01
Coverage ends
A canyon, a hull, a tunnel, a hundred hectares of field, a grid that has gone down. The network has to arrive with you or not at all.
Mesh networking for robots and field operations
Networks that
survive the field.
Software you install on the machines you already own. It turns their radios, cables and tunnels into one network that routes around a dead link by itself — and your existing applications run over it unchanged.
Running today on Nordic and ST microcontrollers, on Linux gateways, and in simulation. Taking a few design partners.
state
converged
The problem
Infrastructure is the single point of failure.
Most software assumes the network is simply there. Past the edge of that assumption, three things break at once.
02
Fixed topology breaks
A star network dies with its hub. Nodes move and hardware fails, so the network has to find another path on its own — in seconds, not on the next maintenance visit.
03
The medium is public
A radio anyone can hear is a radio anyone can transmit on. Without signed membership, a mesh takes routing claims from any device in range.
Where it fits
Built for the places the network does not reach.
Robotics & autonomy
Fleets where the topology changes every second, and what strands a machine is the link, not the battery. Routes re‑form as the fleet moves, and your autonomy stack can see a link going before it has gone. Multicast is carried across the mesh — the traffic DDS and ROS 2 discovery run on, and where a naive mesh falls over.
- Mobility‑tolerant routing
- Multicast for DDS and ROS 2 discovery
- Link quality exposed to the application
- No ground‑station dependency
Industrial & critical infrastructure
Assets kilometres apart with no cellular coverage to lease. A low‑power radio carries the telemetry; the gateway hands it to the systems you already run.
- Runs on the radios you picked
- No recurring connectivity contract
- Bridges to the existing network
Defence & public safety
Comms that do not need a tower, a satellite window or a working grid. People and vehicles carry the network with them, and only signed hardware is part of it.
- Infrastructure‑independent
- Cryptographically gated membership
- Degrades gradually, not all at once
How it works
One network, from the fence post to the server in the van.
Every node learns the next hop toward every other, and none of them holds a map. No hub, so no single failure takes the network with it.
One network, not two stacks
The identical code, from a microcontroller with no operating system to a server. Memory is fixed at compile time, so a node cannot quietly run out of it in week three.
Any link, all at once
The router does not care which link a frame came in on, so a new medium is a driver rather than a change to the routing.
À la carte encryption
Turn on membership, then take as much secrecy as you need.
- Signed membership Only hardware you have signed can advertise a route or address a member. Broadcast traffic is not covered, which is why the tiers below exist.
- Bring your own IPsec, WireGuard or your application’s own, unchanged. We never hold a key to your data.
- Managed for you Nodes with an internet path enrol themselves onto Tailscale, over a Headscale the mesh runs. Radio hops stay authenticated, not encrypted.
Your software can see the network
A mesh sits underneath the software running on it, which otherwise cannot see the network it depends on. One API publishes link quality, throughput and route stability — enough for an autonomy stack to adapt, and it is what the dashboard reads.
Deployment
Four ways to put a node in the field.
A node is one binary and one config file. What changes is the machine underneath it, not the routing.
01
Bare metal
Flash the firmware and it relays. No operating system, no host — a board with a radio becomes a repeater. Range and battery life are the radio’s, not ours.
02
Linux gateway
Ordinary IP applications route across the radios unmodified. This is the machine that joins the mesh to the network you already have.
03
Container
Images for every binary, plus a simulation of a full mesh on one workstation — rehearse your topology and a failover before buying any hardware.
04
Cloud authority
The root of trust runs as a node with no radio: it signs membership, handles enrolment and revocation, and runs the Headscale the others tunnel over. Its root key is generated offline and stays yours.
Operating it
It fails in a field at 2am. Now what?
Every node answers the same API, whether it is a server or a sealed board in a field.
It tells you what is wrong
A flood, a revoked peer still talking, a table filling up. The node tells you, and repeats coalesce rather than pile up.
Logs without a probe
Read them over the API and turn the detail up at runtime. No cable, no reflash, no reboot.
A window, not a command line
A browser dashboard, a terminal one and a CLI — because whoever depends on the mesh rarely built it.
Design partners
We are taking on a limited number of pilots.
It runs today on real boards and real radios. What it has not done yet is your deployment — and the teams we work with now shape what gets built next.
-
1
Scoping call
We will tell you plainly whether a mesh is the right answer. Sometimes it is not.
-
2
Bench bring‑up
Your hardware, your radios, a working mesh — simulated first, then on the real boards.
-
3
Field pilot
A real deployment, with the instrumentation to show what the network actually did.
Engineering work on both sides, not a procurement exercise. Bring an environment and an engineer who knows it. Tens of nodes per segment today; terms we settle on the call.