What the Playground is
The Playground is a test chat with your own Yaplet. You ask it questions the way a visitor would, and it answers from your brand's knowledge. Next to the chat sit Yaplet's settings, so you can change its instructions or its AI model and see the effect on the very next answer, before anything is saved. Visitors keep getting the saved version until you press Save changes.
Nobody else sees a test chat. It never lands in your inbox or your reports, and no one is notified.
Open it
Go to Brand → Playground, the first tab of the brand menu. You need the AI agent permission. When you finish onboarding with your website address you land here, with the setup's progress in the middle of the page until your website has been read.
Try Yaplet in five minutes
- Check the status card at the top of the left column. Trained means Yaplet has something to answer from. Learning your website · 12/25 pages means it is still reading, so give it a moment. Could not read your sources means nothing could be read: open Brand → Knowledge to see why. If the small line under Trained says No knowledge added yet, add some knowledge first.
- Use the test questions. Once your website is learned, a Test questions list appears. Open one to see Answer on your website, then press the send button on its row to ask it in the test chat. If Yaplet's reply says the same as your site, your knowledge works.
- Ask your own. Try three questions you know the answers to, one it should not be able to answer (a product you do not sell: it should decline, not guess), and "I want to talk to a person".
- Start fresh with the round New test chat arrow in the chat header whenever earlier answers could affect the next one.
Change something and see the effect before you save
- Instructions — the text Yaplet follows in every answer. Write your own, or press Start from a template and pick Customer support, Online store, Sales and leads or Bookings. A template replaces what is in the box, and Undo changes puts back the last saved text. While the text is unsaved, a line under the box says "Not saved yet — the test chat is already answering with these instructions." What to write in them is covered in Write a system prompt and tone for Yaplet.
- Model — the AI model that writes the answers. Where your plan lets you choose, pick one and ask again. It is a single setting for your whole workspace, the same as Settings → AI Models, so saving it changes it everywhere. Where your plan does not let you choose, the row is locked and says "Upgrade for more advanced models".
Happy with the result? Press Save changes in the title bar. Not happy? Press Discard. Leaving the page with unsaved changes asks first. To compare two versions fairly, press New test chat and ask the same question again.
What the test chat does and does not do
- It is the real Yaplet, with the same knowledge and the same model. Each answer is paid for like any other AI answer: first the free 14 days' credits, then credit you bought, then your plan. If nothing can pay, the payment window opens (see Yaplet usage and credits).
- When Yaplet would hand the chat to your team, you see a card that says so, and nobody is alerted.
- API tools run for real, so a tool that changes data changes real data.
- Workflows do not start, and tools set to Only for identified visitors are refused, because the test visitor is anonymous. Test those on a real page: see Test your chatbots before going live.
- It answers even if Website chat is switched off under AI agent → General → Answers on. A note above the chat says visitors on your website will not get these answers.
Test a phone call too
The Playground can test your phone agent as well. People who can manage phone agents see a Chat / Call switch above the test area. Press Call to start a test call, in your browser, to the brand's phone agent. It answers from the same knowledge, products and tools as the chat.
- The left column then shows that phone agent's main settings instead of Yaplet's: whether it is answering calls, its voice and language, its greeting and its instructions. They are saved with the same Save changes button in the title bar, and a test call always uses the saved version. All settings of this agent leads to the rest.
- If the brand has no phone agent yet, Call shows a No phone agent yet card with a Create a phone agent button.
- If your plan has no phone, the owner sees Call with a lock and a Yaplet can answer your phone too card with a See plans button. Colleagues without phone access do not see the switch at all.
- A test call is billed like a phone call and stops after 5 minutes.
When you are ready
Press Save changes first, because visitors only ever get saved instructions and a saved model. Then press Install on website in the title bar: it opens your chat widget's Install tab, and Install Yaplet takes it from there. Check also that Enable Yaplet is on, and Website chat is on under Answers on, both on AI agent → General.
Common questions
Does asking test questions cost anything?
Yes, the same as any AI answer. See above for who pays. If you send questions very fast, Yaplet asks you to wait a moment.
I changed the instructions, but my visitors still get the old behavior.
The change is not saved yet. Press Save changes in the title bar. Until then only the test chat uses it.
An answer in the test chat is wrong. What do I fix?
If a fact is wrong or missing, fix the knowledge on Brand → Knowledge (Why isn't Yaplet answering from a source I added? helps when a source is being ignored). If the facts are right but the tone or length is not, change the instructions.
Can my colleagues see my test chat?
A test chat belongs to the brand, not to you. A colleague who opens the Playground of the same brand sees the latest test chat too. Visitors never do.