Answer the phones. Kill the manual work.
One agent, one workflow, one number we agree before anything is built. A written scope, a quote by email, and a live run on your own messy enquiry so you watch it hold or fail before you sign the work off.
How it runs, start to handover.
Three steps. You can stop after any of them, and what has been done so far is written down and handed to you.
-
Pick one workflow, agree one number
A short call, then a written scope. One workflow that eats the most hours, one number that has to move, how it is measured today, and the person on your side who checks it.
Nothing starts until that page is signed by both of us.
-
Build it on your real mess
We build against your actual inbox, forms and records, not a tidy sample. Limits, escalation rules and the handoff to a person go in while we build, not after.
Typically 10 to 14 working days for one workflow.
-
Live run, then keys in your hand
We sit on a call and you pick a real enquiry we have never seen. It runs in front of you. Then you get the prompts, the code, the keys, the runbook and the off switch.
Care after that is optional. The agent is yours either way.
Most agent projects get switched off.
Plenty of agencies are selling a wrapper around a chat box and calling it an agent. Here is exactly what we do differently, in plain terms, so you can hold us to it.
One workflow, one measured number
Not a platform. One job, with one number we write down before the build: first reply time, enquiries handled without a person, hours off a rota. We measure it before, we measure it after, and we show you both. We do not promise a figure, we agree how it gets counted.
A written quote, by email, first
There is no figure on this page because every job is a different size. We scope yours on a call, then email one number with the workflow it covers, the systems it connects and the date it lands. If the scope grows later, we quote the extra and you say yes or no. No hourly meter, no discovery phase that quietly bills for six weeks.
A live run on your own worst enquiry
Recorded demos are theatre. You pick a real message from your own inbox, one we have never seen, and it runs on a call while you watch. If it falls over, you see it fall over and you hear what we are going to do about it.
Human handoff is the design, not a patch
The agent has a confidence line and a list of things it is not allowed to say. Below the line it stops, hands the whole thread to a named person and says so to the customer. That path is built on day one, not bolted on after the first complaint.
When an API changes, here is who fixes it
Models get retired and APIs change. On a monthly care plan we watch it, patch it and write to you saying what changed and why. Care is optional, quoted with the build, and you can cancel it. With no care plan, we quote the fix and you decide. The code and the keys are yours, so any competent developer can also do it.
You can switch it off in one click
There is a kill switch and a fallback that sends everything back to the old way of working. If it is not earning its place, turning it off costs you nothing and breaks nothing.
The risk, said out loud.
A lot of the enterprise agent projects announced in the last two years have quietly been rolled back. The pattern is boring and it repeats: the scope was too wide, no number was agreed at the start, there was no human path when the thing was unsure, and nobody owned it after handover.
We would rather say that than pretend it does not happen. It is also the reason this page reads like a contract instead of a pitch.
- Scope too wide, so we build one workflow and refuse the second until the first is running.
- No agreed number, so we write the number and the counting method into the scope.
- No human path, so escalation and the kill switch ship with version one.
- Nobody owned it, so handover names a person on your side and a runbook they can follow.
- If the live run says the job is not worth automating yet, we tell you and we stop there.
No list price. A quote for your job.
Every agent is scoped to what you actually need, so the number is different every time. We work it out with you on a call and send it by email, in writing, before anything starts.
The work the agent takes off your desk
One workflow, done all the way through, in your own words and inside the tools you already use.
- Answers the phone and WhatsApp at 2am, and on Sunday, in your tone
- Qualifies the enquiry: what, where, when, budget range, who is asking
- Books it straight into the calendar your team already lives in
- Chases the quote you sent and the follow up everyone forgets
- Pulls the copy and paste data entry out of the week: forms to sheet, sheet to invoice, email to record
- Hands the whole thread to a named person the moment it should, and tells the customer it is doing that
- Works over voice, WhatsApp, email, web form or the inbox you already have
It only answers from material you approved, and every run is logged with the reason it chose that answer. What it will not touch is written down before the build, not discovered later.
Get a written quoteWhat you end up owning
At handover it all moves into your name. Nothing is locked to us.
- The prompts, as plain text you can read and edit
- The code and the workflow files, in a repository you own
- Accounts and API keys in your name, billed to you by the provider
- A runbook: what it does, what it refuses, how to change a reply, who to call
- The escalation rules, the confidence line and the kill switch
- A readable log of every run, exportable, kept in your own storage
- A recorded walkthrough for whoever joins the team next
- 30 days of fixes after handover, no charge
Model usage is billed by the provider straight to your own account, so you see that cost yourself and we never mark it up. Care after handover is monthly and optional, and cancelling it does not stop the agent, because it was always yours.
Ask what yours would costHow the number gets decided
We scope it against your job, not a template, so the quote moves with the work in front of us. These are the things that actually move it, and we show you which ones apply to you before you commit.
You get one number, one page, and the date it lands. Nothing goes on an invoice you have not already seen and agreed.
- How many systems it has to talk to, and whether they have a proper API
- The state of your records, and how much cleaning up they need first
- Voice as well as text, which needs more testing than people expect
- A second or third language, and who signs off the wording in it
- Whether every outgoing message needs a human to approve it first
- Whether you want us watching it afterwards, or your own developer takes it
The four things people actually ask.
These are the questions from real calls, with the answers we give on the call.
Is this just a chatbot with a new name on it?
A chatbot talks. This does the steps. It reads the enquiry, looks up your real records, writes the reply, updates the row, books the slot and stops when it hits something it was told not to touch. Every run leaves a log you can read.
If a plain chatbot is genuinely all your job needs, we will say that on the call. It is a smaller build and a smaller quote, and we would rather sell you the smaller thing that works.
What happens when the model or an API changes and it breaks?
It will break at some point. Providers retire models and change endpoints on their own schedule, and no honest builder will tell you otherwise.
On care, we watch it, we patch it and you get a written note saying what changed and what we did. Without care, we quote the fix and you decide whether to take it. Either way the code, the prompts and the keys sit in your accounts, so you are never stuck waiting on us.
What if it says something wrong to one of my customers?
It only answers from your own approved material, it has a written list of things it must never state, and it has a confidence line. Below that line it does not guess. It says it is passing this to a person and hands over the whole thread.
For the first two weeks you can hold every outgoing message for approval, so you read what it wants to send before a customer does. Most people turn that off in week three. Some keep it on for the money questions forever, which is a perfectly sane choice.
Why should I believe your demo?
You should not, and neither would we. Anyone can record a clean run and cut the three that failed.
So we do it live. You pick the enquiry, from your own inbox, ideally the ugly one with a typo, a half question and an attachment. We share a screen and it runs. What you see is what you are buying, and if it cannot handle your mess we would rather find that out on a free call than after an invoice.
Start your agent.
Tell us the four things below and you get a link to the full brief straight away. Fill it in when it suits you — the brief is what we build from, so there is no call you have to sit through first.
Get an agent that answers for you
Tell us where to reach you. We'll send a short brief, then build an agent trained on your business and hand it back within 24 hours.
You're in
We've sent your brief link to WhatsApp and email. It takes about two minutes — the sooner you fill it, the sooner your agent is ready.
Open the brief nowBring us your worst enquiry.
Twenty five minutes on a call. You describe the job that eats the most hours, we tell you straight whether an agent is worth building for it, and you get the quote in writing by email after. If the answer is no, you get that for free.
A person reads every message and a person replies, usually within one working day.
