Why Nexus Atlas

Lose links. Lose nodes. Lose paths. Keep talking.

Everything you operate hangs off a link, and links fail — jammed, cut, faded, saturated, or simply out of range. The tools that claim to fix this notice failure only after it happens, and most of them depend on a cloud that won’t be reachable when it matters. This page is the argument for doing it differently.

The pain-first argumentHonest comparison included

The core problem

Three failures, one root cause: the network is treated as a given instead of a variable.

One link is one point of failure — and everyone knows it

A platform may carry LTE, tactical radio, satellite and Wi-Fi, yet one path carries the active session while the backups sit idle. When that band is interfered with, the weather fades the dish, or the cable is cut, everything stops — the vehicle goes blind, the call drops, the telemetry vanishes.

What Atlas changes: it bonds every link into one encrypted connection, all of them live at once. The mission survives as long as any path survives — not “primary and backup”; all of them.

Failover is binary — and by the time it fires, you’ve already lost

Real links don’t die instantly; they degrade. Latency creeps, loss climbs, the storm rolls in over minutes — while the link still reports “up”. Threshold-based failover keeps pushing traffic into a dying path, then drops every session at once.

What Atlas changes: it measures continuous quality — delay, jitter, loss, capacity, several times a second — and shifts load away proportionally as a link degrades. Where an adapter owns the radio, Atlas also reads what the hardware itself knows — signal strength, noise floor, transmit-buffer depth, error counters — so a fade is visible from inside the radio, and traffic can move on the RSSI trend before the probes miss anything (opt-in). By the time a link actually dies, the traffic already left.

Your resilience plan depends on infrastructure you don’t control

SD-WAN needs its orchestrator. Overlay services need their cloud. Licence servers need the internet. A single satellite provider is one commercial or political decision away from an outage. All of it evaporates exactly where you need communications most: air-gapped networks, offshore in a storm, disaster zones, contested areas.

What Atlas changes: no controller, no cloud, no phone-home. Every node is autonomous; configuration spreads peer-to-peer over the encrypted tunnels; a partitioned network keeps operating on both sides of the split.

“Failover itself causes an outage.”

Even where failover exists, it moves whole connections: sessions reset, video re-buffers, the control channel hiccups at the worst moment. Atlas schedules per packet beneath the application — nothing to reconnect — and critical classes can be duplicated across links so a link loss passes without interruption.

“We pay for backup links that spend their lives idle.”

Expensive satellite, cellular or radio capacity produces no value until the primary fails. Atlas aggregates them for throughput, or holds them as measured, instantly-usable paths — either way the spend produces value every day.

“A link can be ‘up’ and still unusable.”

Rising latency, jitter and loss freeze applications long before conventional monitoring reacts. Atlas makes quality-based decisions, not state-based ones — a degrading link simply carries less until it recovers or dies.

“The topology changes faster than anyone can reconfigure it.”

Vehicles move, aircraft leave range, relays disappear, static routes go stale. Atlas’s mesh floods a shared topology view and recomputes routes the moment a change is detected — ≈1.25 s at default probe settings for a lost node.

“We find out the link is bad when the video freezes.”

Reactive systems learn about geometry when packets stop arriving. Atlas can be predictive: nodes share position and velocity, radio signal trends are projected forward, and traffic moves to a longer-reaching path before the short one fails. Digital elevation models sharpen this further — sampling the ground profile between two nodes for line-of-sight clearance and Fresnel-zone obstruction catches a ridge sliding into the path that relative geometry alone misses. That engine drives the mission simulator today; wiring it into live scheduling is in progress.

“Two stations can see the relay but not each other.”

Where direct reach ends, conventional VPNs end. Atlas routes multi-hop through your own participating nodes — eight hops deep by default, re-encrypted at every hop — so a surviving path through the network replaces the direct path you lost. The eight-hop depth is a loop-protection default, not an architectural ceiling.

“Video starves the command channel.”

On a degraded link, bulk video happily crowds out the small stream that keeps the asset alive. Atlas’s six service classes give command strict priority with reserved floors — with classes enabled, a 4 Mbps video stream cannot crowd out a 2 kbps command channel.

“Retransmitting stale data makes congestion worse.”

Late video is worthless, but conventional transport retransmits it anyway, choking the link further. Atlas applies deadline delivery: too-late frames are dropped, not resent. Video bends; it doesn’t break.

“A blackout means lost data and restarted transfers.”

Hours of silence usually mean an operator restarting uploads. Atlas store-carry-forwards: bulk data is held through the outage — surviving reboots — and drains automatically when any path returns, even via a passing relay node.

“Loss on wireless normally costs a round trip.”

Atlas stripes parity across the bonded links (systematic Reed–Solomon): lost packets — up to an entire link going dark — are reconstructed at the receiver in one trip, with no retransmission and no return channel.

“DIY multipath fails exactly when it matters.”

Everyone has tried a VPN plus keepalive scripts. The detection-and-decision layer is precisely the part that fails in a hurry — and it’s the part Atlas is. One binary, already built, on the same cryptographic family WireGuard and Signal use.

“Every radio vendor ships another closed box.”

Integrators accumulate appliances, scripts and per-vendor failover logic. Atlas is pure software at Layer 3: any IP-capable link joins the bond, specialist radios join through small adapters, and the hardware stays vendor-neutral.

“Fleets are blind.”

Nobody can answer “how good is the satellite link on site 14 right now?” Atlas exposes live per-link telemetry on every node’s dashboard, and the hosted console adds a live map, offline alerts and an audit trail across the estate.

“Security review stalls procurement.”

Memory-unsafe network stacks and opaque crypto invite long reviews. Atlas answers structurally: a memory-safe Rust core, named cryptography, signed releases with SBOM and build provenance, and a hardened deployment profile. Security & compliance

Honest comparison

Where Atlas wins — and where the alternatives do.

Categories, not names. Your advisors can verify every row on the engineering site — including the mechanics behind each claim.

Capability	Single-link VPN	Failover router	Cellular bonding appliance	SD-WAN	MANET radio system	Nexus Atlas
Several links active at once	—	standby only	cellular only	policy-dependent	own radios only	any mix, all live
Bonds dissimilar bearers (satellite + cellular + radio + mesh + wired)	—	—	rare	limited	—	yes
Acts on degradation, not just failure	—	—	partial	partial	varies	continuous quality, gradual reweighting
Application survives a path change	—	—	mostly	mostly	yes	per-packet; nothing to reconnect
Multi-hop relay through your own nodes	—	—	—	—	own PHY only	any bearer, 8 hops by default (configurable)
Works with no controller and no cloud	yes	yes	management cloud	—	yes	yes
Traffic classes with distinct delivery contracts	—	—	basic QoS	policy QoS	limited	6 classes: priority, deadlines, store-carry-forward, FEC
Named modern cryptography	varies	varies	opaque	opaque	proprietary	default suite: Noise IK · X25519 · ChaCha20-Poly1305, 2-min rekey — primitives swappable by design
Runs on hardware you already own	yes	—	—	—	—	Linux, ~15 g board and up

Where they’re stronger. SD-WAN suites bring years of WAN analytics and application steering at enterprise scale. Bonding appliances are turnkey hardware with global support channels. MANET radios bring purpose-built waveforms and ruggedisation Atlas does not replace — Atlas rides on top of them and bonds them with everything else. If one of those is the whole of your problem, buy it. If your problem is dissimilar links, contested conditions, or infrastructure you must own outright — that’s what Atlas is for.

Full architecture comparison, including detection mechanics — nexusatlas.io