Dolphy DocsTry the bot before publishing
The playground is where you talk to your own bot. The answer you get there is exactly what the site widget will produce: the same knowledge, the same retrieval and the same answer pipeline.
Playground conversations are flagged as test: they do not reach the inbox, they create no contact or request record, and they are not deducted from your monthly message allowance. You can use 30 messages per day.
Seeing behind an answer
Each answer has a diagnostics panel beside it. It makes no new model call; it reads data already collected while the answer was produced.
- Which routing branch ran: business information, nearby, live data or out of scope
- How the query was rewritten, if it was
- Which knowledge passages came back and with which similarity scores
- Which model answered
- How long the answer took
This screen is the fastest answer to “why was the reply wrong”. A wrong answer usually comes from a source in the knowledge base. Editing the model or persona settings before looking at the returned passages is usually the wrong place to start.
What to test
Five cases are enough:
- A question with a clear answer in the knowledge base
- A question that needs two sources combined
- A question with no answer
- A question that should produce a request or a handoff
- A short message, in Turkish and in every other language you enable
The fifth case matters most. A short English question must not receive a Turkish answer; if it does, the setting to change is the language, not the answer length.
The release criterion is not how many settings are filled in. It is that answers are grounded, that the agent knows its boundary, and that it hands over to a human at the right moment.
Independent of publishing
The playground is open even when the bot is not published, so you can try the bot without adding the widget to a site. The daily limit of 30 messages is shared with the real widget preview on the Setup screen.