Feature

An assistant that only knows what you taught it

Most business chatbots fail in the same way: asked something they do not know, they invent an answer. WaFi's assistant answers from the documents you upload and nothing else — and when they do not cover the question, it says so and hands over.

Start free See the whole platform

A general-purpose model will answer any question you ask it, confidently, whether or not it knows. In customer service that is a liability: a made-up refund policy or an invented price is worse than no reply at all, because the customer believes it.

WaFi takes the opposite approach. Your documents are the only source. The assistant retrieves the passages that actually relate to the question, answers from those, and is instructed to say when the answer is not there.

What it does

The parts that matter

Grounded in your content

Every answer is built from passages retrieved out of your own documents, so the facts are yours.

It declines rather than guesses

No relevant passage means no invented answer — the conversation goes to your team instead.

It understands follow-ups

"Anywhere else?" is meaningless alone. The assistant reads the previous turns, so short follow-ups resolve to the right subject.

It stands down for humans

The moment a real agent replies, automation goes quiet for the rest of that window. Customers are never answered twice.

One brain, both channels

WhatsApp and Instagram answer from the same documents, so the two never contradict each other.

Your tone, your rules

A system prompt sets voice and boundaries — what to always mention, what to never discuss, when to escalate.

Why "only from your documents" is the whole point

The assistant retrieves the most relevant sections of your knowledge base and instructs the model to answer using those and nothing else. If your documents do not mention pets, it cannot form a pet policy. That constraint is what makes it safe to put in front of customers unsupervised.

No invented facts — prices, policies and dates come from your files
A real "I don't know" — recognised as a non-answer and replaced by your fallback
Visible gaps — questions it cannot answer show you exactly what to add

Context is what makes it feel human

Real conversations are full of fragments. Somebody asks about your London properties, then simply types "anywhere else". Alone, that phrase matches nothing useful. WaFi carries the previous turns into both the retrieval and the reply, so the follow-up resolves against what was just discussed.

Follow-ups resolve correctly — "that one", "how much", "anywhere else"
Same behaviour everywhere — flows, the assistant and Instagram all read context
Scope is remembered — a conversation pinned to one property stays there

Questions

Straight answers

Which AI model does it use?

The platform supports several providers and we configure the one that suits your account. What matters more than the model is the constraint: whichever is used, it answers from your documents and is instructed to decline otherwise.

Will it ever make something up?

It is built not to, and in practice it declines rather than invents. The strongest protection is in your control: keep the knowledge base accurate, and do not upload anything you would not want quoted.

Can I control its tone?

Yes. The system prompt sets voice, language and boundaries — including instructions to always escalate particular topics to a human.

What happens when it cannot answer?

It sends your configured fallback message and the conversation waits for your team in the shared inbox. Nothing is silently dropped.

See it working on your own numbers

We will connect your WhatsApp number, load your documents and show you real answers to your real questions — before you commit to anything.

Start free Talk to us

No card required for the walkthrough.