LINKSPREED Expands Its AI Lab: More Models, Autonomous Agents, and a New Rule — Only Verified Identities Get Into Web4
LINKSPREED is expanding its AI lab with more models and autonomous agents, and extending its UIID identity system so every agent operating inside Web4 can be traced to a verified identity.
United States — October 2026 — LINKSPREED, the company building Web4, today announced it is significantly expanding its artificial intelligence work. The company plans to grow its AI lab, release more models, push further into autonomous agents that can act on behalf of users and communities, and tie all of it to its identity system, UIID, so that every agent operating inside Web4 can be traced back to a verified identity.
The backdrop: a new kind of internet
To understand why this matters, it helps to understand what LINKSPREED is actually building. The company’s stated mission is Web4: an internet where ownership sits with the individual, not with large platforms. Today’s internet largely runs on borrowed infrastructure. People don’t own their accounts, their data, or increasingly, the AI agents that act on their behalf — platforms do. LINKSPREED’s answer has been to build that infrastructure from the ground up, layer by layer: identity, connectivity, computing, naming, software, applications and hardware, each designed so it can be traced back to something a person or organisation directly controls. Its flagship offering lets people build and run their own social networks instead of depending on big technology companies, and its identity system, UIID, sits underneath all of it as the entry point every user and, increasingly, every machine must pass through.
Growing the AI lab
AI has quietly become one of the more active parts of this effort. LINKSPREED publishes its models openly on Hugging Face, under an organisation simply called Web4, describing its focus as AI for social networks. The lab has already shipped close to fifteen public models: systems that write social media captions, generate hashtags in real time, detect emotion in text, score how relevant a post is, support content moderation, and flag hate speech, among others.
What makes this lab distinctive is its philosophy: instead of one giant general-purpose model, LINKSPREED builds many small, specialised ones, each trained to do a single job extremely well, and each small enough to run directly on a person’s device rather than in a remote data centre. Its caption-generation model, for instance, runs entirely inside a web browser with no server involved at all — what the company calls “serverless operation” — keeping everything private, fast and functional even offline. Its newest generation of models, built on Google DeepMind’s Gemma 4 family, goes further still: multimodal systems that handle text and images, support more than 140 languages, and are tuned specifically to run efficiently on phones and laptops instead of requiring cloud infrastructure.
Today’s announcement signals that this is just the starting point. LINKSPREED says it intends to expand the lab meaningfully: more models, covering more tasks, released at a faster pace, while holding onto the same principles that have defined its work so far — small, local, inspectable, and privacy-first by design.
From models to agents
The second part of the announcement is about agents. Over the past year, LINKSPREED has moved from AI that simply generates content to AI that acts. Earlier this year, the company integrated Google’s Gemma 4 model with OpenClaw, an open-source agent framework, inside ATRIUM, its platform for building social networks. This created autonomous agents the company calls “CoHosts,” capable of handling community moderation, knowledge management and more complex, multi-step workflows on behalf of the people and communities that deploy them — with both network operators and individual users retaining control over whether and how these agents get involved.
LINKSPREED now says it is accelerating this push into agents. The company has recently rolled out a wave of new agents built on its platform, extending beyond community moderation into broader tasks across its ecosystem, and it intends to keep building more.
The identity problem agents create — and LINKSPREED’s answer
Agents that can act — post, moderate, negotiate, manage, transact — create a problem that the rest of the internet still hasn’t solved: how do you know who, or what, is actually behind an action? Historically, software agents have operated using borrowed human credentials or static API keys, sitting in a legal and technical grey zone that LINKSPREED has publicly described as a “ghost agent” crisis.
LINKSPREED’s answer is to extend UIID, its identity system, to cover machines as well as people. Since UIID version 2.8, AI agents, hardware devices and organisations can each hold their own cryptographically verifiable digital passport, separate from any human user. Agents can register for these passports through a machine-readable protocol, and a tool called the Agent Factory gives individuals and organisations a single place to issue access tokens, launch new agents, and monitor their activity and trust scores over time. Long-lived secret keys are replaced with single-use or limited-use tokens, so that even if one is intercepted, it can’t be used to compromise an entire fleet of agents. UIID currently recognises six distinct types of identity, covering humans, organisations, AI agents, hardware nodes, digital services and collectives, each with its own enrolment process.
Today, LINKSPREED is drawing a hard line from this groundwork: agents operating inside Web4 should carry a verified identity, not operate anonymously or on borrowed credentials. The company frames this as a matter of basic trust — the same reason, it says, that only identity-verified businesses are allowed to buy ad placement on its discovery platform, HELION, a rule designed specifically to eliminate fraudulent activity. Extending that same logic to AI agents is, in the company’s words, simply the next step.
The underlying identity system that makes this possible was itself built with privacy as the starting point. UIID splits a person’s presence into a master “Core ID” plus separate, context-specific aliases, so that no platform can link someone’s activity across different services. Identities are secured locally on the user’s own device using biometrics and hardware-level security standards, and verification relies on zero-knowledge proofs — a way of proving something is true without revealing the underlying data itself. The combination is designed to make it both impractical and unprofitable to run mass bot networks or fake accounts at scale.
A statement from the founder
“We didn’t build an identity system just for people and then bolt AI on top as an afterthought,” said Marc Herdina, founder of LINKSPREED. “We built it so that anything acting on this internet — a person, a company, or an agent — has to be able to prove who it is and who stands behind it. As we put more AI into the hands of our communities, that’s not optional anymore, it’s the whole point. An agent without a verifiable identity shouldn’t get to act like it has one. That’s the internet we’re building, and it’s why we’re investing more in both the models themselves and the identity layer that keeps them accountable.”
What comes next
Taken together, the announcement points to three parallel efforts moving at once: a growing library of small, specialised AI models built for privacy and on-device use; a rapidly expanding set of autonomous agents doing real work inside LINKSPREED’s products; and an identity system, UIID, extending to cover every one of those agents so that capability never comes apart from accountability. LINKSPREED frames this combination as central to what distinguishes Web4 from the internet that came before it — one where intelligence is distributed everywhere, but nothing gets to act without first proving who it is.