The Agentic Stage lets AI handle part of a conversation for you. You set the goal, give it the tools it needs, and decide when it hands back to your flow. It then talks with each customer until the job is done. It answers questions, recommends products and looks up orders, with no need to script every message or button. This guide covers the three tabs you configure: General, Tools and Exit Conditions.
What you configure
Tab | What it sets | Article |
General | The goal, instructions and advanced settings | |
Tools | What the stage can do: images, orders, products, knowledge base, your APIs | |
Exit Conditions | When the stage ends, where the flow goes next, and reminder nudges |
Advantages
Goal-driven. You describe the outcome, the stage works toward your goal and recognises when it is reached, replacing dozens of scripted messages, buttons and branches.
Flexible setup. Shape how it talks with your own instructions, and choose the tools it needs: product search, knowledge base, order lookup or your own APIs.
Every scenario covered. Exit conditions send each outcome to its own next step, whether the customer buys, asks for a person, isn't interested or stops replying.
Fits into your flows. Combine the stage with every other flow feature for full control, for example to create a ticket or send an additional scripted message.
Preview your flow before going live
Preview flow lets you chat with your flow as a customer would, before any real customer sees it. Nothing leaves the preview: no messages go to contacts, and API calls, contact updates and handovers are recorded but not run.
In the canvas toolbar, click Preview flow. The Preview mode panel opens.
Choose the version to test: Draft (your latest edits) or Live (the published version).
Type a message as the customer would, or use Attach image. Tap any buttons or list options the flow sends back.
Between messages, check what the flow did. Each reply shows the node it came from. When the Agentic Stage ends, the preview names the exit condition it took.
If the preview can't evaluate a condition or a random split, it asks you to choose the branch to continue.
Click Reset to start a new conversation, and try again with different customer behaviour.
Click Exit preview when you're done.
Test at least one conversation for every exit condition, plus one where the "customer" stops replying. For repeatable checks at scale, use Automated tests in the flow menu: simulated customers ask your questions and AI checks the stage's replies against what you expect.