AI Stand-ins
Try it out
Try it out opens a private dashboard conversation with the selected AI Stand-in so its saved instructions, identity, skills, and attached knowledge can be checked before enabling it for public coverage. The test ends when the tab is left or the conversation is explicitly ended.
- Availability
- All plans
- Configured in
- Dashboard → AI Stand-ins → select a Stand-in → Try it out
- Category
- AI Stand-ins
- Reference status
- Current
What the test conversation uses
Opening Try it out mounts a test-chat client for that Stand-in. Messages go through the AI conversation pipeline using the currently saved customer-controlled configuration rather than a separate mock prompt.
- Saved Team, Site, and Stand-in instructions in their normal composition order.
- The selected Stand-in’s saved skills and configuration.
- Knowledge bases that were attached and saved for the Stand-in.
- The Stand-in identity used to label the test conversation in the editor.
Start and end a test
- Save pending Identity or other Stand-in changes before opening the test when those changes should be evaluated.
- Open the Try it out tab and send realistic questions, including questions that should be answered from knowledge and questions the Stand-in should decline or hand off.
- Inspect the actual response rather than treating a successful connection as proof of quality.
- End the test conversation when finished. Leaving the tab unmounts its live client so the editor does not keep an unseen test connection running.
What it can verify
| Useful check | What to observe |
|---|---|
| Instruction behavior | Whether the Stand-in follows its role, tone, boundaries, and conversation approach. |
| Knowledge retrieval | Whether a question grounded in an attached source receives a useful, relevant answer. |
| Unknown information | Whether the Stand-in avoids inventing details when its instructions and knowledge do not contain an answer. |
| Skills | Whether an enabled skill enters the conversation under the conditions its own reference page defines. |
| Language | Whether the Stand-in follows clear visitor-authored language during the test. |
How it differs from a public chat
- It runs in the authenticated dashboard rather than through the installed website widget.
- It does not prove that a Site is installed, enabled, or matched to a real visitor URL.
- It bypasses the public decision among eligible humans and Stand-ins and targets the selected Stand-in.
- It does not demonstrate configured launcher timing, greeting memory, compact display, step-aside behavior, or mobile presentation.
- It should not be used as evidence that Web Push or browser notification delivery works on a rep device.
Availability and limits
- Try it out is available for Stand-ins on every plan.
- The Stand-in can be tested before it is enabled for public routing, which is the intended safety benefit.
- The test uses saved configuration; unsaved editor state can produce a different result after saving.
- AI output remains variable. Test several representative and boundary questions rather than relying on one favorable answer.
Example
The screen below shows the feature in its normal Stand context. Labels and surrounding controls may vary with account state and plan.
