Not “platform capabilities” — what you're busy with today.
The same ten questions every day: “how much is it”, “is it in stock”, “when will it arrive”. It answers from your materials — with your prices and your terms. When it doesn't know, it says so and offers to call you, instead of making up a price you'd then have to stand behind.
Adds the customer to your database, logs the request as a row in a Google Sheet, and sends you a Telegram message when a conversation matters. You describe what it's allowed to do — it decides the rest on its own as the conversation goes.
Wakes up on a schedule — every morning at 9:00 or every hour — does its job and puts the result where it was told: in an email, a chat, a sheet. This kind of assistant has no one on the other end: it doesn't wait for you to write something.
Seven presets — from a website assistant to a search agent. Not thirty you'd still have to choose between.
Answers visitors about products, delivery and terms; collects contacts and calls in a real person when it doesn't know.
Writes a reply and then rereads it: did it promise too much, and did everything come from your materials rather than out of thin air. The customer sees one reply — already checked. Costs about twice as much as a regular one — and the Builder will warn you about that before turning it on.
A table at 19:00 is available. 10% off for new guests.
A table at 19:00 is available. Shall I book it?
Wakes up on a schedule, does its work, and answers no one — there's no one on the other end. For nightly reconciliations and morning summaries.
Nowhere to reply: this assistant has no one to talk to.
A template gives the personality and the way of working. Your details — addresses, keys, names — the assistant fills in during the conversation: there's no way to guess them in advance, and we don't pretend we can.
It knows exactly what you gave it. No more, no less. Three ways to teach it, all equally good — pick whichever suits you.
Give it an address — it'll read the page, or the whole site from its sitemap, and sort it into knowledge itself.
Mon–Sun, 8:00–21:00
Espresso — 45 UAH, cappuccino — 65 UAH
Tables are booked for 2 hours
PDF, DOCX, XLSX, CSV, TXT, Markdown — up to 10 MB each, one at a time. An Excel price list is read as a price list, not as a jumble of characters.
Item: Espresso; Price: 45 UAH
Item: Cappuccino; Price: 65 UAH
Item: Latte; Price: 70 UAH
Just write what isn't written down anywhere: opening hours, return policy, what to say about discounts.
A brand-new assistant asked “how much is delivery?” will honestly say it doesn't have that information. That's not a bug but a safeguard: an honest “I don't know” beats a made-up price you'd then have to stand behind. Add material with prices — and the answer will appear.
A request, a booking, a row in a spreadsheet — right in the conversation with the customer. Four integrations are already set up: all you provide is the access details.
Adds the customer to your database: the name and phone from the conversation go into their card.
Sends you messages: a new request, an important question, a summary. It's not a bot for customers, it's a line in your own chat.
Adds a row to your spreadsheet. You'll need to publish a small script on the Google side once — step-by-step instructions are in the action itself.
Sends an event to any address of yours: Zapier, Make, n8n or your own service.
{ "event": "new request", "details": "Oksana Hryn, +380 67 000 00 03, table at 19:00" }
Then you describe the action you need in words right in the conversation — or give a link to your service's documentation, and the assistant will pull the list from it on its own. This is the part where you'll have to do some work or call whoever set this system up for you. We don't pretend it connects by itself.
Once it creates an action, it immediately offers to run it once with sample data. Say yes: an untested action is a bug that will only surface in a conversation with a customer.
Request created, number 1043
Besides answering and taking actions, the assistant can simply do the work: process data, put together a report, work with your repository.
The limits are stated up front: the environment lives for exactly one run, internet access for scripts is limited to package registries and repositories (pages are read by a separate browser), and a single step is capped at five minutes. Files aren't kept between runs — anything that needs to outlast a run goes into the notebook, the repository, or to you as a file.
The same assistant can be personal: talk to you in your messenger, remember what you've told it, know your email and calendar.
SKILL.md from a link. It starts acting on it only after you click “Save”.An assistant that writes to your customers is not just a convenience but a risk. Here's what keeps it in check: not a polite request in the agent's personality, but a limit it can't get around.
Not from a customer who hasn't gotten a reply for three days.
An assistant that works on its own breaks quietly. The key to your CRM expires, someone else's service stops responding, the credits run out — and from the outside everything looks fine right up until someone opens the log.
So we tell you about it ourselves — in Telegram, by email, or to your service. And not just “an error occurred”, but what exactly to do: “the key expired — enter a new one”, “top up your balance”, “someone else's service is down, nothing to do here”.
One cause, one message — no matter how many times it repeats. Two hundred identical alerts overnight isn't care, it's a way to teach you to ignore them.
The Builder's MCP server: Claude Code or any agent client sets up the bot through the same conversation.
This section is for those who already use Claude Code or another AI agent. If that's not you, just keep scrolling: the dashboard doesn't need any of this.
Everything the Builder does in the dashboard, it also does for your agent: add knowledge, connect a channel, change the personality, set a schedule. One tool — constructor: no list of methods to learn — you just say what to do.
Connecting takes a minute: the agent client will bring you to the dashboard to confirm access itself, or you take a Builder key on the “Account” screen. The key can manage, but it can't pay or delete — those are its limits, not settings.
And yes, the bot can manage itself: connect this same MCP to it, and it will ask the Builder to add knowledge it was missing in a conversation.
claude mcp add --transport http youselfbot \ https://api.youselfbot.com/v1/api/mcp > via youselfbot, add knowledge to the bot about the delivery schedule ✓ Builder: “Done. Saved the schedule schedule to knowledge.”
Free plan: one assistant, 10,000 credits a month, no card.