A new foundation. Every enduring breakthrough rests on first principles. For webAI networking, those principles define how a sovereign AI network should work.
Addressability. Humans, devices, and content must have stable identities across contexts—whether at home, in the cloud, or on the move.
Interoperability. Change happens gradually, not all at once. Bridges into existing systems let organizations adopt at their own pace and unlock more value over time.
Pluggability. Real-world environments are messy. A pluggable architecture adapts to different protocols, topologies, and security needs while staying compatible with today’s stacks.
SaaS-but-Local. The network must “Just Work™” like SaaS—yet remain under your control. Convenience and sovereignty together, not either/or.
The result. A sovereign AI network where any model, dataset, or device can discover, converse, and evolve—without surrendering control.
From our First Principles Series: Essays on the foundational ideas and practices that guide how we build at webAI
Every enduring breakthrough—be it a civilization, a skyscraper, or a technology platform—rests on a handful of inviolable tenets. These first principles are not design options you can swap later; they are the hard constraints that shape every trade-off that follows. At webAI, our networking fabric stands on four such tenets:
These guide each protocol we adopt, every line of code we ship, and every future upgrade we imagine. We may compromise on features or timelines, but never on these principles. Together they form the pillars of a sovereign AI network where any model, dataset, or device can discover, converse, and evolve—without surrendering control.
The internet was built to address computers. But that’s not how humans think.
You don’t want to send an email to your grandma’s laptop. You want to send an email to your grandma.
Even if you did want to reach her laptop, you probably couldn’t. Routers, firewalls, dynamic IP addresses, and device mobility all break global addressing. Your laptop’s IP address changes constantly and usually isn’t routable from the wider internet.
So the way we’ve made the web work is by pushing everything important—our data, our software, our communications—onto stable machines in datacenters. Datacenters provide globally routable, unchanging addresses. That’s the hidden constraint that has shaped the way we build, ship, and use software today.
But we believe the future of AI is local. Personal AI agents should live with you, accountable to you, on your own hardware. Which raises new questions: when you’re at home having a conversation with your AI running on a MacStudio, how do you pick it up at the coffee shop? When you walk into a space with ambient intelligence, how do you discover it and interact with it? What does a “share” button mean when the thing you’re sharing is an AI chat that lives only on your phone?
This is why we say we’re not building “web3.” The web itself is doing fine. We’re building something different: internet2. A new addressing layer where humans, their devices, and their content are all first-class citizens. Where your “address” as a person is the same across devices. Where a file, a chat, or an AI model has a single stable address no matter where it’s stored. Where a device is reachable no matter where in the world it moves.
But shifting the foundation of the internet isn’t a binary moment. You don’t wake up one day and move from the old paradigm to the new. It’s more like steering a ship: Small adjustments to the rudder angle can gradually, but dramatically, change the course of the vessel.
People, companies, and governments already have systems they depend on. Databases that assign employee IDs, passports and social security numbers that identify citizens, web servers that assign URLs, compliance frameworks like SOC 2 or ISO 27001 that codify practices. These are deeply entrenched, and they aren’t going to update overnight.
That’s why interoperability is essential. We don’t expect people to jump wholesale to our way of doing things. We meet them where they are, and we build bridges. Our networking layer doesn’t demand to always be the single source of truth; it can wrap, defer to, or integrate with what’s already in place. Over time, as the value becomes obvious, the switch flips—those systems begin deferring to this network instead.
In this sense, interoperability is the rudder angle of the sovereign AI network. Minor adjustments for big impact down the line: It’s how we shift course without asking the entire world to move all at once. It’s how we build progressively, allowing customers to adopt just what they need now, and unlock more value as they go deeper.
In an ideal world, we could just pick a clean stack—say ed25519 over QUIC with our new routing layer—ship it, and be done.
But the real world is messy.
There’s a dizzying number of networking protocols, topologies, and security frameworks, each with its own constraints. That makes it impossible to build one universal approach to providing stable addresses for humans, their devices, and their content.
So we have to design for mess. We need a network that works in your living room, in an air-gapped aircraft hangar, or on the front lines of a battlefield—and where your device address stays the same across all three. The only way to get there is a pluggable architecture.
Our encryption curves are pluggable: ed25519, RSA, P256—we can support what your requirements demand.
Connectivity is pluggable: IPv4, IPv6, QUIC, WebRTC, TLS, even Tor.
Routing strategies are pluggable: different network environments call for different approaches. NAT traversal, VPNs, or private-only routing can all be swapped in or out depending on security needs.
This isn’t just flexibility for its own sake. It’s the only way to reliably serve networks with wildly different constraints. Pluggability is how we make purpose-built networks that remain compatible with current stacks while adapting to whatever the world throws at them.
Finally, accessibility. Local and self-hosted software has existed for decades, but let’s be honest: it takes a skillset and time commitment most people don’t have.
My grandma isn’t going to self-host a deployment. If we demand that, she’ll swipe her card with a cloud vendor. And she won’t be alone. If local AI requires sysadmin skills, adoption stalls after the early enterprise crowd.
That’s why our real competition isn’t other local software—it’s SaaS. The bar is Gmail, Google Drive, ChatGPT. These things Just Work™. And that’s the experience we’re building: the convenience and usability of SaaS, but delivered locally, under your control.
When we design our network—how it stores, queries, and moves data—our benchmark is the SaaS experience. We aren’t trying to recreate self-hosted software; we’re building something that feels as smooth as the cloud, but runs under your roof.
Because if we ever force people to choose between convenience and sovereignty, many will pick convenience. So we don’t force the choice. We deliver both.
These four principles don’t sit in isolation. They build on each other.
Addressability makes local AI possible—it gives humans, devices, and content stable identities.
Interoperability ensures those new identities can live alongside the systems the world already depends on.
Pluggability makes the whole thing adaptable to any environment, no matter how constrained.
And SaaS-but-Local makes it usable by everyone, not just experts.
Together, they form a coherent worldview: A sovereign AI network where intelligence lives with you, not in someone else’s datacenter.
