Skip to content
Zapito
Open App

The canvas (advanced view)

Behind every Zapito agent is one flow: a small graph of nodes on a canvas. The setup form generates and wires it for you, so you never start from a blank canvas — and most people never open it at all. Opening a bot lands on the simple view (test chat, knowledge, channels); the canvas is one click away behind Advanced: edit the flow on the canvas.

WhatsApp Trigger ── AI Agent ── Confidence Check ──(Confident)── Send Reply
▲ └─────────(Unsure)────── Human Handoff
Knowledge Base

One trigger per platform you picked, an AI Agent grounded by your Knowledge Base, a Confidence Check that routes sure answers to Send Reply and unsure ones to Human Handoff, which emails you the transcript. A message arrives at a trigger, walks the wires, and a reply (or a handoff) comes out the other end.

  • Filter a trigger by keywords, or only fire on a conversation’s first message.
  • Tune the confidence threshold on the Confidence Check.
  • Attach a Custom Tool so the agent can read live data from your API.
  • Add a Wait for a more human reply rhythm.
  • Or just to see exactly what your agent does — every node is inspectable, nothing is a black box.

The canvas has an AI copilot panel: describe a change in plain language — “only answer messages that mention orders”, “raise the confidence bar” — and it edits the flow for you. You review the result on the canvas; nothing applies behind your back.

  • Pan by dragging the background, zoom with the scroll wheel.
  • Add nodes from the node library (the panel on the left).
  • Connect nodes by clicking an output port, then the input port of the next node. Curved wires show the message path.
  • Click a node to open its config drawer — every field is described in the node reference.
  • Flow wires (solid) carry the conversation: trigger → agent → condition → reply / handoff.
  • Resource wires (dashed, into the agent’s bottom port) attach capabilities: a Knowledge Base or a Custom Tool. They don’t carry messages — they tell the agent what it knows and can use.

The Test chat panel runs your flow against the real engine — same code as production channels. Each turn shows a trace: which trigger fired, what the agent replied and with what confidence, which branch the condition took. Wires animate along the path while a test runs.

Changes save automatically (debounced — closing the tab flushes a final save). From the bot card menu you can pause a bot: paused bots are silenced on every channel until you resume them. Deleting a bot removes its flow and channel connections.