Note
Access to this page requires authorization. You can try signing in or changing directories.
Access to this page requires authorization. You can try changing directories.
[This article is prerelease documentation and is subject to change.]
The Preview tab in the new experience in Microsoft Copilot Studio provides an interactive testing environment where you can chat with your agent, debug its behavior, and review conversation history—all before you publish.
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.
What you can do in Preview
The Preview tab provides a comprehensive set of testing and debugging tools:
- Interactive chat: Send messages to your agent and review its responses in real time. The chat experience mirrors how users interact with your agent after it's published.
- Prompt suggestions: The preview displays conversation starters from your agent's configuration, so you can quickly test common scenarios.
- Activity trace: View a detailed chain-of-thought (CoT) trace that shows how your agent processed each message, including which knowledge sources it consulted, which tools it invoked, and the reasoning steps it followed.
- Conversation history: Review past test conversations with your agent.
- Response feedback: Rate agent responses with thumbs up or thumbs down, and provide written feedback to track quality issues.
Preview layout options
The Preview tab supports different views to fit your workflow:
- Maker view: The default view that includes the reasoning card within the chat, so you can see the agent's reasoning as you test.
- End user preview: Toggle End user preview on to see the agent as end users experience it, without debugging information.
You can switch between views using the End user preview toggle in the preview header.
When to use Preview vs. Evaluate
| Scenario | Use Preview | Use Evaluate |
|---|---|---|
| Quick, interactive testing during development | ✓ | |
| Debugging a specific response or tool call | ✓ | |
| Iterating on instructions or knowledge | ✓ | |
| Measuring quality across many test cases | ✓ | |
| Repeatable, automated testing before publish | ✓ | |
| Comparing results across agent versions | ✓ |
Both approaches are complementary. Use Preview to explore and refine, then use Evaluate to validate at scale.
Key features
Chat capabilities
When testing in the Preview tab, you can:
- Send text messages and conduct multi-turn conversations
- View agent responses rendered with rich formatting
- Copy agent responses for review or documentation
- Rate responses with thumbs up or thumbs down
- Provide detailed feedback through the feedback popover, including written comments
- Start a new conversation to test with a clean context
Activity trace
The activity trace shows you exactly how your agent processed each message:
- Chain-of-thought steps: See the reasoning nodes your agent followed
- Knowledge source attribution: View which knowledge sources the agent consulted and the citations it used
- Tool execution: See when tools, connectors, flows, or skills were invoked and what results they returned
- Error details: Identify and troubleshoot errors that occurred during processing
Select any source or tool pill in the trace to navigate directly to that component in the Build tab for editing.
Learn more in Use the activity trace to debug your agent.
Conversation history
The Preview tab maintains a history of your test conversations. You can:
- Review past conversations and their outcomes
- Filter conversations by status (All, Failed, In Progress, Completed, Auth Required, or Input Required)
- View a previous conversation's transcript and reasoning steps to verify agent behavior
- View detailed metrics for each session, including steps/turns, duration, last step, and the model that processed the request
Learn more in Manage preview conversations.