Vivollo

Build your first flow

Build a real support flow from scratch — greet on open, answer from your knowledge, and hand off when needed.

Let's build a flow you'd actually ship. By the end of this page you'll have a support agent that greets people as they arrive, figures out what they need, answers from your knowledge, and pulls in a human when it's out of its depth.

We'll build it one block at a time and explain the why behind each one. The single most important idea — the one to internalize before anything else — is where each piece connects.

Start vs Fallback: the one thing to get right

A flow has two entry points, and knowing which is which is the whole game:

  • Start runs once, when the conversation opens, before the customer has typed anything. It's for a proactive greeting.
  • Fallback runs when the customer sends a message. It's the main handler — where your answering logic belongs.

So the AI that answers customers connects to the Fallback, not the Start. Get this right and the rest is easy.

What we're building

Here's the shape of the finished flow:

Start     → Send Message (proactive greeting, before they type)
 
Fallback  → Agentic AI (answers, searches, remembers, looks things up)
              → if it can't help → Send Message ("connecting you to the team")
                                 → Assign User / handoff

Two simple branches that cover what every support conversation needs: a warm opening, real help, and a graceful exit to a human.

Step 1 — Greet them on Start

Open Flows and create a new flow. You'll see two entry blocks already there that you can't delete: Start and Fallback.

Connect a Send Message to the Start. This is your proactive opener — it fires the instant a conversation opens, before the customer has said a word.

Set the message type to Text and write something welcoming:

Hi! I'm here to help with orders, products, and anything else.
What can I do for you?

Two small touches that make a big difference:

  • Use a variable if you know who they are: Hi {{customer_name}}!
  • If you serve customers in more than one language, turn on Auto-translate so this greeting arrives in each visitor's own language automatically.

On some channels, the customer's very first action is a message — so there's no "open" moment and Start is skipped. That's fine: the greeting is a nice touch when it fires, and the real work happens on the Fallback next.

Step 2 — Answer messages on Fallback

Now the heart of the flow. Connect the Agentic AI action to the Fallback — this is what handles whatever the customer types.

This single block is an autonomous agent. Whenever the customer sends a message, it will:

  • Understand what they're actually asking
  • Search your collections for the answer
  • Remember details they share (and recall them later)
  • Look up customer context and orders when relevant
  • Decide, on its own, when it's stuck and a human is needed

Point it at the assistant you want it to use — that's where its persona, model, and knowledge come from. For a first flow, the default assistant is fine.

This is the step people get wrong: they connect the answering logic to Start and wonder why nothing happens when customers write in. The handler for what customers say always hangs off the Fallback. When in doubt, ask: "does this run before the customer types, or after?"

At this point you already have a working agent. You could stop here, and many people do. But let's make the handoff graceful.

Step 3 — Hand off with grace

When the Agentic AI decides it can't help — a tricky complaint, a request for a human, something outside its knowledge — it triggers a handoff. Let's make sure that moment feels good.

Add a Send Message so the customer knows what's happening:

Let me bring in a teammate who can help with this — one moment.

Then add an Assign User action to route the conversation to the right person or team. The conversation moves to the Messages inbox carrying everything — the full history, what the AI learned, the customer's context — so your teammate never has to ask the customer to start over.

→ Learn more in Live handoff.

Step 4 — Test it like a customer

Before anyone real sees it, save and talk to your own agent. Use the in-panel preview or open your website widget, and ask the questions your customers actually ask:

  • "What's your return policy?"
  • "Where's my order?"
  • "I want to talk to a person."

Watch what happens. If the agent stays silent when you write, double-check that the Agentic AI is on the Fallback, not the Start — that's the usual culprit. If an answer is wrong, it's usually a knowledge gap, so add the missing information to your collection and try again. If the handoff fires too early or too late, tune your assistant's instructions.

You shipped a flow

You just built a complete support agent: it greets on arrival, understands, answers from your real content, and hands off cleanly. That's a genuinely production-ready pattern — plenty of live Vivollo agents are exactly this.

From here, you can add structure wherever you want more control: