Interaction buttons now route directly to the AI agent by default — no scripted text required. Click a button, get an AI response. Scripted replies remain available as an explicit opt-in, and the admin UI makes it obvious which mode each button is in.
Quick-reply buttons are one of the highest-engagement elements in a chat widget. Users click them — so they need to work.
The original model put the entire burden on you: write a static answer for every button, in every language you support. Miss one, and the button added a user message bubble and then did nothing. The conversation stalled silently. This was a fundamental mismatch: you chose an AI assistant because you didn't want to hand-author every reply, yet the button system required exactly that.
Clicking an interaction button routes directly to the AI agent by default. The button label becomes the user's message, the AI receives it, and it responds — exactly as if the user had typed those words themselves.
No configuration required. No scripted text to maintain. Buttons work on day one.
If you need a guaranteed, word-for-word reply for a specific button — an exact pricing figure, a policy statement, a legal disclosure — that's still fully supported. You opt into Scripted Reply mode per button, write the text, and it behaves exactly as before.
By default the AI receives the button's label as the user message. If you want the AI to interpret the click differently — more context, a different framing, a specific question — you can set a custom message per button. Leave it blank and the label is used as-is.
The button editor now presents a clear mode toggle:
No ambiguity about what a button does when clicked. The mode is explicit, not inferred from whether a text field happens to be filled.
Flows continue to work exactly as before. They've been moved into a collapsible "Keyword-triggered flows (legacy)" section so the primary config path stays clean. If flows serve your use case, they're still there — most new setups won't need them.
Every button already configured keeps working without any changes. Buttons with scripted text continue to show that scripted text. The only difference is that buttons which previously had no response — and therefore did nothing — now trigger an AI reply. This is the intended behavior in virtually every case where scripted text wasn't written.