Test an agent (preview)

[This article is prerelease documentation and is subject to change.]

Use the Preview tab in the new experience in Microsoft Copilot Studio to interactively test your agent before you publish it. Interactive testing lets you verify your agent's responses, check knowledge source accuracy, and confirm that tools and skills work correctly.

Note

This article reflects the new agent experience in Microsoft Copilot Studio, which is currently available as a production-ready preview. Learn about the two experiences in Classic vs. new agent experience.

  • Production-ready previews are subject to supplemental terms of use.
  • Some capabilities available in the classic experience aren't yet available in the new experience.
  • Agents created in the new experience can't be converted to the classic experience.

Prerequisites

  • An agent created in the new experience
  • At least basic instructions configured for your agent

Start a test conversation

  1. Open your agent in Copilot Studio.

  2. Select the Preview tab.

  3. Type a message in the chat input at the bottom of the preview pane and press Enter or select the send button.

  4. Review the agent's response. The agent processes your message using its instructions, knowledge sources, and available tools.

  5. Continue the conversation to test multi-turn interactions.

Tip

Use the suggested prompts that appear in the preview pane to quickly test common scenarios. These prompts are derived from the conversation starters configured in your agent's settings.

Test specific capabilities

Test knowledge responses

  1. Ask your agent a question that should be answered using one of its knowledge sources.

  2. Review the response for accuracy and completeness.

  3. Toggle End user preview to switch between the maker view and the end user view of the preview.

  4. Check that citations and source attributions are correct.

Test tool invocations

  1. Ask your agent to perform a task that requires a tool (for example, looking up data or performing an action).

  2. Review the response to confirm the tool was invoked correctly.

  3. In the activity trace, verify the tool execution node to see the tool that was called and the result it returned.

Turn on debug mode

To see detailed information about how your agent processes each message:

  1. In the Preview tab, toggle End user preview off to use the maker view.

  2. The reasoning card appears within the chat messages, showing the agent's chain-of-thought processing steps for each interaction. The reasoning card is expanded by default in the maker view.

  3. Select the reasoning card within a chat message to expand or collapse the chain-of-thought details for that response.

Provide feedback on responses

  1. Below any agent response, select the thumbs up or thumbs down icon to open the feedback popover.

  2. Enter a written comment describing the issue or positive behavior.

  3. Submit the feedback. Use this feedback to track quality patterns as you iterate on your agent.

Start a new conversation

To reset the conversation context and start fresh:

  1. Select New chat in the preview header.

  2. The current conversation is saved to your conversation history.

  3. A fresh conversation starts with no prior context.

  • Last updated on