AI chatbot

Modern chatbots are two technologies in one wrapper. Scripted flows handle the predictable work — collecting an email, checking an order status, booking an appointment — deterministically, the same way every time. A language model handles the open-ended questions the script did not anticipate, ideally answering only from the company’s own content rather than from the model’s general knowledge.

Why it matters

A chatbot is often the first thing a customer touches, and the first place a business puts AI in front of the public. The design questions are practical: can non-engineers build and change the flows, can a bad publish be rolled back, does the bot know when to stop and hand a person to a human, and is there a cap on what it can do to your data.

How chatbots work in Oneop

Oneop’s chatbot builder is a visual canvas executing roughly 55 block types: messages, validated form input, branching and switches, loops and delays, AI blocks for intent classification, sentiment and knowledge-base-grounded answering, and native actions that create CRM contacts and deals, open tickets, look up orders, check stock and book appointments. Seventeen starter templates ship in the box, and you can generate a bot by describing it in plain English.

Publishing is governed. Every publish validates the whole graph and refuses to ship a broken flow, snapshots into an append-only version history with one-click rollback, and the live widget always serves the published snapshot — never a draft. A/B testing is real: two published versions run against a traffic split with sticky per-visitor bucketing, so each visitor sees a consistent experience.

Safety limits are code, not policy: no eval in expressions, a loop guard, and a hard cap of five write actions per conversation per ten minutes. When the bot reaches the edge of what it should handle, a handoff block routes the visitor to a human, with an optional business-hours gate that takes an offline branch after hours — see live chat handoff.

The bot’s AI answers run on Google Gemini and draw on your knowledge base — see knowledge-base grounding. Full detail on the product page: AI chatbot.