The **Playground** tab lets you check, before a real customer sees it, which fragments of your knowledge base the AI would retrieve for a given question. It doesn't generate a chat reply: it shows you the exact sources — the document fragment — it would rely on. If your agency has several agents with tag-restricted access (section 3 and manual 23 · AI agents), the **Agent** selector lets you test the query as if you were one of them, to check exactly what each one would see. **Requirements:** having at least one document uploaded and processed for the test to make sense (with an empty knowledge base, there's nothing to retrieve). 1. Go into **Knowledge base** and tap the **Playground** tab. 2. Type an example question, as if you were one of your customers. 3. In the **Agent** selector, choose **"Global (no agent, full catalog)"** to test against every document, or a specific agent to check exactly what that agent would see based on its tags. 4. Tap **"Test query"**. **Result:** the **"Retrieved sources"** panel lists the document fragments retrieved for that question and that agent (or the full catalog), with the source document's name and the retrieved text itself — so you can verify, before the question actually arrives over WhatsApp, Telegram or the web chat, exactly what information the AI would use to answer, and quickly spot if a document is missing a tag it should have.