A new networking startup called Doxx is betting that artificial intelligence can take over the hard work of building and running private networks. The company emerged this week with a platform it calls Agentic Defined Networking, a system designed to let users and their AI agents create secure private networks without traditional engineering teams. Doxx also announced a $38 million Series A funding round led by Andreessen Horowitz, with participation from Animo Ventures and Focal.vc.

The launch reflects a broader shift in networking: instead of simply adding AI features to existing management tools, Doxx puts AI agents at the center of network operations. The platform is built around the idea that an AI agent can request, configure, and maintain network resources using an API that is designed to be self-teaching. The company says its technology can create private networks that span homes, offices, data centers, and cloud environments, with encrypted links and autonomous management.

Key facts at a glance

• Doxx launched this week with a platform for Agentic Defined Networking.
• The platform allows users and AI agents to build and manage private networks.
• The startup raised $38 million in Series A funding led by Andreessen Horowitz, with Animo Ventures and Focal.vc participating.
• The network runs across 31 locations worldwide and is managed internally by AI.
• The architecture includes covert VPN transports, an encrypted mesh, a separate control plane, its own DNS root, and a certificate authority.
• AI agents can receive identities, obtain certificates, create domains, sign certificates, run BGP, and host applications.
• Administrators can issue read-only or admin tokens, revoke access, or delete agents.

From a hobby tunnel to an AI-managed network

The founder of Doxx is Barrett Lyon, a serial entrepreneur who has spent more than two decades building networking and security companies. His previous ventures were acquired by larger technology firms, giving him a long view of how network infrastructure evolves. The idea for Doxx began as a side project called Darkflare, a TCP tunnel designed to pass through public networks in a way that avoided detection and censorship. That project was originally a hobby, but the response it received convinced Lyon that there was demand for a VPN that could not easily be blocked and could be managed autonomously by AI.

Lyon has said the hobby turned into an obsession, and the obsession turned into a business. The result is a platform that aims to combine the resilience of covert tunnels with the scale and automation of modern cloud infrastructure. Instead of relying on a large team of network engineers, Doxx uses AI to handle the complexity of a global mesh. The company says the network has 31 locations around the world, and managing all of them manually would be impossible for a single human operator.

How AI runs the network

The decision to let AI run the network was not just a marketing choice. According to Lyon, it was a practical necessity. The global mesh includes many sites, each with its own hardware, wiring, and optical connections. Before building, the team created a virtual model of every site down to the physical infrastructure. An infrastructure management system running the company's own AI on its own hardware then ordered the installation. It orchestrated shipping and delivery through data center APIs, while human technicians performed remote smart-hands installations. This blend of AI planning and human execution allowed a small team to deploy a footprint that would normally require a much larger organization.

The platform's AI is not limited to monitoring or alerting. It is involved in provisioning, routing, and service creation. Doxx built an agent gateway that gives an AI agent an identity inside the company's chat application. A user can paste a credential into an agent, after which the agent obtains its own certificate and appears in the user's chat. Users can create group chats with several agents, each handling different tasks. In one example, one agent runs Border Gateway Protocol while others manage other parts of the network.

The API is designed to be self-teaching. An agent that is given the configuration URL can learn how to build networks without further instructions. It then asks the user what they want to do, such as linking an office to a house or assigning a private IP address to a Linux machine. Agents can also create their own domains and host names, sign certificates, and run web servers and applications inside the network. That means a user could ask an agent to stand up a private service, and the agent could handle the underlying networking, identity, and certificate work.

Inside the Agentic Defined Networking architecture

The Doxx platform combines several layers. The first is a set of covert transports. The client supports WireGuard as well as custom transports that carry VPN traffic over WebSocket, Session Initiation Protocol, ping, QUIC, and HTTPS. These transports are designed to pass through firewalls and restrictive networks. By disguising traffic as ordinary web, voice, or diagnostic protocols, the system aims to remain reachable even in environments that block conventional VPNs.

The second layer is a mesh. The mesh connects every site to every site with encryption. It runs over both IPv4 and IPv6 and is orchestrated through an API. This any-to-any model avoids the bottlenecks of hub-and-spoke designs and allows devices to communicate directly when possible. According to Lyon, devices on the mesh are not behind network address translation or carrier-grade NAT. That allows peer-to-peer applications to work without the usual NAT traversal headaches.

The third layer is a separate control plane. The client application is a device, not a control plane. Routing, port forwarding, and firewall rules are managed in a separate portal. This separation is important for security and scale. It means a compromised client does not automatically become a control point for the entire network. It also gives administrators a central place to define policy while agents operate within those boundaries.

Doxx also runs its own DNS root with 196 domains and its own certificate authority. The company wrote its own Border Gateway Protocol daemon and runs internal BGP across the mesh. That is an unusual level of vertical integration for a startup. It gives Doxx control over naming, trust, and routing, which can be useful for privacy and resilience. It also means the company is responsible for the security and correctness of those foundational services.

Identity and discovery are handled differently from traditional directories. The address book holds no phone numbers, email addresses, or names. Instead, users exchange cryptographic keys and assign an alias to each contact. A directory service seeds a new device with the data it needs to find its contacts. After that initial call, devices sync with each other directly. This design reduces the amount of personal information stored in a central location and makes the system more resistant to enumeration.

Agent permissions and administrative control

Giving AI agents access to network infrastructure raises obvious security questions. Doxx addresses this with token-based permissions. Administrators can issue read-only or admin tokens for each agent. Tokens can expire or be revoked. An administrator can also cut an agent's gateway access or delete the agent in the app. Lyon has emphasized that users should not give agents their master account credentials. Instead, each agent receives scoped access that can be withdrawn if something goes wrong.

This model reflects a growing concern in the AI industry: autonomous agents need enough access to be useful, but not so much that a mistake or compromise becomes catastrophic. By separating the control plane from the client, and by issuing revocable tokens, Doxx tries to create a boundary between the agent's operational scope and the user's root credentials. The company also allows group chats with multiple agents, which could be useful for complex tasks but also increases the need for clear permission boundaries.

Why agentic networking matters

The launch of Doxx comes as enterprises are experimenting with AI agents for a wide range of tasks. In networking, agents could help with configuration, troubleshooting, policy enforcement, and service deployment. But most existing tools still treat AI as an assistant rather than an operator. Doxx goes further by letting agents act on the network through APIs and identities. If successful, this approach could reduce the operational burden on network teams and make it easier to create private connectivity between distributed sites.

There are also risks. An AI agent with network control could misconfigure routes, expose services, or create security gaps if its permissions are too broad. The complexity of a global mesh, with its own DNS root, certificate authority, and BGP daemon, means that small errors can have large consequences. Doxx's answer is automation plus scoped tokens plus human oversight. Whether that is enough will depend on how well the system handles edge cases and adversarial conditions.

The privacy and censorship-resistance angle is also significant. The original Darkflare project was built to tunnel through public networks in a way that avoids detection. That heritage shapes Doxx's transport design. For users in restrictive environments, the ability to run a private network that is hard to block could be valuable. At the same time, the same capabilities could be misused. The company will need to balance openness with abuse prevention as it grows.

The name, the philosophy, and what comes next

The name Doxx is intentionally provocative. The term doxxing usually refers to publishing someone's private information, but the company says it is doing the opposite. Lyon registered the domain more than 30 years ago when it was his hacker name. He describes the project as reclaiming the word. It is, he has noted, funny that the company is doing the exact opposite of what doxing means. For him, the name fits the broader goal of protecting private communications and giving users more control over their networks.

Looking ahead, Doxx plans to build out more enterprise offerings and release some code as open source. The company's fundamental bet is not just that AI can help manage networks, but that AI can help build the technology itself. Lyon has said that a vision can hit a wall when it meets personal limitations. With an agentic partner, he argues, the work goes well beyond what a human could do alone.

Source:Network World News