Local coordination: protecting connected devices with one internet gateway

I have been talking about the fragility of the internet since I started Offline Protocol in 2024. This week it showed: breaches across PHP, Railway, eMudhra and LiteLLM landed within days of each other, and each one reached systems that were only exposed because they were connected to everything else.

We have spent two years building the alternative. Telecom was the first network and the internet the second. The third is the network devices form among themselves: local, encrypted, self-identifying, and connected to the wider internet only where and when it is needed.

AI-driven attacks on connected devices

Most connected systems give every sensor, camera, controller, phone and laptop its own path to the internet. Security has relied on the fact that finding a weak one took time and skill.

That is no longer true. Models now find bugs faster than people can patch them, so the cost of an attack keeps falling. Code written with AI assistants ships faster and carries more vulnerabilities. Machine traffic has already passed human traffic on the web, as Cloudflare Radar shows month after month. Faster patching does not keep up with that. Fewer paths in does.

Self-replicating AI worms

A recent essay, Defense Against the Super-Worms, describes where this leads: worms that run language models on the compute they capture, rewrite themselves as defenders respond, and spread to everything they can reach. Its conclusion: “I think the time of the world wide web is coming to an end.” The author argues that critical infrastructure will have to move onto smaller intranets, joined to the wider world by narrow links that carry only small, expected packets.

We already build those intranets. For the past year we have helped enterprise clients run local coordination, and it does what that essay asks for today.

Local coordination with a single gateway

Take a site with a hundred devices: sensors, drones, machines, computers, handhelds and smartphones. Today each one connects upstream on its own. With our stack, one gateway machine holds the uplink, and the other devices form a local network with our SDK, which gives them:

• Device-issued identity: keys generated on the device itself, with no central account to steal
• Nearby discovery: devices find each other without a directory server
• Encrypted communication: end to end, between the devices themselves
• Conflict-free replicated data (CRDTs): every device converges on the same state without a central database
• Telemetry: collected on site and sent upstream when it matters
• Optional on-device intelligence: models run where the data is
• Upstream reconciliation: the gateway syncs with the organization’s own system of record

Protection from cyber attacks and unauthorized access

Going from a hundred paths in to one changes what an attacker can reach. There is one gateway to harden, watch and audit. Devices prove who they are with keys they generated themselves and only talk to peers that can do the same. Data moves encrypted between devices that are physically on site, and it leaves only as expected, verified updates through the gateway.

That gateway is the airlock the Super-Worms essay describes. A worm spreads by reaching its next target. A fleet that coordinates locally and reconciles through one narrow gateway leaves it no open route between sites and no exposed device to land on, so a self-replicating attack cannot take hold across it.

Lower backhaul costs and outage resilience

Security is the newest reason to coordinate locally. Cost is the oldest. When a hundred devices stop sending everything over fiber or satellite and one gateway sends only what matters, the backhaul bill drops, and devices next to each other talk directly instead of through a distant server.

The same design keeps a site running when the uplink fails. The devices keep coordinating, and the gateway catches up when the link returns.

Offline Protocol in production

In two years we have built device-issued identity, resilient online and offline networking, replicated data, agent and machine coordination, and most recently private on-device intelligence.

This year we demoed all of it on one stack: local IoT communication, robotic coordination, an agent-to-agent services marketplace, delegated access to local inference, private model training runs, smart sensor telemetry and autonomous fleet communication. Our infrastructure has been used on almost 500,000 devices, and we ship new features and bindings for connected hardware almost every week.

Getting started with local coordination

The risk from an always-connected internet does not stop at software. It reaches factories, fleets, grids and facilities, where downtime and intrusion cost real money. Those systems do not need to share an open road with the rest of the internet.

If you build connected products or run operations that have to stay safe and keep moving, start with our documentation or talk to us. We would love to help more products and organizations become resilient.