How it works
- Signals — the SDK derives friction signals entirely in the browser: time on a page, inactivity after a started action, back-and-forth navigation, and rage clicks. Your developers can add two more: form-field error counters, and the page state they already expose to the agent (for example a funnel step left incomplete). No field values are collected, only counters and page paths — no personal data.
- Triggers — your developers declare triggers in the app code: “on this page, after this much inactivity, with this page state, propose help.” Clear-cut cases show a fixed message instantly; ambiguous cases ask the AGO agent to make the final call (“should we intervene here, and to say what?”).
- Nudge — the intervention is a small bubble near the chat launcher. It is never a blocking pop-up, always has a close button, and can offer a concrete one-click action (for example “Want me to pre-fill this for you?”) or open the chat with full context.
Built-in restraint
Proactive mode is designed to never feel like a pushy chatbot:- Each trigger has a cooldown, and a global cap limits nudges per session and per day.
- A nudge the user dismissed stays hidden for 24 hours by default (configurable per trigger).
- Only one nudge can be active at a time, and nothing is shown while the user is already chatting.
- Users can opt out entirely, and you can turn the whole feature off per agent or for the entire workspace at any time.
Enabling proactive mode
Proactive mode is off by default.- Turn on the workspace switch — an administrator enables proactive mode for the workspace. Until this is done, the SDK never shows a nudge, whatever the app code declares.
- Review per-agent settings — optionally disable proactive mode for specific agents via the agent’s proactive configuration.
- Declare triggers in your application — your developers declare the triggers in your application code with the AGO SDK:
