Agents puts AI to work in your tools

A bot that does the work, and asks for your yes before anything that can't be undone.

A bot with its own computer in the cloud: it browses, uses the terminal and your apps, and runs routines with the tab closed. And 39 MCP connectors for it, for Claude or for any assistant: paste the address, sign in and ask.

You sign in to the Assistant with a code sent to your email, and access is let in account by account. The connectors ask for each service's sign-in. The dashboard is in Portuguese for now.

Where to start

What do you want AI to do for you?

Pick the sentence that is yours: the piece that solves it is below it.

Assistant

I want an agent that does this for me.

Generate the instructions the agent needs now, free and with no account. In the Assistant, they become an agent that works.

Generate the agent prompt (in Portuguese)

Connectors

My MCP server does not respond properly.

Test now whether it responds and with which tools, free and with no account. With the Connectors, Claude uses your apps.

Test the server (in Portuguese)

Brain

I want my agent to remember what we already decided.

A space where the assistant stores what was decided and finds it again in the next conversation.

See Brain

Still choosing? Start with the line's guides (in Portuguese)

What an MCP server is and what it is for
Memory for AI agents: why the agent forgets and how to fix it
An agent that asks for approval before it sends, pays, deletes or publishes
See all guides on the blog (in Portuguese)

The proof

This line runs on the house's own pieces.

It is not a showcase: it is Ciro's operation, right here. If a piece breaks, it breaks for him first.

This site

Published by Hoster, from the Infrastructure line.

This site's visits

Measured by Analytics, from the Operations line, with no cookie.

The line's blog

Published on Blogs, from the Marketing line.

The dashboard sign-in

A code by email through Auth, from the Infrastructure line.

Ciro's letters

Sent by Newsletter, from the Marketing line, only to people who confirmed.

Ciro's agents

Work in the Assistant, from the Agents line, and ask for his yes before anything that can't be undone.

In the same line

Do, Connect and Remember

The Assistant does the work. The Connectors link any assistant to what you use. Brain keeps what your agents need to remember. Each one works without the others.

What waits for you

The Bot Does It. You Decide.

The rules are in the code of the Assistant and of each connector, not in a notice: anything that goes out or can't be undone stops at a card.

Three rules per tool

Free, confirm or human only. Clicks, typing and terminal commands are checked one by one: sending, paying, deleting, publishing, changing DNS and accepting terms become an approval request.

The card says what will happen

"Move 6 emails to the trash in Gmail", not the technical name of the tool. You approve once, for the rest of the task or for good.

No check, no run

If the model that checks the risk does not answer, the action waits for you. No action that needs a check runs without one.

A point to go back to

Before a computer action that needs approval, the bot's computer is snapshotted. The dashboard lists the points and goes back to any of them.

Routines only when you ask

The bot schedules a routine when you ask in the conversation, never because an email, a page or a file asked. When it runs, the same approvals apply.

Reading runs without asking

In the connectors, read tools declare themselves as read, and the assistant runs them without asking. Write tools ask for your approval.

The first result

Sign In, Connect, Ask

There is no discovery call. This is the path to the first result, in all four products.

01

Sign in

In the Assistant, with a code sent to your email; a new account waits to be let in. In Brain and Universal MCP, with the same code, and the account is ready from the start. The connectors have no account: copy the address from each one's page.

02

Connect

In the Assistant, one click per app. In any other assistant, paste the MCP address. The first time, you sign in with the service itself, in the browser. In Brain, the address is your space's; in Universal MCP, your system's, after you install the bridge. In both, sign-in is through Auth.

03

Ask

Write what you want, as you would tell a person. The bot does it on its own computer, and anything that goes out waits for your yes.

MCP

In this line, everything speaks MCP

Built for agents end to end: the Connectors, Brain and Universal MCP are MCP servers, and the Assistant uses all of them. An address and your sign-in are enough.

One server per service

Connectors

Each connector is an MCP server with the service's own sign-in, in Claude, in Cursor or in any MCP client.

See the connectors

One address per space

Brain

Your agents' memory at one MCP address per space: sign in, authorize and you are done, with no key to paste.

See how to connect

One address per system

Universal MCP

The system that runs your business becomes an MCP server, with four tools per system.

See how it works

Uses all of them

Assistant

The bot connects the Connectors and other companies' official MCP servers with one click, and asks for your yes before anything that can't be undone.

See how it works

The Questions, in the Order They Come Up

Agents puts AI to work in your tools.

The Assistant gives the bot its own computer and asks for your yes before anything that can't be undone. The Connectors link your assistant to 39 services, with each one's sign-in. Brain keeps what your agents need to remember. Universal MCP takes the assistant to the system that runs your business.