Copilot Studio: Create Custom Knowledge

Key insights

• OnKnowledgeRequested trigger lets Copilot Studio call external sources so you can build custom knowledge into a Copilot agent.
  The video demonstrates filtering a ServiceNow knowledge base as an example, and notes the approach works with any knowledge source.
• Copilot Studio connects multiple types of data so generative responses stay grounded in facts.
  Supported sources include public sites, SharePoint, OneDrive, Dataverse, file uploads, and real-time connectors.
• Basic steps: create or open an agent, add and name your knowledge sources, and attach them at the agent or topic level.
  Clear names and descriptions help the system pick the right content for answers.
• Use the create search query node and the OnKnowledgeRequested trigger to filter results and keep the conversation context aligned with your custom documents.
  This ensures Copilot returns relevant items instead of unrelated content.
• Key benefits include enhanced relevance, broader information coverage, simpler topic authoring, and near real-time updates from enterprise systems.
  You can deploy agents across productivity and collaboration channels to meet business needs.
• 2025 updates expanded source types and integration with Microsoft 365 Copilot, and added stronger governance features such as auditing and tenant controls for safer deployments.
  These changes make custom knowledge more powerful and easier to manage at scale.

Overview of the video

The YouTube video by Dewain Robinson walks viewers through creating custom knowledge sources inside Copilot Studio using the OnKnowledgeRequested trigger. He demonstrates the technique with a practical example that filters content from ServiceNow knowledge bases, though he stresses the approach is not limited to that system. The presentation aims to show how makers can hook external or internal content into Copilot agents to improve relevance and context for users.

Robinson explains the underlying mechanics so viewers understand both the setup and how the conversation context is preserved. He also highlights the role of the create search query node and how it helps keep answers grounded in the chosen knowledge items. As a result, the video positions this feature as a helpful but underused capability within the studio environment.

Overall, the content targets admins, makers, and analysts who need Copilots that respond accurately to domain-specific queries. Robinson keeps the pacing practical, showing configuration steps and then testing to verify the agent returns the desired knowledge. Therefore, the video is useful for teams that want to adapt Copilot behavior without authoring every response manually.

How the OnKnowledgeRequested trigger works

The core idea Robinson presents is that the OnKnowledgeRequested trigger lets a Copilot ask for external knowledge at runtime, rather than relying only on pre-authored responses. When a user query runs, the trigger can call custom knowledge sources so the agent retrieves targeted information. This mechanism helps the agent ground generative responses with live or curated content.

In the demo, Robinson links the trigger to a search node that builds queries tailored to the conversation context, ensuring answers stay relevant. He shows how filters limit results to specific knowledge bases in systems such as ServiceNow, which is handy for support desks that need to reference only approved documentation. Consequently, developers can reduce noise and increase precision by narrowing the search scope.

Practical demo steps and common use cases

Robinson starts the demo by selecting or creating an agent in Copilot Studio, then adds one or more knowledge sources at the agent or topic level. He emphasizes giving each source a clear, unique name and a detailed description so the system can orchestrate which content to use. After wiring the OnKnowledgeRequested trigger to a create search query node, he tests by posing a query and showing how the agent pulls in matching items.

The video highlights use cases such as IT support, where an agent might only reference a specific knowledge base; sales teams that need current contract clauses; and compliance teams that must limit responses to approved guidance. Robinson also points out that you can mix public websites, SharePoint, OneDrive, file uploads, Dataverse, and real-time connectors to broaden coverage. This flexibility makes the pattern useful across many business scenarios where specific, up-to-date answers matter.

Finally, he demonstrates troubleshooting steps when the agent returns irrelevant items, showing how to refine the search query and improve filtering logic. By iterating on query parameters and testing with realistic prompts, teams can tune recall and relevance. Thus, the demo balances practical setup with hands-on tips to shorten the learning curve.

Tradeoffs and implementation challenges

While custom knowledge improves relevance, Robinson also notes tradeoffs around performance, cost, and complexity. Calling external sources at runtime can add latency, especially when multiple connectors or large data sets are involved, and teams must weigh responsiveness against depth of coverage. Additionally, maintaining many custom sources increases administrative overhead and requires clear naming and documentation to avoid confusion.

There is also the risk of amplifying incorrect or outdated content if synchronization and indexing are not managed well, so teams must plan update cycles and quality checks. Overreliance on retrieval can cause hallucinations if the search returns weak matches, which is why Robinson recommends combining retrieval with clear fallback behaviors in the agent. Therefore, balancing freshness, accuracy, and speed remains a central challenge.

Governance, security, and recommended practices

Robinson covers governance concerns and points to available controls that help enterprises manage risk, such as activity auditing and tenant-wide inventory. He mentions how integrations with tools like Microsoft Purview and Sentinel can provide logging and monitoring so teams can trace data use and detect issues. For sensitive information, customer-managed encryption keys and strict connector permissions reduce exposure.

As a practical checklist, Robinson suggests clear naming conventions, detailed descriptions for each knowledge source, and regular testing with representative queries to measure relevance. He also recommends monitoring performance and setting retention or refresh policies for connectors that feed the agent. In summary, the video offers a hands-on roadmap for teams that want to build controlled, context-aware Copilots while navigating tradeoffs around cost, latency, and security.

Keywords

Copilot Studio custom knowledge, create custom knowledge Copilot Studio, Microsoft Copilot Studio tutorial, add custom knowledge to Copilot, build knowledge base in Copilot Studio, Copilot Studio knowledge creation guide, Copilot custom knowledge setup, how to create knowledge in Copilot Studio