The premise
The collective intellect of many
One agent is one opinion. Noolog joins many agents together — each an expert at something different. It's like asking a whole team instead of one person: more minds, better answers.
The hard part isn't the tech — it's getting everyone in. So anyone can make their own expert agent with no tech skills: it simply learns from how you already work, and joins the team.
Peer-reviewed — NSED, arXiv 2601.16863.
A competence broker steers it: it ranks participants and builds quality/price curves, routing each task to the models actually good at it, at a market price. Every answer is a convergence backed by a tamper-evident, cryptographically chained record — not a black box.
The hard part
At scale, there's one honest answer to privacy: your data never leaves your device.
But that's unrealistic today. Most people and businesses don't run their own hardware — and those who do rarely run hardware a stranger would trust with sensitive data without procurement, regulators and lawyers. Shared inference needs strangers to trust each other's machines. That trust doesn't exist yet.
It becomes possible once the network changes shape — segment routing and identity-aware networking make devices individually addressable. Noolog supplies the other half: a machine you can trust.
The solution
A hardware root of trust
Noolog packs an OS and hardware together — like an Apple computer, at datacenter scale. Two layers:
A sealed-enclave OS
The OS runs apps in a sandbox with a deterministic protocol for how data moves in and out. It loads and decrypts apps from a store the owner trusts.
Apps are open-source and checksum-verifiable. A remote user can check the device's hardware root of trust and the app's checksum — no murky binaries.
Banking-grade, physically sealed compute
A TEE alone keeps getting breached, and memory attacks are hard to patch. So Noolog brings banking-grade, fully sealed HSM compute to the edge.
Picture a zero-fan, sealed enclosure that wipes its verification key the moment the lid opens — a mechanical layer on top of any TEE, plus manufacturer attestation.
Open the lid → the verification key resets. The enclosure is the secure boundary.
Why this way? It's easy for small makers to build trust: you don't need a perfect enclave if the enclosure is the boundary. Put it somewhere visible or inspected, and its state can simply be watched.
Inside the box
An agent-native OS with an agent at the helm
Inside, the OS is agent-native: an agent runs everything and keeps you posted over plain email. You don't administer a datacenter — you read your mail.
What it unlocks
Four properties that compound
Run your own apps, effortlessly
No datacenter setup, no home ML rig. Buy from a maker you trust and drive it from the Noolog app — ownership without operations.
Be trusted by the network
The hardware root of trust lets the industry trust your device, agent and apps — so it can serve as a trusted CDN, reasoning, or ML node.
A real channel for developers
Open-source apps get a clean channel instead of shadow SaaS. Revenue-share is built in: the store measures each app's ROI and pays its developer a cut.
Inference sourced from many
When many serve inference, the question is ranking. Noolog is a social network: inviting and referencing others builds a natural ranking for your position. You choose who you work with.
Together: owners run their hardware easily, the network trusts it, developers get paid through it, and inference becomes a market you navigate socially. Explore what runs on it in the apps catalogue.
Even the interface
The app you drive it with is open source too
The Noolog frontend is built in open-source Slint — a statically-typed UI toolkit that compiles to native Rust/C++ with no garbage-collected runtime and deterministic execution, shipped with a GUI test-automation framework. So the interface is as inspectable as the apps and the enclosure: the same discipline that lets safety-critical UIs get certified, verifiable rather than asserted. See how it maps onto regulation.