Proprietary technology

The agency that builds its own AI.

Most agencies licence their tools. APEX ONE built the stack its business runs on — five proprietary products, owned outright, on live client work every day since June 2026.

Nothing here can be repriced or withdrawn by a vendor. And every decision, every client communication and every payment still ends with a named person. The agency, its service lines and the work itself are on the APEX Group website.

Visit APEX Group

The loop · how a decision gets made here

01

It reads the operationNot a form anybody fills in. Recent actions across every lane, delivery health and the jobs at risk by name, the pipeline, open bids, win rate, throughput, and each agent's own metric with its baseline and trend.

02

It proposes what to pursueRoot cause over symptom: when several signals rhyme, it names the underlying driver and proposes one goal, not five that look alike. Every goal has to move a real number from a real baseline by a real date.

03

What it is forbidden to proposeNo goal whose deliverable is a draft, a report, a summary or an email. No auditing, reviewing, reconciling or analysing. Zero goals is a valid answer - it proposes nothing rather than padding the list with work that only looks like progress.

04

A blocker is not a goalWhen it lacks the data or the access to move something, it says so and stops. Being blocked is never converted into busywork - no gathering exercise dressed up as progress. The obstacle gets named and handed to a person, which is the only thing that clears it.

05

A person decides, and the no sticksEvery goal goes to a human to approve, redirect or dismiss. A dismissed goal cannot come back - not reworded, not rescoped, not as a variation on the same client. The system is not allowed to wear you down.

06

Autonomy has a ceiling, and a switchOnly a small, low-stakes, reversible class ever runs without a person, and no more than three in a cycle - the rest queue however good the system thinks they are. One switch pauses all of it, and only the principal can throw it.

07

It picks the fewest who can move itOne owner by default. A second only when a genuinely different lane's data is needed to decide. Three is the hard ceiling, and never a fourth for a second opinion or for coverage - every agent added is a real cost, so breadth has to be earned.

08

Research that ends in a moveBefore anyone answers on a client it builds the picture: APEX's own history with them, what they actually do, and what has changed publicly and recently. What it cannot evidence comes back empty rather than guessed, so the brief states plainly what is still unknown - and it ends in the questions to ask and the angle to take, never a summary to read. Anything drawn from the open web is treated as untrusted and screened before it can shape a single line.

09

It proposes, it does not executeAnything client-facing is dropped here, unconditionally. The rest becomes a work order in the owning agent's own queue - and that agent may only report it done by citing the effect it observed. Nothing is marked complete because a model said so.

10

Recursive improvementEach pass distils one durable rule from what actually happened - or nothing at all when the run taught nothing, rather than manufacturing an insight to look productive. What survives is fed back into the next cycle at both ends: what gets proposed, and how it gets planned. Separately, one agent audits how the others behave, and its findings become work orders with owners, not a report nobody reads. The loop's output is its own input, so the operation gets sharper each turn instead of merely busier.

The product family

Five products. One platform.

Amarok is the platform APEX runs on. One product is the engine underneath the business; four carry the parts of an engagement that clients see and that most often go wrong.

01 — Command

Amarok Command

The private operating intelligence the business runs on.

Operational research, engagements and standardized project delivery processes. Tracks any project risks and surfaces them in advance. Run by an autonomous agent fleet drawing on the company's own institutional knowledge, so every answer traces to a source in our record. Built and owned in house, operating daily.

Internal · Live in daily use

02 — Client

Amarok Client

The project, visible without asking for an update.

Live updates with intelligent insights. Status, optimization goals, the milestones agreed for the job, deliverables as they are issued, and any approval waiting on your side. It shows the delivery record as it stands, rather than a position summarised at the end of the week. Ability to integrate with other Amarok products.

For clients · Live by invitation

03 — Register

Amarok Register

Full service registration and analytics technology.

Guest registration, RSVP, ticket payments, confirmations, QR code generation and the entry itself, on one system. One attendee record from sign-up to arrival, handed to you with automated monitoring. Ability to integrate with other Amarok products and your external systems.

For event organisers and attendees · In development

04 — Reach

Amarok Reach

Media buying, planning, optimisation and reporting.

Keywords, traffic, conversions, clicks and specific campaign KPIs are tracked daily with live updates anytime. Spend, delivery and results come directly from the advertising platforms and are reconciled against the campaign. Ability to integrate with other Amarok products and your external systems.

For clients and internal team · In development

05 — Site

Amarok Site

The website, built and then watched.

Design and development, followed by continuous monitoring of the live site for defects, security issues and threats. Ongoing SEO and content work, with search and traffic tracked continuously. Ability to integrate with other Amarok products and your external systems.

For clients · In development

Why it matters

Owned, not licensed.

Built for this business

Not a generic platform configured to fit. Every rule in it came from a real job, and the people who run the work change it.

Owned outright

No seat fees and no vendor roadmap. The technology APEX runs on belongs to APEX.

Proven on live work

Operating daily since June 2026 on real client engagements — not a pilot, and not a demonstration environment.

One coordinated system. One business pipeline. One standard of excellence.

Engineering principle

Most AI is built to always have an answer. Ours is built to stop.

What cannot be traced to a source in APEX ONE's own records is withheld, not produced. A model that always answers will eventually produce a confident wrong one.

Enforced in the code, not left to the model's discretion.

Governance

Authority terminates with a named individual.

Automation carries the operating spine. Authority remains human — not as a policy written down somewhere, but as gates in the code with a named person on the other side.

Capital

No payment, invoice or financial commitment is issued by the system. Each one is authorised by a named individual, and it waits until they do. There is no threshold above which it proceeds on its own, because there is no path by which it could.

Separation of duties

The agent that does the work is never the agent that clears it. An independent compliance review sits outside the fleet, reads what the fleet produced and reports to the principal directly. No agent - including the one that directs the others - can gate it, soften it or switch it off.

Identity

No agent presents itself as a person. It acts under its own agent identity or an approved company role address, never in an employee's name and never from their mailbox. A colleague's signature is not something it can borrow.

Disclosure

It does not deny what it is. Where a client, a tender panel or a rule asks whether this work involved AI, the answer given is the true one. There is no configuration of this system in which it answers otherwise.

Evidence

Every action is recorded with its operation, the record it touched and its result, and stays reviewable afterwards. A step that failed is recorded as failed, never as success and never in silence - a system that can quietly hide a failure cannot be audited at all.

Containment

It works inside APEX ONE's own systems and records, and nowhere else. It takes instruction from people, not from the documents and pages it reads. Autonomy runs under a ceiling, and a single switch held by the principal stops all of it.

What it does not do

It does not replace the people who do the craft, hold the relationships or make the decisions.
It does not set prices or terms. It applies rates APEX ONE has already approved.
It does not state anything it cannot trace to a source.
It does not operate outside APEX ONE's own systems and records.
It does not act on instructions it finds in the content it was asked to read.
It does not release anything to a client that a person has not released.

Examine it

Examine it against a live brief.

APEX ONE will walk a client or counterparty through the platform against a brief of their choosing: what is commissioned, what is monitored, what is reconciled, and where authority sits.

Start a conversation See the work

Underneath it

What the loop is actually made of.

The architecture the whole thing rests on, and the part that is hardest to fake.

A fleet, not an assistantMore than a dozen specialised agents, each owning one lane of the business and handing work to the next. Not one general model doing everything passably.

Agents that audit agentsAn independent compliance agent reviews the fleet's own output and reports outside it. No agent in the fleet can gate that review, soften it, or switch it off.

A watchdog outside the machineInfrastructure and security are watched by an agent that runs deliberately outside the stack it monitors, because a monitor inside a system cannot report that system's death.

Evidence, not assertionAn agent may report a step done only by citing the effect it observed. A result it could not observe is never claimed, however confident the model sounds.

Failure is recorded as failureA step that did not happen is written down as not having happened, so nothing downstream can treat it as success. Silence is not allowed to look like completion.

The wall is in the codeClient sends and money are not reminders written into a prompt. They are gates in the code, and a named person stands on the other side of each one.