AI Automation
AI Agents & Chat Assistants
A chat assistant that answers from your documents, says when it does not know, and hands the hard ones to a person.
What this actually is
An assistant that knows your business because it was trained on your documents, not the open internet.
A general chatbot will confidently tell your customers something that is not true. That is not a tuning problem — it is what a model does when it has nothing real to answer from. So we ground the assistant in your own material: policies, price lists, manuals, the answers your support team already sends every day.
Every answer points back to a source you can check. Where the assistant is not sure, or the question involves money, complaints or anything sensitive, it stops and passes the conversation to a person. It can live on your website, on WhatsApp, or inside a staff-only tool — wherever the questions already arrive.
How we work
Four steps, in this order.
Gather
Collect the policies, manuals and past answers the assistant should know.
Ground
Tie every answer back to a source document, so it can be checked.
Guard
Set the limits: what it may answer, and when it must hand over to a person.
Deploy
Put it where the question gets asked — website, WhatsApp, or an internal tool.
What you get
Every engagement hands over these.
- A deployed assistant on your website, WhatsApp or internal tool
- A searchable knowledge base built from your own documents
- An escalation setup that routes hard questions to a named person
- A conversation log you can review, correct and use for tuning
Who it’s for
This is usually the right fit when…
- Support teams answering the same handful of questions all day while real problems wait.
- Businesses whose customers message on WhatsApp at night, long after the office has closed.
- Organisations with good documentation nobody reads — policies, manuals and guides buried in a shared drive.
Questions
The things people ask before starting.
Will it make things up?
It can, if it is built carelessly. We reduce that by only letting it answer from your documents and by having it say plainly when it does not know. We do not claim it will never be wrong — no one honestly can. That is why every answer shows its source and sensitive topics go to a person.
Our documents are out of date and scattered. Is that a blocker?
It is the normal starting point, and part of the work. We collect what exists, find the contradictions, and bring them back to you to settle — because the assistant will repeat whichever version it is given. Tidying the source material usually helps your human staff as much as it helps the assistant.
Who can see what customers type into it?
You decide that before we build. Conversations can be stored in your own systems, redacted, or kept only briefly. If your assistant handles account details, complaints or payments, we set the boundaries on paper first and build to them — rather than discovering later that something private went somewhere it should not.
When is an assistant the wrong answer?
When the questions are genuinely varied and need judgement, or when the real problem is that your product is confusing. An assistant will answer a bad policy faithfully and at scale. If most of your inbox is a handful of recurring issues, fixing the issue beats automating the reply.
Have a project in mind?
Tell us about your project, study or idea — and let's see what's possible.
Get in touch