Skip to content

AI chatbots and agents for WhatsApp and the web

We build WhatsApp chatbots and AI agents that answer enquiries, qualify leads and help your team find information, on WhatsApp, on your website or inside your business, using your own documents and data as the source.

An assistant is only useful if its answers are right. Before we build, we collect real questions from your business with their correct answers, test the assistant against them and show you the results. Questions it should not answer go to a person on your team, with the conversation so far.

AI agents and chatbots
Illustration of ai agents and chatbots work, with sample content

For example: a WhatsApp assistant that answers order-status questions from your own system.

An illustration of the kind of work we do, not a client project.

Order questions today

Before

Customer: Where is my order?

Your teamLooks it up and replies by hand
After hoursNo reply until morning

WhatsApp assistant

After

Customer: Where is my order?

Assistant: It has shipped. Here is your tracking link.

Answered from your order system.

A question it should not answerHanded to your team

What chatbot development includes

  • Real questions collected first

    Questions your customers or team actually ask, with the correct answers, gathered before we build.

  • An assistant where people already are

    On WhatsApp, on your website, or in an internal channel your team uses.

  • Answers from your own documents and data

    Product details, policies or order status, drawn from your sources rather than guessed.

  • Tested against your real questions

    Accuracy measured on your examples, with the results shared before launch.

  • Hand-over to a person

    Questions it should not answer go to your team, with the conversation so far.

Want this for your business?

Tell us what you need. A founder reads every enquiry and replies with questions or a proposed scope.

Start a project

Who it’s for

  • Businesses answering the same questions all day

    Opening hours, availability, order status and bookings, asked many times a day.

  • Sales teams that want enquiries qualified

    The assistant asks the first questions, so calls start with the facts.

  • Teams searching their own documents

    Policies, manuals and past work found by asking a question.

What you get at handover

Everything is handed over in your name, so you can run it, change it or move it to another team.

Handover record Checked by a reviewer
  • The assistant, connected to your WhatsApp number, website or channel
  • The test questions and the accuracy results
  • A guide to updating the documents it answers from
  • The hand-over rules: which questions go to a person
  • The expected monthly running costs
Illustration of a handover record

What we need from you

A project moves fastest when these are ready. We help you put together anything you do not have yet.

  1. 1Real questions with their correct answers, as many as you can collect
  2. 2The documents and data the assistant should answer from
  3. 3A WhatsApp Business account, if it will run on WhatsApp
  4. 4Agreement on which AI tools may see your data
  5. 5A person or team to receive hand-overs

How an engagement runs

  1. Scope document

    Scope: customer portal

    What we deliver

    • Customer sign-in, with roles
    • Order status page
    • Invoice downloads
    Not includedMobile appDone whenTested with your real orders

    Agreed by you before the build starts

  2. This week’s build

    This week’s increment

    Order status page

    Ready to view

    feat: order status page

    test: status for each order type

    docs: how statuses are calculated

    Next increment: invoice downloads

  3. Review checkpoint
    Our reviewer
    • Checked against the scope
    • Tests read and run
    • Ready for you

    Your approval

    Approve milestoneRequest changes
  4. Handover pack

    Code repository

    In your account

    Credentials record

    Transferred to you

    Documentation

    What was built and how

    Runbook

    Deploy, roll back, fix common problems

    Maintenance available on a monthly plan

What affects the cost

We quote after a scoping conversation, so there is no fixed price on the site. These are the factors that move the cost of this work up or down; the proposal states the scope, what is not included and the price.

Technologies we commonly use

  • WhatsApp Business messaging
  • Website chat widgets
  • AI models chosen per task, including open-weight models
  • Search across your own documents
  • Python
  • Connections to your order or booking system
Channels
WhatsApp, a website widget and an internal channel each need set-up and testing.
Sources it answers from
Documents only, or live data such as orders and bookings from your systems.
Languages
Each language needs its own test questions and checks.
Volume of conversations
More conversations mean higher model and messaging costs.
Hand-over design
How and when a person takes over, and in which tool.

How we keep AI reliable

Every AI service follows the same five rules, on top of the review every project gets.

Checklist for every AI projectBefore handover
  1. Real examples first

    We collect real examples from your business, with the correct answers, before we build.

  2. Tested against them

    We test against those examples and show you the results.

  3. A person approves

    Anything that sends messages, moves money or changes records has a person approving it, unless you decide otherwise.

  4. Running costs stated

    You know the monthly running costs before handover.

  5. A manual fallback

    Every automation has a documented manual fallback.

Questions

Before we build, we collect real questions from your business with their correct answers. We test the assistant against them and show you the results. Checking the format of an answer is not the same as checking it is correct.

Tell us what you want to build.

Describe the work. We will scope it with you, or tell you plainly if it is not work we should take on.