What an agent may do is written in a drawing you can read
Drag steps, set conditions, catch errors and test before it goes live.
What this is about
Flowbuilder is the visual workbench in which a process is recorded. You drag steps onto a canvas, connect them and put conditions between them: if the quantity is above 500, request a purchase price first; if the article is not found, send it to a human. Error handling is part of it, so a flow does not stall halfway without anyone noticing.
In test mode you run a flow with test data and see per step what came out, before anything reaches a customer. There is also a scheduler: daily, weekly or monthly, or on an event from outside. And this is where it is recorded what Magic Agent may and may not do — not in a model left free to roam, but in steps you can inspect and adjust.
Flowbuilder is the visual editor for creating conversation flows. Build complex chatbot scenarios, connect to APIs, trigger workflows and guide customers through any process — without writing a single line of code.
Flowbuilder combines visual simplicity with programming power. Build complex conversation flows without writing code.
Who we built this for
The administrator or team lead who wants to record a process without a programmer, and anyone who wants to be able to check why an agent did something. A flow is documentation that executes: it describes the process and it is the process, so it cannot quietly go out of date.
What you get with it:
What actually happens
What makes a flow usable in practice is not the dragging but what surrounds it: a test mode, error handling and a log per run. Without those three a drawing is a nice picture that tells you nothing during an incident.
Drag and connect
Steps on a canvas instead of lines of code, so a process can be read by the people who carry it out.
Conditions and choices
If-then, several outcomes and repetitions, so one flow handles several situations instead of five near-identical flows.
Error handling
A step that fails is caught and reported instead of leaving the flow standing still.
Test mode
Run with test data and see per step what came out, before it goes live and before a customer receives anything.
Scheduler and webhooks
Run at fixed times, or start as soon as a message arrives from outside.
Log per run
When it ran, how long each step took and where it went wrong. That turns a discussion about odd behaviour into an answer.
Multilingual
A flow can answer in several languages, so a visitor is helped in their own language without a second flow.
From design to production
Design visually, test safely, deploy to production. Flowbuilder makes workflow automation accessible.
Canvas Designer
Visual workflow design with nodes and connections.
Conditional Nodes
If/else logic, switch statements, loops.
Test Mode
Run flows with test data, see results per step.
Scheduler
Cron-based triggers: daily, weekly, monthly.
How it fits with the rest
Flowbuilder determines what Magic Agent may do and in which order, and Widget Server routes every conversation to the right flow. For WhatsApp, Meta Server does that. Magic Connect and Magic Workflow make the connections a flow pulls its data from, and the MCP servers in Magic Orb govern what an agent may reach.
So the agent is not a model roaming free: it walks through steps you can inspect, with a log afterwards. That is also the answer to the question clients ask most often — not whether it works, but who is responsible when it gets something wrong. The answer is in the flow.
What it delivers:
Tell us one process that currently lives in someone's head
We will draw it in Flowbuilder and run it in test mode on your own examples. That shows you the process and the places where it runs into an exception at your company.
Often half the steps already exist and the work is in the exceptions. Those you can then maintain yourself instead of requesting them from us.