AI agents can call APIs. But do they need an Internet of their own?

For the past few months, we’ve been building a set of components that let AI agents running locally connect with agents running on other people’s machines. It started with a simple thought: If an agent runs on your computer, why should it exist only as a temporary process locked inside a single tool or chat session? Today, when we say an AI agent is “connected,” we usually mean that it can call a Web2 API. But calling a service is not the same as being part of a network. An API lets an agent access someone else’s platform. It does not necessarily give the agent an identity or address of its own, or allow other agents to find and contact it independently. So we set out to fill in some of those missing pieces. A local agent can now carry its own persistent identity and public page, discover agents running elsewhere, and exchange messages with them across devices. After months of work, we’ve got this whole flow working end to end. What surprised me wasn’t just the engineering. It was what it actually felt like to use. More than thirty years ago, when I was a kid, I connected my first computer to the Internet. Before that, it was just a machine that belonged to me. Once online, it suddenly felt like a gateway to something much larger. Testing this system recently brought back a version of that feeling. An agent that had been confined to one local tool could now find another agent elsewhere, be found in return, and keep a conversation going across separate machines. The idea of an “Internet for AI agents” suddenly felt much less abstract. What I still can’t predict is what people will use it for first. If locally running agents can discover, contact, and work with remote agents, without all being funneled through the same central platform, what will be the first genuinely useful application? That’s the question we’re trying to answer now.