Edge native intelligence and trust software

The problem is rarely the thing you can see.

Meet AirGuard Core See everything we build

AirGuard operations dashboard showing network health, wireless client metrics and device status

Backed by

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.

The tooling shows that something changed, but finding the cause still means a drive to the site with a spectrum analyser.

Network events

23:46 site-14 link down cause unknown

23:47 212 sessions lost cause unknown

The dashboard shows the failure, not the cause. Finding out late means a truck roll, an SLA credit and an afternoon with a spectrum analyser.

Agents touch production without the metering and attribution needed to price, bill or authorise their work.

Agent calls, last hour

11:04 research-bot, customer.export owner unassigned

11:09 support-triage, crm.lookup cost untracked

4,182 calls this month. No tenant, no price, no policy.

A log file cannot prove what happened when nothing shows whether it is original, complete or unchanged.

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.

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.

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.

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.

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

↔ 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

↔ 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.

AI operations analyst

Bobo

Ask it like a colleague.

Ask what happened in the words you would actually use. Bobo answers from your own operational data, shows the events it drew the conclusion from, and proposes the next step. Nothing reaches a live system until a person approves it.

In developmentTalk to us about a pilot

BoboSite 14 · 02:07

YouWhy did site 14 drop three times last night?

Bobo

All three drops line up with a rise in the noise floor on channel 149. The pattern starts at 23:41 and repeats on the hour, which usually means a duty cycled emitter rather than a failing radio.

23:41 noise floor -78 to -61 dBm
23:44 link margin under 6 dB, 2 sectors
23:46 session drop, 212 subscribers
00:41 same signature, repeats

Suggested: move sector B to channel 157, clear on the last six hours of sweeps. Needs your approval before anything is applied.

Metering and governance

ToolGates

Every call, priced and accounted for.

Everything your agents call passes through one place. Usage is metered and attributed to the customer, team, or workload it belongs to, so it can be priced and billed. The same gateway enforces policy and screens for prompt injection and data loss on the way through.

In developmentTalk to us about a pilot

Usage, current cycleTenant: Northwind

Agent	Tool	Calls	Cost
support-triage	crm.lookup	4,182	84.20
support-triage	kb.search	2,940	29.40
billing-agent	ledger.write	318	47.70
research-bot	customer.export	1	blocked

Policy: export contains PII, denied161.30 USD

Verifiable evidence

Veridikt

For the day somebody asks you to prove it.

Ordinary logs can be edited and nothing about the file says so. Veridikt hash chains and signs each record as it is written, so an auditor, a regulator, or a customer can check for themselves that the sequence has not been rewritten since.

In developmentTalk to us about a pilot

Evidence chainSegment 0184

#000182prev 9f2c41ab this 41ab77e0signed

#000183prev 41ab77e0 this 77e0b3d5signed

#000184prev 77e0b3d5 this b3d52c19signed

184 of 184 records verifiedchain intact

Change assurance

BugBot

Review that has read your architecture.

Instead of generic rules, BugBot argues from your own decisions. It tells you what this specific change touches, how far the damage would travel, and the document it is citing when it says so.

In developmentTalk to us about a pilot

services/billing/webhook.tsChange 4471

const client = new Gateway({

- retries: 3,

+ retries: 0,

timeoutMs: 8000,

})

Retries removed on the only path that settles payments

This gateway is the sole writer to the settlement ledger. With retries at zero a single upstream timeout drops the event silently, and the reconciliation job will not see a gap.

Blast radius: 3 services, 1 external processor · Cites ADR-014, incident 2026-03-11

Or use the arrow keys

01 / 04

04 / Services

Custom systems, designed and deployed with your team.

Week 0

First conversation

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

Weeks 1 to 3

Architecture and plan

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

Weeks 3 to 8

Integration

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

Before go live

Runbook and handover

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

Ongoing

Support

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

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