Test the Copilot
By the end of this page you will have had a real conversation with your Copilot and know how to inspect what it’s doing.
Prerequisites
- A brain bound (Bind a brain). Without one, the preview only shows a styling sample.
Chat in the live preview
Once a brain is bound, the preview on the right becomes an interactive chat. Type a message and the bound agent responds — streaming its reply, running tools, and showing activity exactly as it will for your users.
Send a message
Use the preview’s device toggle (Fullscreen, Desktop, or Mobile) to see the widget at different sizes, then chat. For an Aurora Support Copilot, try Where is my order 10248?.
Watch human-in-the-loop steps
If the underlying agent pauses for approval (human-in-the-loop), the request appears as a card inside the chat. Respond to it right in the preview — the same way an end user would.
Inspect the event stream
The Test tab is a read-only viewer of what’s happening under the hood — it isn’t a second chat box. You still chat in the preview; the Test tab shows the events that conversation produces.
It shows:
- Agent — the bound agent (or system) handling the chat.
- Thread — the current conversation’s id.
- Latency — TTFT (time to first token) and Total response time, in milliseconds.
- Event stream — a color-coded list of events as they arrive:
| Event | Meaning |
|---|---|
text-delta | A chunk of the streamed reply |
tool-call | The agent invoked a tool |
tool-result | A tool returned a result |
finish | The turn completed |
error | Something failed |
Turn on Raw Payloads to expand any event and read its full JSON. Use Clear to reset the list.
Use the event stream to debug: if the agent isn’t calling a tool you expect, you’ll see it here. Fix the behavior on the underlying agent, then chat again.
Common issues
- The preview shows a canned conversation — No brain is bound. Add one on the Brain tab.
- The Test tab is empty — It only fills as you chat in the preview. Send a message first.
- Responses look wrong — The Copilot only relays the agent. Adjust the agent’s skills, tools, or prompt, then retest.