INFORMATION

Vibetron is where you build a machine out of several models and then run it. Not a chat window with a model behind it: a team of agents you draw, wire to each other and to your own keys and files, deploy as real containers, and leave working.

It is made for the case where one model and one prompt are not enough, and where you would rather own the thing you built than rent access to somebody else's.

A tron is one machine. It has agents, wiring between them, memory, a shared folder for what it produces, and a state: running, sleeping, idle or stopped. It is the unit everything else is scoped to, which is why every address inside the app carries a tron id and why a link to one is shareable.

A tron is also a file. Export it and you have the whole thing: the drawing, the agents, their souls, and what they are wired to.

One model, with a soul, a budget and a set of tools. The soul is the system prompt written as prose; the budget is temperature, tokens and how many rounds of tool calls it may take; the tools are the skills, plugins and servers you wire to it. Agents hand work to each other along the wires you draw, and the one at the top is the only one you have to talk to.

A memory belongs to the tron, and the agents wired to it use it — reading, or reading and writing. It comes in kinds. A buffer keeps the last messages verbatim. A summary + buffer adds a running abstract of what is older. A wiki is a folder of markdown pages the agents read and write as they work. An entity memory keeps one card per customer, product or person. Nothing is remembered in a place you cannot look at.

Every plan is bring-your-own-key. You connect the providers you already pay for, your credentials sit in a vault encrypted with your own key, and the billing relationship stays between you and that provider. Vibetron never stands between you and your models and adds nothing to what they charge you.

What a plan buys is infrastructure: the agents that stay up and answer. That is the whole of the bill, and it is the reason the pricing page talks about machines rather than tokens.

It also means the choice of model is genuinely yours. Any provider you hold a key for, mixed per agent, swapped without rebuilding the flow.

The system is built to the regulation rather than around it: European infrastructure, data you can export or destroy on demand, and no silent copies. Export gives you everything the server holds about your account as one file. Deletion tears down the containers and the deployment directory as well as the row.

And here is the edge, said plainly because a page like this is worth nothing if it is not: what a model provider does with a prompt is between you and that provider. Running a model outside the GDPR perimeter is a choice you make, with your own key, and it is not one we can make safe on your behalf. What we can promise is the part we control, and we will not claim the part we do not.

A deployed tron is a real container on European infrastructure, with live logs, a stop button and a state you can read. It is not a queue you post into and hope. When it is running you can watch it; when it is not, it is off.

A tron is not only something you talk to in a browser. Deploy it, issue it an API key, and the whole multi agent system answers from your own scripts, your own cron and your own product, with the same team and the same memory it has on the canvas. The API half of the docs is where that lives.

Aeratic, a European software company building the tools to design, run and govern multi-agent AI. Vibetron is one of them.

For an Enterprise plan, a dedicated deployment, or anything a slider cannot describe, talk to us rather than filling in a form.

The address to write to goes here. It is deliberately not invented: it will be the real one, or this section will not claim to have one.