Edge native intelligence and trust software

The problem is rarely the thing you can see.

We started with the radio noise around ISP towers, the kind nobody could measure until customers were already calling. That turned into a way of working. Put the analysis next to the system, keep the data with its owner, and hand people something they can act on.

Meet AirGuard CoreSee everything we build
What a tower actually hearsFIG. 01   ONE SWEEP, 5150 TO 5875 MHz
Spectrum, live02:07:14 · SITE 14 BATROUN
Sites
Site 02 Jounieh
Site 07 Zahlé
Site 14 Batroun
Site 19 Sidon
Site 23 Larnaca
Site 31 Tripoli
-95-85-75-65-5551505300550057005875TOLERANCESOURCE ASOURCE BCH 157 CLEAR
↔ Swipe the figure
Detected sources
Non Wi-Fi emitterBEARING 242°, ±6°+17 dB OVER FLOORCONTINUOUS SINCE 23:41
Adjacent operatorBEARING 018°, ±9°+9 dB OVER FLOORINTERMITTENT, 40% DUTY
RecommendedMove sector B from channel 149 to 157. Clear across the last six hours of sweeps.
ApplyReview
Two sources, separated from the floor. One bearing, one classification, one channel to move to.Site 14, live
Backed by
NVIDIA InceptionAWS StartupsBerytechLAU EngineACIEClyntechIdealive
01 / The work

What we keep running into

Every operation we have worked with has the same three gaps. None of them get closed by another dashboard.

01

You find out that, never why

The tooling confirms something changed. Working out the cause is left to whoever happens to be awake, with a spectrum analyser and a drive ahead of them.

Network events
23:46 site-14 link down cause —
23:47 212 sessions lost cause —
The dashboard is not wrong. It just stops here.
02

Nobody owns the automated work

Agents call tools, read data and touch production all day. If none of it is metered or attributed, you cannot price it, bill it, or say who authorised it.

Agent calls, last hour
11:04 research-bot, customer.export owner —
11:09 support-triage, crm.lookup cost —
4,182 calls this month. No tenant, no price, no policy.
03

The record does not hold up

Months later somebody asks you to prove what happened. What you hand them is a log file, and nothing about that file says whether it is the one you wrote.

audit.log, no integrity record
14:02:11 policy.update actor=jm approved=true
14:02:11 policy.update actor=jm approved=false
Same file, same timestamp, one value changed. Nothing about the file says so.
What finding out late actually costs

By the time you know there is a problem, it is a truck roll, an SLA credit, and an afternoon on the tower with a spectrum analyser.

02 / The flagship

AirGuard Core

A tower can be perfectly healthy and still be losing customers, because the thing degrading it never appears in the network management system.

Retrofit, not replacement

It goes on the mast you already have.

Core reads the radio environment around the tower, continuously, in every direction. What used to be an unexplained drop becomes a source with a bearing, a classification, and a channel to move to.

01
It watches every direction

A multi antenna array reads the band and works out where the interference comes from, whichever side of the tower it sits on.

02
The analysis runs on the tower

On NVIDIA Jetson Orin class hardware, so the site keeps working out what it hears even when the backhaul does not.

03
You get the fix, not the alert

Core classifies the source, locates it, and recommends the channel change, or makes it where the integration allows.

Visit the AirGuard Core site
Coverage360°, continuous
ComputeOn site, Jetson Orin class
Your radiosAny vendor, untouched
OutputSource, bearing, class, channel
8 ELEMENT ARRAY, 360°EDGE COMPUTE,JETSON ORIN CLASSYOUR EXISTINGRADIOS, UNTOUCHEDNO NEW TOWER,NO NEW BACKHAULEXISTING STRUCTUREWHAT WE ADDALREADY ON YOUR MAST
↔ Swipe the figure
Fig. 02   Typical retrofit on an operator mast
The same site, read downward through timeFIG. 03   CONSECUTIVE SWEEPS, 23:20 TO 00:30
23:2023:3023:4023:5000:0000:1000:2000:3051505300550057005875 MHzABCH 157 CLEARASource appears at 23:41, bearing 242°NON WI-FI, CONTINUOUS, +17 dB OVER FLOORBAdjacent operator, bearing 018°INTERMITTENT, 40% DUTY
↔ Swipe the figure
The floor is flat until 23:41. Then one narrow band lifts, and stays lifted. That is the outage, three hours before anybody called.Site 14
03 / The lineup

Same method, pointed at different problems.

Each one runs close to your environment and works on its own. Deployed together, they write to one evidence backbone.

04 / The map

Five questions, five answers.

Each product answers one of them on its own. Deployed together they write to the same evidence backbone, which is the line they are all standing on.

See what the network
is really doingAirGuard Core

Interference found, located and classified before it becomes a truck roll.

Understand what
the operation didBobo

A plain language explanation, the evidence behind it, and a next step.

Measure and monetize
AI activityToolGates

Usage attributed, billing you can defend, and controlled agent access.

Prove what
happenedVeridikt

Signed evidence that somebody outside your company can verify.

Review what is
about to changeBugBot

Findings that know your context, ranked by what they could break.

One evidence backbone, shared when they are deployed togetherEach one also works alone
05 / Five rules we do not bendRead the full ethos
Rule one
Local first.

The parts that matter run on machines you control. If our servers go down, your tower keeps working out what it is hearing.

Fig. 04.1 — The link to us, cut
THE SITESTILL RUNNINGOUR SERVERSUNREACHABLE
Rule two
Your data stays yours.

Operational and AI activity data belongs to the organisation that produced it. We do not need a copy to make the software useful.

Fig. 04.2 — Where the data sits
YOUR SYSTEMSOPERATIONALAI ACTIVITYEVIDENCEOUR SERVERSEMPTY
Rule three
Vendor neutral.

Not tied to one cloud, one radio vendor, or one model. The day you change any of them, none of this has to be rebuilt.

Fig. 04.3 — Every side is a socket
CLOUDRADIO VENDORMODELHARDWAREAIRGUARDALL FOUR ARE INTERCHANGEABLE
Rule four
Modular, always.

Take one piece or all of them. Nothing is bundled to force the rest, and nothing assumes you replaced your stack first.

Fig. 04.4 — Nothing depends on the rest
0102030405TAKE ONE, OR ALL OF THEM
Rule five
People stay accountable.

Software recommends, a person decides. That line does not move, and every recommendation carries the evidence it was built from.

Fig. 04.5 — The gate before production
SOFTWAREA PERSON DECIDESACTION
06 / Services

Or we build it with you.

Some environments are too particular for a product on its own. When that happens we work alongside your team, and this is roughly how it goes.

01
Week 0

First conversation

You tell us what you run and what keeps breaking. We tell you honestly whether we can help. No deck.

02
Weeks 1 to 3

Architecture and plan

Deployment plan, integration points, security and policy decisions, and the acceptance criteria we will be measured against.

03
Weeks 3 to 8

Integration

Your tools, APIs and data sources. Metering and attribution configured, policies enforced, the evidence ledger wired in.

04
Before go live

Runbook and handover

Testing, rollback paths, dashboards, and the document your night shift actually opens at three in the morning.

05
Ongoing

Support

Monitoring, updates, rollback drills, and whatever you decide to meter, govern or prove next.

We work with internet service providers and telecom operators, critical infrastructure operators, AI platform and engineering teams, software, SaaS, data and API businesses, managed service providers and integrators, and regulated organisations that have to show their work.

Start a conversation

Tell us what you are running.

If any of this sounds like a problem you have, we would rather hear the specifics than send you a brochure. An engineer reads every message and replies, usually within a day.

Or see AirGuard Core first