Skip to content
System onlineAI Command Core

Under the hood

From your voice to the computer doing the work

You ask for something from the web, your phone or your watch and the result comes back to you: a file, a spoken answer or the job done. Behind it there is a list of pending jobs, several computers sharing them out and, on each one, an agent with real tools. This is the full path of a request.

  1. 1. You askFrom the web, phone or watch. Typed or spoken: “Hey AMSET”, shaking the phone or tapping the orb.
  2. 2. AMSET decidesChecks your permissions, answers simple things straight away and picks the computer and project based on what each can do and how busy it is.
  3. 3. Onto the waiting listThe task is saved and, meanwhile, it says a short line so you know it is on it.
  4. 4. A computer picks it upAn agent does it with access to files, programs and tools, only inside the folders it is allowed into.
  5. 5. Back to youYou see what it is doing live and get a spoken summary, buttons to decide and the files it produced.

Messages in the same conversation are handled in order, one after another.

The computers

Computers that share out the work by themselves

Adding one means switching it on

Every few seconds each computer reports what it has (processor, memory and graphics card) and what it can do. To add one, you just start the program. If more power is needed, one can also be created in the cloud.

Each job goes to one that can do it

The conversation goes to a computer able to handle it: the one with WhatsApp linked, the one with a graphics card or the one that looks after that project. If they are all busy, the task waits its turn instead of failing.

Several things at once, without clashing

Each computer can run several tasks at the same time. If two work on the same project, the second one uses a separate copy. Two messages from the same chat are never processed at once.

They pass jobs to each other

One computer can ask another for something, and the result comes back by itself to the conversation that asked, whether it went well or badly.

They fix themselves

If a process goes down, it restarts. Updates never cut a task off halfway, abandoned tasks are cleaned up and, when an account runs out of quota, the next one takes over. On top of that, the connections to other services are checked every day, with a warning if one stops working.

Cancel really stops it

When you cancel, the computer stops the task and everything it had started. It disappearing from your screen is not enough.

What AI is underneath

Several models, each doing what it does best

No AI model is best at everything, and some data should not leave the company. That is why AMSET combines several.

  • For anything that must stay inside, a model installed on our own computer with a graphics card sorts requests, chats and shares out the work. There is no per-query charge and the text never leaves the network.
  • Voice and image reading run on that computer too: transcribing, speaking and understanding images. The cloned, emotional voice of the replies does go through the cloud.
  • Each task picks which AI does the work: Claude, ChatGPT (Codex), Gemini, OpenRouter or the in-house model. The Claude accounts take turns by themselves when one hits its limit.
  • For the heavy lifting, a coding agent that works on its own: it reads, writes, runs and publishes changes.
  • Cloud models as backup, for web searches with cited sources and for creating images.
  • A council of AIs: the same question to Claude, ChatGPT and Gemini, and a single plan built from what each one said.
  • Its own memory, which finds things by meaning rather than exact words, kept separate between each project and each user’s personal notes.
Illustration: several small computers and a server connected by light beams to a central core

While it works

You see everything it is doing

Every step the agent takes is saved and shows up straight away: which tool it uses, what it gets back, when it is thinking and when something fails. The app panel also shows the connected computers, how busy each one is and which task it is on.

See the interface

This is the kind of thing we build

What drives AMSET under the hood (work queues, agents, live results and automatic publishing) is something we can apply to your product.

Let's talk about your project