In short
- Follow-ups just work.Short questions like "and for Canada?" are understood in the context of the chat.
- Long chats stay coherent.Recent messages are kept word for word, and older ones are summarised.
- Next questions it can answer.Suggested follow-ups are checked against your content before they are shown.
- Asks instead of guessing.When a question could mean different things, it asks which one.
People do not talk in perfect, standalone questions. They ask "do you ship to Canada?" and then "how long does it take?". An agent that forgets the first question gives a useless answer to the second. OneType keeps track of the conversation so every reply makes sense in context.
Follow-ups that make sense
Before searching, OneType rewrites the visitor's latest message into a full question using the recent conversation. So "and how much is it?" becomes a search for the price of the thing you were just discussing, and finds the right page. The original question is searched too, and the results are combined.
Memory for long conversations
The last 20 messages go to the AI word for word. When a conversation runs longer, older messages are summarised: what the visitor asked, what they were told and what is still open. The summary is refreshed as the chat goes on, so even a long conversation stays coherent without losing the thread.
Each message can be up to 4,000 characters.
Suggested next questions
After an answer that cites your content, the agent can suggest two or three short follow-up questions in the visitor's language. They are not random. Each one is checked before it is shown:
- It must be new, not the question just asked.
- Every product, plan or place it names must appear in the answer or its sources.
- It must not link anywhere the answer could not.
- A quick search must show your content can actually answer it.
So a tapped suggestion leads to a real answer, not to "I couldn't find that".
Clarifying questions
Some questions have different answers depending on what the visitor means: which plan, which store, which region, which version or which product. In those cases the agent is told not to guess and not to answer every option at once, but to reply with one short question that names the options, and answer once the visitor picks.
Asking for what it needs
Actions are coming soon. When they arrive, the same care applies: if the agent is missing a detail it needs, such as an order number, it asks one short question instead of guessing or inventing one.
What it does not remember
Memory lasts for one conversation. A new conversation starts fresh; the agent does not recall what a visitor asked last week. Clarifying questions are an instruction the AI follows, not a guaranteed step, so a clear enough question may just get an answer. Suggested follow-ups appear only after an answer that cites your content.