We're building AI infrastructure, written as code.
For business owners who want AI to do real work and keep control. The HausAI Fabric is the standard we are building, in code, for every model to run through. We are building it so that only you can approve deleting or combining records. Open source first, then hosted.
Tell us what you're automating Ask about your data protection and security Just want updates?
This is a picture of what we are building. Nothing is released yet. You are the circle in the middle, with five agents around you, each in its own colour for a different model. One ring of five gates surrounds them, and the world is outside. Two agents send actions toward the world. Each meets a gate, which checks it, opens, lets it through and shuts behind it, so the rules sit outside the agent. A gate refuses an action, shown by a red bar and a cross, and the action goes back to you. An agent's finished work goes to a reviewer, a square standing apart from the agents, because no agent grades its own work; the verdict is recorded. An action that would delete something is held at its gate under a ring, and a request comes to you. Only your mark lets it out. Last, you pull one switch: a ring spreads from you, every gate shuts at once and everything holds, so you can stop it. Above you, small marks form the record: each decision adds one. It repeats in a loop. Click it or press Enter or Space to stop and start it. Stopped, it shows everything at rest: the record full, one gate refused, one held, all shut by the switch.
What it is
In the Fabric we are building, you give an agent a job, it works with only the access you grant, and the rules it must follow (its rails) sit outside the agent, as code that checks each action.
When your agent does something, it doesn't do it as itself. It does it as you.
Nothing is released yet.
When AI acts, who is responsible?
AI is not accountable. Humans are accountable for how we use it.
That's what the rules are for.
How we are building it
These are the principles we are building into the Fabric.
- The rules can't live inside the model. They have to sit outside it, in a separate part that only checks whether an action is allowed.
- No agent grades its own work: a separate reviewer looks at what actually changed, and that verdict gets recorded.
- Done means written down and checked by someone who didn't do the work.
- The important decisions get written down. We are building the same for every decision to override a check.
- A claim either carries its evidence or says plainly that it has none.
The aim: you can hand work to an agent and still know who decided, on what evidence, and how to stop it.
Three ways to reach us
Here from one of our posts? You're in the right place: we're building the Fabric so you can hand work to an agent and still know who decided and how to stop it.
Tell us what you're automating
One honest answer, no pitch.
A few sentences are enough, and it doesn't need to be a finished plan.
Ask about your data protection and security
One specific answer, free.
We reply personally. Keep it general, with no names and no real data; an outline is all we need. Practical, not legal advice.
Just want updates?
We write when there's something worth reading. Nothing more.
Sign up
Sending…
Thanks, you're in.
We will reach out when there is something worth reading.
Thanks, we got your message.
It is recorded and we read every one. We reply personally. You will not get an automatic email from us.
Ticket #…
- Please enter a valid email address.
- This address is already on the list. Nothing more to do.
- Too many requests from your IP. Please try again in a few minutes.
- Please check your inputs: the email or message looks incomplete.
- Send failed. Please try again later.
- Network error. Please try again.