Links move, degrade and disappear
Coverage, congestion, interference, mobility and physical damage turn a stable path into an operational variable.
Resilient multi-link communications software
When a link fails, the connection doesn’t.
Nexus Atlas is multi-link bonding software: satellite, cellular, radio, Wi-Fi, mesh and wired paths become one encrypted connection that applications never notice. Every path is measured continuously; traffic shifts away from trouble gradually, not on a binary up/down trigger. Nodes also relay for each other: multi-hop mesh routing keeps more than one path open between any two nodes, so a lost node or a broken hop is routed around rather than taking the connection with it. No appliance, no controller, no cloud. It works fully offline.
Lose links. Lose nodes. Lose paths. Lose peers. Keep talking.Bond satellite. Bond cellular. Bond radio. Bond mesh. One tunnel.Any radio. Any link. Any terrain. Any storm. One encrypted fabric.Single binary. No cloud. No controller. No phone-home.Self-healing mesh. Sub-second reweighting. Predictive routing.Gradual failover — never binary up/down.Routing around failures that haven’t happened yet.GNSS-denied. Spectrum-contested. Link-degraded. Still routing.Contested spectrum. Cut cables. Weather fades. Still connected.
Compatible existing Linux computeVendor-neutralCustomer-controlled operationOffline-capable data plane
Any links · one connection
Two nodes · six kinds of bearer each · one encrypted connection.
- RadioVendor Acarrying
- CellularCarrier Bcarrying
- SatelliteProvider Ccarrying
- MeshVendor Dcarrying
- Wi-Fi / HaLowLocal networkcarrying
- Fiber / WiredExisting LANcarrying
Every path is measured continuously, so traffic moves off a link that is degrading — before it fails.
Applications stay connected
Software-onlyRuns from servers to compatible single-board computers weighing around 15 g; no mandatory Atlas appliance.
No fixed link limitConfigure the links the deployment and node resources require.
Vendor-neutralCombine compatible paths from different equipment and service providers.
Field-demonstratedEight-node mixed fleet under controlled link degradation at Hemus 2026.
The operating reality
The operating reality
Most networks assume links stay up. Operational environments do not.Read the full argument — including where the alternatives win
Backup failover reacts after disruption
A backup that waits for an outage can still drop the voice call, control session or video feed it was meant to protect.
Vendor silos fragment the outcome
Each radio, carrier and network is managed separately instead of contributing to one communications objective.
Hardware resilience adds integration burden
Another box introduces mass, power, mounting, cabling and qualification work—especially on vehicles and aircraft.
What Nexus Atlas does
One logical network. Three layers of resilience.
Atlas does more than move a connection between links. It coordinates the available paths, protects different traffic according to its purpose and can route across a wider participating topology.
Across links
Measure each path and use aggregation, best-path selection, duplication or gradual reweighting as conditions change.
Across traffic
Protect control, voice and telemetry from bandwidth-intensive video or bulk transfer through class-specific policy.
Across topology
Forward through authenticated mesh peers and reform routes when direct links or intermediate nodes change.
Across operations
Keep the data plane working without a mandatory cloud controller while retaining local visibility and control.
Beyond backup failover
Respond to change before a path becomes unusable.
Traditional backup designs often wait for a binary failure. Atlas continuously measures the path set and can change how links contribute while applications continue to use the same logical connection.
| Conventional backup | Nexus Atlas |
|---|---|
| One primary path and one waiting backup | Multiple paths can contribute concurrently |
| Acts after a threshold declares failure | Reweights as measured path quality changes |
| Moves the connection as one undifferentiated flow | Policy can differ for control, voice, video and bulk data |
| Usually scoped to an endpoint or appliance pair | Can participate in a wider multi-hop logical topology |
Response cadence: typically configured between 0.1 and 0.75 seconds.The interval controls how often traffic weighting can be reconsidered as conditions change. It is deployment-specific—not a universal dead-link or application-recovery guarantee.
See the full honest comparison — category by categoryWhat the platform does
Fifteen capabilities, one communications objective.
Fifteen things the platform does for the mission — each anchored by the number that makes it concrete. The complete catalogue explains all of them in business language, with engineering deep links for your advisors.
01Multi-link bonding
Every link active at once — aggregate for throughput or duplicate for certainty, per traffic class.
6 dissimilar link types bonded on one field node02Continuous link sensing
Quality measured continuously; degradation is seen and acted on before failure.
~4× per second, per link, on its own clock03Per-packet scheduling
Eight strategies, including interference-adaptive redundancy across simultaneous paths.
1→2→3 paths as interference worsens, then back04Signal-trend prediction
The trend of a fading radio is projected forward; traffic leaves a degrading link while it still carries.
hand-over before the link fails — not after05Self-healing mesh
Multi-hop relay through your own nodes; routes recompute when nodes join, move or are lost.
8 hops (default, configurable) · reconverges on detection — ≈1.25 s at defaults06Six traffic classes
Command, voice, position, telemetry, video, bulk — each with its own delivery contract on one tunnel.
a 4 Mbps stream can’t starve a 2 kbps channel07Encryption always on
The Noise protocol family — X25519, ChaCha20-Poly1305 — with automatic key rotation. The primitives sit behind the protocol: swapping the whole suite for FIPS-validated or national algorithms is a roadmap build mode, not a redesign.
2-min rotation · zero measured loss08Reachability service
CGNAT traversal with hosted or private relays; direct paths are discovered and upgraded automatically.
reachable in <1 s · EU + North America live09One binary, your hardware
Linux and Android, from datacenter servers to single-board computers and handsets, with a web dashboard on every Linux node.
~15 g compute floor · 100% software10No controller, no cloud
Every node is autonomous; configuration spreads peer-to-peer over the encrypted tunnels themselves.
0 cloud dependencies · runs air-gapped11Delivery through blackouts
Bulk data addressed to an unreachable peer is held — surviving reboots — and drains the moment any path returns.
no end-to-end route required — data still arrives12Loss repaired, not resent
Reed–Solomon parity striped across the bonded links reconstructs lost packets at the receiver — even a whole link going dark.
1 trip · 0 retransmissions · erasure coding (FEC)13Location & velocity aware
Nodes share GNSS position and velocity in the routing flood; the scheduler projects each track and raises the cost of links about to stretch out of range.
routes bend before the geometry breaks14Terrain awareness
Elevation-model line-of-sight and Fresnel-zone checks flag a link sliding into terrain shadow before the radio confirms the fade.
in development · watch it live in the mission simulator15Hardware that talks back
Custom adapters read what the radio itself knows and feed it into the same decision loop — per link, continuously.
RSSI · noise · SNR · buffer depth · error countersThe idea underneath
Route around failures that haven’t happened yet.
The point was never merely joining several links. It’s giving the network enough knowledge — its radios, its motion, its topology, eventually its terrain and its map of contested spectrum — that a failure can be seen forming, and the traffic moved first.
01What the network knows
Radios report their signal trends through the same adapters that drive them. Moving nodes share GNSS position and velocity. The mesh holds the whole topology. Elevation data adds the terrain in between — that part is in development.
02What it projects
Each link’s risk, a few seconds ahead. Failover — even fast failover — answers yesterday’s question: which paths work right now? The operational question is which paths will still be working moments from now.
03What it decides
Where every packet goes next. Every projection feeds the same per-packet scheduler and the same mesh route costs — so traffic leaves a dying link seconds before it fails instead of milliseconds after.
radio & link telemetry+topology · position · velocity · terrain (in development)→projected link risk→per-packet scheduling & routing
Labels first. Signal-trend and position/velocity prediction are shipped and opt-in; terrain awareness is in development; interference-zone awareness is a research direction. And prediction only ever advises — live measurement remains the sole authority on whether a link is alive. How prediction works — nexusatlas.io · What we’re building next
Where continuity matters
Different missions. The same structural weakness.
Each platform carries a different combination of links and traffic, but all depend on communication paths that can degrade, saturate or disappear.
Uncrewed and autonomous systems
Keep command, telemetry and video moving as a UAV, UGV or robot changes range, coverage and direct reachability.
Explore the use caseDefence and mobile command
Coordinate radio, cellular, satellite and mesh paths for mobile headquarters, vehicles and field teams.
Explore the use casePublic safety
Maintain incident command across vehicle relays and surviving infrastructure when public networks are congested or damaged.
Explore the use caseCritical and remote infrastructure
Combine dissimilar carriers and bearers so one provider outage or weather-affected path does not become an operational incident.
Explore the use caseKeep the installed platform
Use compatible onboard Linux compute, radios and network investment already present.
Add Nexus Atlas software
Expose compatible paths to one measured and policy-governed logical network.
Add genuinely independent links
another vendoranother carrieranother bearer
Specialist links are welcome.Atlas directly consumes IP-capable paths. Small adapters can expose serial or specialist radios and their operational telemetry without tying the platform to one manufacturer.
Vendor-neutral expansion
Keep the links you trust. Add the links you need.
Atlas is not tied to one radio, carrier or equipment supplier. Preserve existing equipment, introduce independent alternatives incrementally and strengthen the logical connection with every genuinely diverse path.
Retain existing links→Add Atlas→Introduce independent paths→Reduce shared failures
Diversity must be real. Several routes over the same modem, carrier or backhaul can still share one physical failure domain.
Review the integration modelEvidence before adjectives
Demonstrated behavior, with the boundary stated clearly.
At Hemus 2026, an eight-node mixed fleet ran control and video traffic through repeated physical path cuts and injected loss, delay and jitter.
It has also flown. A ground station, a mast carrying omnidirectional and directional antennas, and two aircraft were all members of one mesh — each relaying for the others, with links made and lost by distance and geometry rather than by a cable being pulled. Multi-link bonding, the scheduler in broadcast, traffic classes and carrier-grade NAT traversal all ran in the air.
Both were trials run by our own team — controlled link degradation at Hemus, real geometry in the air. Neither involved live RF jamming, and neither is customer deployment evidence or a production-availability claim.
Review the evidenceOperational patterns
Hard problems, composable answers.
Seventeen field patterns operators build on Nexus Atlas — some pure product capability, some combining Atlas with surrounding systems, some concepts stated as such. Three examples below, and forty operations they serve in the mission library:
Multi-constellation satellite bonding Pattern
Bond terminals from different constellations and orbits so one operator’s outage — commercial, political or atmospheric — is absorbed by the others.
Store-carry-forward through blackouts Capability
Bulk data waits out an outage — surviving reboots — and drains automatically the moment any path returns, even a passing relay vehicle.
Pre-positioned mesh relays Pattern
Battery-powered relay nodes staged along a dead-zone route wake, hop traffic forward and sleep — range without carrying SATCOM.
Memory-safe Rust coreThe implementation direction recommended by 2024 NSA/ONCD guidance for security-critical software.Cryptography, namedNoise IK · X25519 · ChaCha20-Poly1305, with two-minute zero-loss key rotation.Verifiable releasesCycloneDX SBOM, embedded dependency manifest, signatures and SLSA Level 3 provenance.EU-built and sovereignBuilt in the EU, ITAR-free, no phone-home, runs air-gapped.
One platform family
The right depth for every audience.
The business case, engineering detail, Traversal suite, operations console and application downloads each have a clear public home.
nexusatlas.euBusiness — you are here
Explore the full ecosystem — every product, one pageOutcomes, markets, evidence, integration and partnerships.
nexusatlas.ioEngineeringArchitecture, interactive technical explanations and implementation detail.
nexusatlas.devDevelopersDocumentation, configuration and API reference, release verification, evaluation access.
nexusatlas.netTraversalDirect discovery, CGNAT traversal and public or customer-controlled relays.
console.nexusatlas.netOperationsEnrol nodes and run the estate: live map, per-link telemetry, alerts, audit.
nexusatlas.appDownloadsThe Linux and Android application tours, galleries and evaluation builds.
Programme updates
Stay informed when there is something worth sharing.
Occasional briefings, demonstrations and milestones. No marketing cadence.