Build an in-app assistant

An in-app assistant lives inside your product (a dashboard, a customer portal, a booking app). Your users ask "How do I create an API key?" and the agent explains the steps, opens the right page — or creates it for them after they approve. AnytimeReply's own dashboard assistant ("Ask AI") is built exactly this way.

It combines four things you already have:

1. 1

   Train the agent on your docs

   Add your docs site or sitemap on Train → Sources, and add Q&A for anything tricky.

2. 2

   Load the chat in your app, signed in

   Create a short-lived token on your server (see Identity verification) and hide the floating bubble so the chat opens from your own "Help" button:

   Your app

   <script>
     window.AnytimeReplySettings = {
       user: { token: "JWT_FROM_YOUR_SERVER" }, // { user_id, email, name, attributes: { plan: "Pro" } }
       launcher: false,
     };
   </script>
   <script src="https://anytimereply.com/widget.js" data-agent="YOUR_AGENT_ID" data-launcher="none" defer></script>
   <button onclick="AnytimeReply.toggle()">Ask AI</button>

3. 3

   Add website actions for the tasks

   On Build → Actions → + Website action, describe each task and its inputs (e.g. open_page with page, create_api_key with name). Then register the handlers in your app:

   Your app

   AnytimeReply.onAction("open_page", async ({ page }) => {
     router.push(PAGES[page]);
     return { ok: true, opened: page };
   });
   
   AnytimeReply.onAction("create_api_key", async ({ name }) => {
     if (!(await confirmDialog(`Create the API key “${name}”?`))) return { ok: false, cancelled: true };
     const key = await api.createKey(name);   // runs as the signed-in user
     showSecretOnce(key);                     // show secrets in your UI, never return them to the chat
     return { ok: true, name };
   });

4. 4

   Tell the agent how to behave

   In Instructions, explain when to guide and when to offer doing it, e.g. "Explain the steps first, then offer to open the page or do it for them. Never ask for passwords or card details."

Keep it safe

• Confirm every change in your own UI (like the dialog above). Your handler decides — the agent can only ask.
• Run tasks with the user's own session and permissions. Your server's normal checks (plan limits, roles) still apply; return the error and the agent explains it.
• Never return secrets (API keys, tokens, passwords) to the chat. Show them in your app.
• Leave payments, deleting data and changing passwords to your own pages — the agent can open them.