A research program in federative agentic AI, built in the open — and why it puts the human back at the center.
Noogram is an attempt to build, for artificial intelligence, the kind of shared instrument that science already knows how to build for physics or astronomy: a single instrument that many actors fund and use together, that none of them owns, and whose every result returns to the common record. Not a company’s lab, not a national champion — an instrument.
There is a quiet assumption underneath most of today’s AI: that the technology’s job is to replace the human — to take the work, to automate the task, eventually to build its own successors with the human pushed to the edge, reduced to oversight.
Noogram is built on the opposite bet.
The human is not displaced by the system; the human is amplified by it. Federative agentic AI, in this program, is not a substitute for an operator — it is an extension of human judgment: a companion to attention, a calculating prosthesis for discernment. The final decision, the audit, the signature, the veto, the deliberation — these stay human, by design.
There is a simple test for whether we have succeeded: remove every human operator. If the system keeps running on its own, we have failed. If it stops, we have succeeded. A race toward a system that runs without people is, for this program, the failure mode — not the goal.
Everything the program learns is published before it is valorized. Not after, not maybe — before. The chronology itself is the safeguard: a program whose record is public from birth cannot be accused of having privatized what public research paid for, because the timeline forbids the reading. Open is the invariant; closing would be the error.
This is why the work will be released under copyleft licenses, where openness is a matter of enforcement, not of goodwill — and why every artifact is timestamped against an independent record, the way an open software archive makes public anteriority verifiable by anyone. It connects to the one hard problem the whole field is now circling: not capability, but verifiability — when an agent acts and a human reviews, what is the smallest record that makes the act checkable by anyone, with no one to trust? An instrument built in the open answers that by construction.
The model to hold is not a company’s research lab. It is closer to a single shared instrument that many nations build and run together, and that none of them owns. Real-world domains come into residency on the instrument — they are the experiments that make it run — but they are not the product, and they do not take the instrument home. What they leave behind, on the way out, is substrate: published results, shared data, a signed and timestamped record.
The point is not to be against anyone. It is to reverse an arrow. Today, value flows from the commons into private hands — public research becomes proprietary product, public talent becomes private advantage. This program runs the arrow the other way: private effort, temporarily resident, leaving a commons behind it.
The instrument is meant to outgrow whoever starts it. Today there is one person at the center, for the simple reason that there is only one person. The design goal is the opposite of that: a federation where validation always comes from someone outside the work being validated, where the center fades as the network grows — less by decree than by the way a mycelium spreads. Membership is earned by concrete output: a dated, public artifact that does real work, not a stated intention. The aim is decentralization not as a slogan but as a robustness property — a system whose guarantees come from its members, not from a guardian.
This is a trajectory, not an achieved state. Saying otherwise would be the one contradiction the design is built to avoid: a single guardian decreeing that there are no guardians.
Two conversations are converging. In Europe, voices are calling for a shared, public instrument for AI — built together, at the scale of the biggest science the continent has ever pooled. Noogram resonates with that call but is not bounded by it: it might be hosted in Europe for historical reasons, while aiming to be global, not the instrument of any single bloc. In parallel, the most safety-minded labs are warning that AI is starting to build itself, and asking how the world could ever verify a coordinated slowdown when training runs are easier to hide than missile silos.
Noogram is one possible operational answer: an independent commons — open by construction, federative by design, human-centered by test, verifiable from the first commit. Not a project asking to be absorbed, and not a manifesto about what should be — a thing being built, in the open, that you can watch.
If that’s interesting, we look together.
An agentic-AI project, built in the open. Instruments for human cognitions. To be released under open licenses; this is a report on a program under construction, not a finished claim.