Skip to content

Agents that tell you what they do not know.

Hire one for your customers, for your household, or for nobody but you. Each one turns up ready for the work instead of waiting to be built. Where one answers your customers, it spends its first two weeks writing replies you read before anybody sends them.

From $39 per agent per month. No contract, and one click stops it.

Try it

Three of them, and they answer differently on purpose. Switch between them and watch where each one stops.

The reason you have not put an AI in front of your customers is that you cannot watch it first.

One wrong answer to a real person costs more than every question it got right. So the question was never whether it is clever. It is whether you saw it work.

How this is different

You have seen AI answer questions before.

So the useful question is not whether this one is clever. It is what it does the first time it reaches the edge of what it knows.

A chatbot you build yourself
You draw the flows, you write the answers, and what it gets wrong is what you built. The work is yours and so is the blame.
An enterprise agent
Excellent, and you find out how excellent after procurement, a pilot and a signature. The evaluation is a project of its own.
A tab with ChatGPT in it
Answers anything, remembers nothing about your business, and nobody sees what it told your customer.
OpenClaw or Hermes, run by you
Both are open source and free, and they are what your agent runs on here. Run one yourself and you own the server, the upgrades, the model bill, the signature checks on every channel, and the writing of the role itself. There is no queue to read and nothing holding the send button.
Arsita
A role that arrives already knowing the job, on an engine we did not build and could swap. The one that answers your customers spends two weeks writing replies to your real messages and sending none of them. You read what it would have said. If it is not good enough, it never speaks to anyone.

Before it answers a stranger

Nothing reaches your customer until you have read it.

This is the role that answers people who do not know you, so it is the one that waits. The agents that work for a household or one person start from the first day, because somebody who can correct them is already in the room.

A customer's message reaches the agent, which writes a reply. The way on to the customer is locked. The reply goes to a queue for you to read instead. A message Your agent Your customer A queue you read You
WhatsApp · •••• •••• ••31 21:14

hi, are you open tomorrow? need to pick up my order

Draft

We are open 9am to 5pm tomorrow. What name is the order under? I will have it ready before you arrive.

Held back for you. Nobody read this

A real draft in the queue. Good replies, written to real messages, that nobody outside your account has seen.

01

Hire

Pick a role. Nothing to install, nothing to build, no call to book.

02

Teach

What you sell, your hours, and what it must never promise. Then point it at a channel.

03

Read everything

Two weeks answering your real traffic and sending none of it. You read every draft and mark the wrong ones.

04

Release it

You decide when. Every reply stays on the record, and one click stops it again.

Thirty-three seconds. Sound on, or read the captions.

Who each one works for

Not all of them work for a business.

Some answer strangers on behalf of a company. Some keep a household's week. Some work for a single named person and nobody else. These are not tiers of the same thing, and who a role answers changes what it may do, so each card names that first.

Replies are paid from credits on top of the monthly price, so a quiet month costs less than a busy one. You set a spending cap per agent.

Fourteen ways in. Your customers install none of them.

Nothing to set up Website widget
One value, and it is running Telegram · Webhook
An account you already have WhatsApp · Instagram · Messenger · LINE · Slack · SMS · Google Chat · Zalo · Twitch
A server you run Matrix · Mattermost
One channel reaches one agent, which reaches one queue. A second channel reaches a second agent. The two are not joined. A channel One agent Your queue Another channel Another agent Your queue
An agent answers on one channel. A second channel is a second agent, at the same monthly price as the first.

Where it runs is yours to choose

By default it runs alongside your other agents hired for the same job, and the skills you add reach all of them. For twenty dollars more a month it runs on a machine of its own. No other business ever shares a machine with you. Sharing here means your own agents, hired for the same job.

Instead of testimonials

What you can check before you believe any of it.

Every promise on this page is either something you can see in the first week or something you can read right now.

The send path is closed, not discouraged

A locked door, not a rule the agent was asked to follow. It does not hold the key.

Every reply is kept with the knowledge that produced it

A wrong answer has a cause you can find rather than guess at.

What it could not answer is a list, not a guess

Gaps for you to fill, and the most useful thing you will read in month one.

The engine underneath is not ours

Each role names the one it needs and we can swap it without you moving. You are not betting on one company’s AI.

The price is on this page and the cap is yours

No call to find out what it costs. A runaway month costs you the cap.

Leaving is one click, and your data comes with you

Knowledge, conversations and every reply sent, exportable the day you ask.

And what is not here. There are no logos on this page and no resolution rate, because we have not earned either yet. Our own rule is that nothing goes on the price list before it has answered real messages, and that rule applies to claims about ourselves first. The two weeks exist so you do not have to take our word for any of this.

Questions

Before you hire one.

The ones this page does not answer further up. If yours is not here, ask it rather than guessing.

Do I need a technical team?

Not for the widget, and not for Telegram, which is one token from @BotFather. Most of the rest is values copied out of a console you already administer. WhatsApp is the long one, because Meta asks that paperwork of every business, and you can leave it until after you have watched the agent work somewhere easier.

What does it cost when it is busy?

The monthly price covers the agent. Replies are paid from credits on top of it, and you set a cap per agent, so a runaway month costs you the cap rather than a surprise. A quiet month costs less than a busy one.

How long before it is answering?

Minutes to have it running and a few more to tell it about your business. On Telegram or the website widget you can watch it draft its first real reply the same afternoon. WhatsApp takes as long as Meta takes, because that paperwork is between you and them.

What language does it answer in?

English and Indonesian, in the same agent. It answers in whichever one it was written to, without being told to switch.

What if I want a role that is not on this page?

Then it does not exist yet, and we would rather say so. What is here is what has run on real messages. It can be built for you instead, on the same machinery and with the same fortnight before it is allowed to do anything.

How a custom role works

Hire one that already knows the job.

Tell it what it needs to know about you, and it starts. Where it answers strangers, you read two weeks of its work before anybody else reads a word of it.