Pre-launch · flagship programme

Software that runs the company, not software the company runs.

hurukan is an operating system for a business: one record, one login, and agents that carry work out instead of waiting to be told about it. Built from nothing as an AI-native system of action, on a sovereign open-source platform, on infrastructure you own.

Join the waitlist Read the blog

One record, one login
Agents act, people approve
Sovereign and open source

The problem

The company runs on eleven systems and a person between each pair

Nobody chose this. It is what thirty years of buying the best tool for each job adds up to.

The seams cost more than the systems

Each product is competent. The expensive part is the space between them — the exports, the reconciliations, the person who knows which number is right. That work has no owner and never appears as a line item.

Every system records, none of them acts

They hold an accurate account of what already happened, kept current by people typing. Nothing prepares the next step, so the organisation's speed is set by how fast its staff can operate its software.

Nobody has the whole picture

Sales sees the pipeline, operations sees the schedule, finance sees last month. The complete view exists only in a meeting, once a week, assembled by hand and out of date by the time anybody acts on it.

The architecture

One record, agents that act, people who answer

Three commitments, and the third is what makes the first two safe to deploy.

01
One record
Every customer, order, item, machine and person exists once, in one system, with one identity. Not synchronised between products — the same object, read by whatever needs it. Most integration work exists to simulate this, badly.
02
Agents that act
Work is carried out rather than reported: the quote drafted, the replenishment ordered, the schedule rebuilt when a machine goes down, the invoice raised and chased. Inside explicit bounds, with the reasoning shown next to the result.
03
People who answer
Anything irreversible — money leaving, a customer contacted, a commitment made — waits for a person. Everything is logged in a form a human can read afterwards. Autonomy without a record is not autonomy, it is an unowned liability.

The system

One system where a company usually runs four

The ambition is stated plainly because it should be judged: the categories below are what hurukan is built to replace, not sit alongside.

One record of the business

Customers, suppliers, items, orders, stock, people, machines and money as one connected model. The place where an ERP core, a CRM and a WMS stop being three answers to the same question.

Work that executes

From enquiry to cash and from order to delivery, carried out by the system with a person confirming the parts that matter. The unit of work is the job, not the module it happens to live in.

Digital workers beside real ones

Every person has an agent working in the background on their behalf — preparing, chasing, watching for what they would otherwise miss. Assistance attached to a role rather than a chat window.

A decision log for everything

What was observed, what was concluded, what was done and who approved it. Written permanently, readable by a person, and the reason any of this is defensible to an auditor.

Physical operations included

Stock, movements, production and delivery as first-class parts of the same record rather than a warehouse product bolted to the side. Companies that make and move things are the harder case and the intended one.

Runs where you decide

Your hardware, your cloud account, or a provider you chose. One login for your people, and no obligation to keep your company inside somebody else's product to keep operating.

The Grid

Compute that finds its own capacity

AI work is bursty and hardware is idle most of the day. The Grid lets the machines a company already owns — and machines belonging to others who opt in — offer and take AI capacity from each other, with agents negotiating the terms and every exchange settled and recorded.

No token, nothing to buy. Settlement is an accounting record between participants. There is no coin, no sale and no investment of any kind.
Optional by design. A company running on one machine never touches the Grid. It is an efficiency, not a dependency.
You own the hardware. Capacity is offered from equipment you control, on terms you set, and withdrawn whenever you want it back.

The Factory

The system is largely built by the system

hurukan is developed in an autonomous software factory: specialised agents write, test and integrate continuously, so a change reaches a running deployment in a fraction of the usual time. That is a real advantage and it is also the part most likely to be oversold, so here is where the automation stops.

Tests and review are not skipped. Generated code goes through the same gates as any other. Faster gates, not fewer.
Releases have a human owner. A person decides what ships to a running company. That will not be automated away.
Speed is the point, not headcount. The claim is a shorter path from a defect to a fix, and it should be judged on that.

The picture

Intelligence, in the narrow and lawful sense

The system builds a continuous picture of the conditions a company is operating in — public information, geospatial data, market and supply-chain signals, its own telemetry, and structured reports from its own staff. The intelligence vocabulary is borrowed for the method, and it stops well short of what those words mean in their original setting.

No interception. Ever. Nothing reads anybody's communications, networks or systems. The only signals used are your own.
No covert collection about people. Structured input means your own staff writing down what they saw. It is not surveillance, and no dossiers on individuals are built.
Public sources stay public. Open information and licensed data, used within its terms, under GDPR, with a named controller.

Sovereign, composable, and open where it counts

A system that runs an entire company is the deepest dependency that company will ever take on. That makes the licence and the architecture more important than any feature: the parts you depend on must be readable, replaceable and runnable by somebody who is not us. Composable means you can adopt one domain at a time and keep what already works.

Open source at the core. Readable, forkable, and maintainable by another team. Announced in a licence file rather than in a headline.
Runs on your infrastructure. Your machines or an account you hold, with no path by which a commercial decision elsewhere stops your operations.
Adopt one domain at a time. Orders first, or stock first, alongside what you already run. Replacing everything at once is how these programmes fail.
An exit that is written down. What you get, in what format, in what window. On a system this deep it is the most important clause in the contract.

What a system this large has to promise

The more a system does on its own, the more it has to be able to explain and undo. Those two properties are what separate an operating system a company can be answerable for from an impressive demonstration, and they are harder to build than the autonomy itself.

Bounded by default. Every agent has an explicit scope: what it may touch, up to what value, and what it must ask about.
Reversible where possible. Actions carry their inverse. Where an action cannot be undone, it is one that waits for a person.
Explainable after the fact. Every decision reconstructable months later, because the first serious dispute will require exactly that.
Interruptible from outside. A stop that does not run on the system it stops, exercised on a schedule rather than kept for emergencies.

Where the programme is

Ambitious scope, early stage, said in that order

The scope described above is what hurukan is being built to do. This is what it actually does today.

Now
Running our own operations
The record, the agents and the decision log carrying real work in companies we are responsible for. Not a demonstration environment, and not yet anybody else's business.
Next
First companies, one domain at a time
A small number of businesses, starting with a single domain alongside what they already run, chosen for fit rather than size, and watched through a bad month.
Later
The last system standing
Replacing the remaining products rather than integrating with them, which is the actual test of the idea and the part nobody should be asked to believe in advance.

FAQ

Frequently asked questions

What is hurukan in one sentence?

An AI-native operating system for a whole company: one record instead of several systems, agents that carry out work rather than reporting on it, and a decision log that makes the whole thing answerable.

Does it really replace ERP, CRM and WMS?

That is the design intent and the honest state today is that it has not replaced them anywhere except for us. It is built to be adopted one domain at a time alongside what you run, which is also how you would find out whether the claim holds.

What is the Grid?

A way for hardware to offer and take AI capacity from other hardware instead of everything running in one place. Agents negotiate the terms and every exchange is settled and recorded. It is optional — a company running on one machine never uses it.

Is there a token, a coin or anything to invest in?

No. Settlement between Grid participants is an accounting record, not a transferable instrument. There is no token, no sale, no allocation and nothing to invest in, and if that ever changed it would carry regulatory obligations we would state before anything else.

What do the AI engineers in the Factory actually do?

Write, test and integrate code continuously, under the same review and test gates as any other change, with a person deciding what reaches a running company. The claim is a much shorter path from defect to fix, not software without engineers.

What does an agent-avatar do for an employee?

Works in the background on that person's actual role: preparing the things they would otherwise write from scratch, watching for what they would otherwise miss, and asking rather than assuming. It does not evaluate them and it is not a monitoring tool.

Is the intelligence capability surveillance?

No, and the boundary is stated on this page rather than in a policy nobody reads. No interception of anybody's communications or systems, no covert collection about people, no dossiers on individuals. Public sources, licensed data, geospatial information, your own telemetry and your own staff's reports.

What will it cost, and can we leave?

Pricing is not published because quoting a system of this scope before the first engagement would be a number invented to look credible. The exit is decided: what is handed over, in what format and within what window, agreed at the start, because on a system this deep that clause matters more than the price.

Start with one domain, not with a transformation

The waitlist form goes live soon. Until then the blog carries the argument, including the parts where the automation deliberately stops.

Contact Read the blog

No form yet, no data collected — just come back.