Skip to main content

Internal Chat Tester

Communication → Internal Chat is a sandbox for talking to your own bot the way a client would. You type, the bot answers, and you see precisely what a real client would have seen — except nothing leaves the dashboard. It's the safest way to check a prompt change before your next lead arrives.

The page states its own contract at the top: "Runs the real chatbot · nothing is sent to a real number · uses live AI credits". Those three facts are the whole story, and each one is worth unpacking.

It Runs the Real Assistant​

This is not a preview or a simplified imitation. The tester runs the same assistant your clients talk to, with your real prompts, your real catalog, your real project photos, and your real routes. If the bot would send a product list, it sends the product list. If a route is switched off or has no data behind it, the bot behaves in the tester exactly as it would on WhatsApp — which is what makes the tester trustworthy as a check.

Nothing Is Sent to a Real Number​

No message you type here reaches a WhatsApp number, and no client is contacted. You can be as blunt, repetitive, or experimental as you like: ask the awkward questions, push on pricing, try the phrasing that broke things last week. Nobody is on the other end.

It Uses Live AI Credits​

The tester is free of consequences for your clients, but not for your balance. Because it runs the real assistant, every reply is billed the same way a real conversation is.

Testing Costs the Same as a Real Conversation

Each bot reply in the tester consumes credits exactly as it would on WhatsApp. Long exploratory sessions are genuinely worth it before a big prompt change — just don't leave the tester running as a toy. Check your balance on Credits if you've been testing heavily.

Controls​

  • The message box — labelled Message the chatbot as a client…. Type here as if you were the client and press send. There's no separate "as staff" mode; everything you type is treated as an inbound client message.
  • Show debug / Hide debug — reveals the behind-the-scenes detail behind each reply, then puts it away again. Leave it hidden for a clean client's-eye read of the conversation, and turn it on when a reply surprises you and you want to understand why.
  • Reset — clears the test conversation and starts fresh. Use it between experiments so the bot isn't still carrying context from your last round of questions.

Before You Can Test​

Your project needs a WhatsApp configuration in place before the tester will run. If it doesn't respond, check the provider setup on Configuration → Advanced first.

The Tester Interface Is English-Only

The internal chat screen is currently shown in English no matter which language your dashboard is set to. This affects the labels around the conversation only — the bot itself still replies in whatever language your prompts and your client's messages call for.

What It's Good For​

  • Checking a prompt change — edit your prompts on Advanced → AI System Prompts, then come straight here and ask the questions those prompts were meant to handle. Reading a prompt tells you what you hoped for; the tester tells you what you got.
  • Confirming a route is switched on and returns data — a route is only offered to the bot when it's enabled and has data behind it, so the fastest way to verify one is to ask for it. If the bot can't answer a pricing question or produce a floor plan, see Routes & Capabilities.
  • Previewing how media arrives — ask for photos of a specific space, or for a product list, and watch how they land: how many arrive, in what order, and what the bot says alongside them. This is much easier to judge in a real conversation than from a settings page.
  • Rehearsing the awkward cases — the vague opener, the client who only says "info", the one asking about something you don't sell. These are the messages that expose a thin prompt.
Test the Question, Not the Setting

Before you change a prompt, ask the bot the question you're unhappy with and keep the reply. Make your edit, hit Reset, and ask the exact same question again. Comparing two real answers tells you far more than re-reading your own instructions — and it stops you rewriting a prompt that was never the problem.