your topics and what happen inside.

Good news :

- loop are close to coming, they are available inside public preview, they will help you to keep user inside you'r topic loop
- knowledge local tunning will come very soon too (check build 2025 demo)

Topics are always trigger before knowledge source, if trigger is not called, the prompt is not covering you r use case.

if it help to understand, a quick click on final answer is great and important :)

Was this reply helpful? Yes No