Skip to content

AI-native business software

Your business already has a system.

It just happens to live across spreadsheets, inboxes and people. We turn it into software.

Cyclotron designs and builds AI-native business systems for growing companies — from internal operations and approvals to customer portals, management intelligence and agentic workflows.

Scattered workStructured operating system
Discovery
Paid, standalone
Build
Fixed scope
Ownership
Your code & data
AI actions
Permissioned

The transformation

Same process.
Finally a system.

We do not ask you to change how the business works in order to fit software. We record how it already works, then remove the parts that only exist because there was no system.

01 — Today

Master tracker.xlsxApprovals groupSite photosVendor rates v7Inbox: RE: RE: POHandwritten log

The process exists. It is just distributed across tools and people.

02 — Cyclotron system

  • Modelled entities & states
  • Roles, permissions, audit
  • Approvals & escalation
  • Integrations to existing tools
  • Bounded agents

03 — Operating state

Approval cycle
Status visibility
Manual handoffs

Observable, measurable and improvable — because it is now a system.

Diagram: scattered spreadsheets, messaging and email are consolidated into a Cyclotron system, producing an observable operation.

Outcomes, stated plainly

The changes companies notice within the first quarter of running on a system designed around them.

01
Fewer manual handoffs
Work moves because the system moves it, not because someone remembered to forward a message.
02
One source of truth
A single record per order, job, site or customer — with everything else pointing at it.
03
Faster approvals
Authority encoded once, decisions taken in seconds, every one of them auditable.
04
Searchable institutional knowledge
Contracts, SOPs and correspondence become answerable, with the source attached.
05
Measurable workflows
Cycle times, ageing and exceptions are observed continuously rather than reconstructed monthly.
06
Controlled AI actions
Agents act inside defined boundaries, with permissions, audit trails and human approval gates.

The gap

Software where SaaS stops fitting

Off-the-shelf software is usually about seventy per cent right. The last thirty per cent is where your margin, your differentiation and your daily frustration all live.

Most companies respond by bending the business to fit the product, then rebuilding the missing part in spreadsheets anyway. We take the opposite position: keep the systems that genuinely work — your accounting package, your ERP, your email — and build the operating layer that sits between them and reality.

What SaaS does well
Standardised, regulated, commoditised functions: ledgers, payroll, mailboxes, storage.
Where it stops
The coordination between those functions — approvals, exceptions, evidence, handoffs, judgement.
What usually fills the gap
A spreadsheet, a messaging group and one person who understands both.
What we build
That gap, as software you own — integrated with what you already run.

Where the work happens

Software built around the floor, the site and the field — not the org chart.

We spend the first phase where the process actually runs. The system that follows is shaped by what we observe there, not by a template.

Reserved — operational photography

Documentary wide shot of a working operation — production floor, warehouse aisle or active site — observed, unposed, natural light, people mid-task.

Reserved placement — pending approved photography

How it compounds

Start from proven patterns. Productise what repeats.

Every engagement begins with your specific workflow — but not from an empty repository.

  1. 01

    Reusable modules

    Identity, permissions, audit, approvals, document handling, notifications, reporting and agent supervision are already built, tested and hardened. We are not rewriting them on your budget.

  2. 02

    Configured, not templated

    Your entities, states, roles and rules are modelled specifically. Two clients in the same sector rarely get the same system, because they rarely run the same process.

  3. 03

    Productised where it repeats

    When the same need appears across several companies, it becomes a durable product capability — which means it keeps improving after your project ends.

Applied AI

Agents that are supervised, not trusted blindly.

AI is not the product. It is a capability inside the workflow — and it earns autonomy the same way a new employee does.

  1. 01Read

    Inboxes, documents, tickets, records and operational messages.

  2. 02Analyse

    Classify, compare against history, detect variance and gaps.

  3. 03Prepare

    Draft the reply, the comparison, the summary, the brief.

  4. 04Route

    Send it to the right person, queue or stage with context attached.

  5. 05Follow up

    Chase commitments, re-check conditions, escalate on ageing.

  6. 06Execute

    Take a narrow, enumerated set of permitted actions — nothing else.

The boundaries, stated up front

  • Permissions: an agent can never read a record its requesting user could not open.
  • Audit: every run records input, action, outcome and approver.
  • Determinism: calculations, eligibility and policy stay in code, not in a prompt.
  • Human approval: money, contracts, external commitments and deletions require a person.
  • Escalation: low confidence hands work to a human, already prepared.

How we work

Five phases, each with an output you own

No open-ended discovery, no perpetual retainer to reach a first release.

The full process
  1. 01

    Process Audit

    1–2 weeks

    We sit with the people who run the process and record how it actually works — including the parts that live in messages, memory and workarounds.

  2. 02

    System Blueprint

    1–2 weeks

    The design: data model, users and permissions, workflow states, integrations, automation opportunities and explicit AI boundaries.

  3. 03

    Working Prototype

    2–4 weeks

    One critical workflow, running with your data and your users. Not a slide deck, not a clickable mock — a system you can operate.

  4. 04

    Build & Integrate

    6–14 weeks

    Production build: the full workflow set, integrations with existing systems, migration of live data, access control, testing and staged rollout.

  5. 05

    Operate & Improve

    Ongoing

    Monitoring, support and a standing improvement cadence — plus the intelligence layer, which usually only becomes valuable once real data has accumulated.

Global delivery

Built for distributed businesses, wherever they operate

Multi-entity, multi-currency, multi-language and multi-timezone are design assumptions, not paid extras.

Our systems are designed for companies whose teams, sites, vendors and customers are not in one place. That shapes everything: access models scoped by entity and region, asynchronous approval flows, audit trails that hold across jurisdictions, and interfaces that work on a phone with poor connectivity as well as on a desk.

We build for operational reality rather than an idealised head-office scenario: intermittent connectivity, people working across shifts and time zones, approvals that must survive an audit months later, and reporting that has to reconcile across entities. Systems designed against those constraints hold up wherever a business operates.

Entities

Multi-company, multi-site

Access

Region and role scoped

Delivery

Async-first, distributed

Industries

Where we are starting

Six sectors where the gap between real operations and available software is widest. Starting points, not limitations.

Industry detail

Start here

Show us the process you run today.

Send us the workflow that is costing you the most time. We will tell you plainly whether it is worth building, buying or leaving alone.