PowerButton.ioPowerButton.ioThese are full user agents — the same class of thing the other platforms sell, and not a website builder with a chatbot bolted on. Each one has its own key, its own space, its own memory and its own schedule; what it may reach was declared by the listing it was installed from, and the store refuses it everything else — instead of the model agreeing to behave. Your panel shows what each one did, and what it cost. You do not need a website for any of that: start free and your agents are there, whether or not you have a site with us. If you do host a domain here, your Builder can write that site and watch it as well — a convenience for people who want both, never the price of admission.
Below is what people are already pointing them at — the obvious, the useful, and the ones that make somebody say “I did not know it could do that”. It is a starting list, not a limit.
Better is not a boast about a benchmark; it is about four things you can check. Limits that are enforced: an agent can only see what its listing declared, and the store refuses the rest in writing — not because we asked it nicely. A memory that is a record: everything an agent does is written down, so “why did it say that” has an answer. One bill, and a choice: our models at $1 per million tokens from prepaid credit (the default), or your own OpenRouter account for $5 a month and their whole catalogue, billed by them. And agents you can edit — more to do or less, at any time, by you, your Planner or your Builder — instead of a black box that only its maker may touch.
Every one of these is a job an agent does on your behalf, with its own space, its own key and its own history. Where the platform does not act by itself — sending mail, moving money — the entry says so, because a promise a tool cannot keep is the one thing worth less than nothing.
Your assistant watches the things you already own: the alarm, the heating, the door sensor, the server room thermometer. Readings arrive as records with the time on them, so “when did the freezer warm up” is a question with an answer instead of an argument. It reaches your kit through a machine you register with us, and only the commands you allow.
A camera agent watches a feed you own and writes down what matters — a van at the gate at 2am, the yard light on all day — with the moment it happened. Footage stays on your kit; what reaches your hive is the note, not the video, unless you decide otherwise.
Temperature, humidity, power draw, tank level, footfall: an agent reads the sensor history, spots the drift before it becomes a breakdown, and tells you in a sentence. Thresholds are yours, and it says which sensor crossed which one.
Your Planner and your Builder write the site — your pages, your words, your forms — on the domain you host with us, to your specification, and put it up when you say. Then they watch it: what changed, what broke, what a visitor asked.
A campaign that runs itself: built once, sent on your schedule, written per customer from what your own records say about them. It is fully automated and built not to be spam — only to addresses you allowed, one message per person per week, a daily ceiling, an unsubscribe link in every one, and a bounce that takes the address off the list for good. Your ordinary mail to customers is not bound by the marketing rules: the weekly gap is for campaigns, a reply to somebody who wrote to you needs no list at all, and one person may hear from you up to five times a day — enough for a real exchange.
Point it at the mailbox you own: it reads what arrives, works out who should handle each one, drafts the reply in your voice, and puts the job where the right agent will find it. By default everything it writes waits for you — one press in your panel and it goes. Lift the gate and it answers by itself, through your own mail server, to the people who wrote to you and to the addresses you allowed — and to nobody else — with every message kept so you can read what was said. A busy day is hundreds of replies; that is ordinary mail, and the marketing rules do not apply to it.
A question that needs ten sources and an afternoon: an agent gathers it in parts, keeps the address and the time of everything it read, and assembles it into a deck or a web page you can present — with the sources named under each claim.
The messy export, the stock list, the job costs, the contact book nobody trusts: an agent sorts it, fills the gaps from your own records, and hands back a sheet you can hand to an accountant.
One reminder per unpaid invoice — which invoice, what is owed, when it was due — drafted, kept in your workspace, and left for you to send.
What your work has cost, what you are subscribed to, and what your customers owe, read from your own Stripe with a key you create.
The shoebox becomes categories and totals, with the ones it could not place left out in the open rather than guessed at.
What is due in and out over the next thirty days, and the week the two cross.
The same quote at your rates, from your own price list, in the time it takes to answer the phone.
Your product list brought up to date from your own documents, with what changed marked.
The question a customer asked, answered from what you actually wrote — policy, manual, warranty — with the document named, and “it is not in what I have” as a real answer.
The case, the timeline, the drafted reply that does not apologise for the wrong thing, and the rule for when to escalate to you.
The diary kept, the confirmations and the reminders drafted, and the customer who always reschedules noticed.
What was said about you this week, where, and a drafted reply for the ones that need one.
The welcome, the details you need, the first job booked — the same every time, and never forgotten.
Eligible work found and read against your criteria, with the deadline and what is missing from your bid written down.
What a contract says, what changed between two drafts, and what is missing before you can decide — with the clause quoted.
What happened on site today, in the words and pictures you send, kept so next month’s question has an answer.
Licences, certificates, insurance, inspections: what expires, when, and the paperwork that has to be ready first.
Who is on, who is off, who to ring at short notice — and the message drafted for them.
Incidents logged as they happen, patterns found across months, and the report for the inspector drafted from the records.
The same part from three suppliers, their price this month against last, and what a rival quietly changed on their site.
What is running low against what you actually use, and a reorder list before it stops the job.
Who said they would do what, by when — and the list that survives the meeting.
Applications read against the requirements you wrote, with the reason for each decision kept in writing.
The manual your staff can actually follow, written from how you really do it.
The eligibility read first, then the draft, then the deadline — and a straight answer when you do not qualify.
The pack a buyer or a bank asks for, built from your own records, with the gaps named.
Every answer an agent gives, and every document it read, kept as records you can search — so the memory does not walk out of the door with anyone.
Facts, decisions, documents and what it read on the web: all records, addressed and searchable, so the agent that helped you in March still knows what it learned.
Each agent gets exactly the access its listing declared — the roster, your ledger, your domains, the web — and the store refuses anything else, with the refusal written down rather than an empty answer.
Not in a tab you have to keep open: a schedule is a record, and the platform runs it — every morning, every hour, or the first of the month.
Any agent you own can be edited to do increased or decreased actions, at any time, by you, by your Planner, or by your Builder.
The workspace is shared: the triage agent drops a billing case where the bookkeeper finds it, and a research question where the web agent does. You can read the same workspace from your panel.
Every call is metered and written to your ledger, so “what did this cost” is a query rather than a guess — at $1 per million tokens, or on your own OpenRouter account if you add the module.
If you can describe the job — what to watch, what to gather, what to draft, how often — your Planner will build the agent that does it. Tell it in your own words; it will ask what it needs, and it will tell you plainly what it cannot do.