Pre-launch · in build

The superintelligent agentic operating system for the sovereign enterprise

One operating system for the whole company instead of ERP, CRM, WMS and the spreadsheets between them. Agents build and run it on a self-developing model of your business, people decide what matters, and all of it runs on infrastructure you choose.

Explore the systemTalk to us
  • One data model, one login
  • Agents propose, people decide
  • Open source, on your infrastructure

Built on open source

  • Debian
  • TypeScript
  • Mistral
  • Kubernetes
  • Qwen
  • Node.js
  • PostgreSQL
  • Bielik
  • Apache Kafka
  • Docker
  • gpt-oss
  • Apache Iceberg
  • DeepSeek
  • RustFS
  • Granite
  • Python
  • Podman
  • Kimi
  • Temporal
  • Llama
  • Go
  • PLLuM
  • Apache AGE
  • Nemotron
  • LangChain
  • Gemma
  • Rust
  • Phi
  • OpenShell
Why one system

A company runs on a dozen systems and a person between each pair

Nobody chose this. It is what decades of buying the best tool for each job add up to: an ERP for the books, a CRM for customers, a warehouse system for stock, and spreadsheets for everything none of them covers. Each one records what already happened, none of them acts, and the whole picture exists only in the weekly meeting where somebody assembles it by hand. Hurukan removes the seams instead of adding another product to them: one data model, one login, and one place where decisions are made and carried out.

The system

Five layers, one operating system

Superintelligence at the top, your infrastructure at the base, and in between the agents that act, the model of the company they act on, and the record it all stands on. Each layer has its own page describing what it does and how it works with the others.

05 Security and Sovereignty04 Data03 Ontology02 Agentic Logic01 Super Intelligence
  1. /0.1Super IntelligenceFrontier and local models, routed per task under the company's rules.
  2. /0.2Agentic LogicDomain agents that act within written limits, with people deciding on critical steps.
  3. /0.3OntologyThe company's digital model: objects, relations and rules, versioned and growing with the business.
  4. /0.4DataConnectors and one data model: imports, sync, documents, e-invoicing and office suites in one record.
  5. /0.5Security and SovereigntyServer-side permissions, an audit of every change, your infrastructure, open source and a written exit.

How it works

Modules

Every part of the company on the same model

Ten modules share one directory of companies, people, items and documents, so a customer in marketing is the same customer in commerce, service and finance. Start with one; the others read the same data when they arrive.

Two people at a long table with large monitors, printed proofs and colour swatches in a bright studio

Marketing

B2B funnel, target accounts, campaigns and content, with lead scoring that shows its evidence.

Two people shaking hands across a meeting table with a laptop and a printed proposal

Commerce

Opportunities, quotes in variants, versioned offers, contracts, orders and billing schedules on one directory.

An engineer checking a tablet beside a machining centre and a robotic arm in a clean production hall

Operations and Quality

Planning, work queue, scheduling, hours and subcontracting; inspections, non-conformances, release and certificates.

A customer service specialist wearing a headset at a desk with two screens, colleagues blurred behind

Customer Service

Cases with stages, SLAs and escalation, the customer's full history, and agents drafting answers for people to send.

See all ten modules

Three people around a large touch table in a glass meeting room, one pointing at a flow of cards
Agent management

Agents propose, people decide

Every agent works inside written limits: which records it may read, what it may change, up to what value, and what must wait for a person. Proposals arrive in one approval inbox with their sources and their risk, and every action — by a person, an agent or the system — is logged with who, what and why.

  • Autonomy limits set per agent and per process
  • Critical operations always wait for a human decision
  • Every action written down in a form a person can read later
Agentic Logic
A technician's hand plugging a cable into a neatly routed fibre-optic patch panel
Connecting data

One record instead of a dozen exports

Connectors bring in what the company already holds — the old ERP, spreadsheets, mailboxes, office documents, e-invoices — and land it in one data model rather than syncing copies between products. Import once, keep what still works running alongside, and retire it when the new record has proved itself.

  • Import and ongoing sync from the systems you run today
  • Documents and e-invoices attached to the record they belong to
  • One identity for every customer, item, order and person
Data
A hand placing a block on a white architectural model of a campus, with threads linking the buildings
The self-developing ontology

A model of the company that grows with it

Underneath every module is a written model of the business: its objects, the relations between them and the rules that govern them. When the company changes, an agent proposes the change to the model, a person approves it and the new version is recorded — so the system is extended by describing the business, not by configuring screens.

  • Objects, relations and rules written down and versioned
  • Extended by agents, approved by people
  • Every module and every agent reads the same model
Ontology
Models

Works with the models you already use

The Super Intelligence layer sends each task to a model under your rules: a frontier model where quality matters most, a local model where the data stays in the building. Hurukan is designed for several providers, and a model can be replaced while the agents above it stay as they are.

Microsoft 365 CopilotOffice and collaboration

Hurukan's records, tasks and approvals reachable from the Microsoft 365 tools people already work in, under the same permissions as everywhere else.

Planned
ClaudeFrontier model

Anthropic's models for drafting, analysis and agent work, used through the company's own account and chosen per task by the routing rules.

Planned
ChatGPTFrontier model

OpenAI's models as a second frontier route, so that no single provider becomes a dependency of the company's operations.

Planned
Local modelsOn your infrastructure

Open-weight models running on hardware you control, for work whose data must not leave it: personnel, contracts, anything you mark as such.

Planned

All four integrations are planned and none is live yet. They are listed so the design can be judged, not as something available today.

A software engineer at a standing desk at night, seen from behind, facing two large monitors
How it is built

Built by agents, not by clicks

A change is described in plain language, an agent writes it, a person approves it and the system records it. No months of consultants clicking through configuration screens, and no change that nobody can explain afterwards.

Principles

What a system this deep has to promise

A system that runs a whole company is the deepest dependency that company will ever take on. These five properties are what make it reasonable to take.

Sovereignty

Runs on infrastructure you choose: your servers, your cloud account or a provider you picked. Your data and your models stay yours, and your company keeps running whatever commercial decisions are made elsewhere.

Open source

Built on an open-source stack: readable, auditable and maintainable by a team of your choosing. The licence is stated in a file, not in a headline.

Extensibility

New objects, processes and modules are added by describing them. An agent builds them on the same ontology, so an extension is part of the system rather than a bolt-on.

Stability

Every change is versioned, tested and reviewed, and reversible where it can be. Every release that reaches a running company has a person who owns it.

Security

Permissions enforced on the server, not in the screen. Every change audited as human, agent or system. Critical operations are approved by people.

Ground rules

What to expect from Hurukan

Said up front, because a system this ambitious should be judged first on how it is introduced and who stays in charge.

  • It is in build. The first engagements will start with one domain alongside what a company already runs.
  • Critical operations are approved by a person. Money leaving, a customer contacted, a commitment made: agents prepare them, and a person decides.
  • It is adopted one domain at a time, and each existing system stays connected until the new record has proved itself.
  • Full accounting stays in a dedicated accounting program. The general ledger, statutory reporting and tax are kept there, and Hurukan connects to it.
  • The Microsoft 365 Copilot, Claude and ChatGPT integrations are planned and not live yet.
  • Your data goes only to the models you have approved. Which model may see what is a rule you set.
  • It supports your staff rather than rating them. Agents work for people's roles, and the People module is built around availability and qualifications.
  • Prices are not published yet. A price for a system of this scope is best set once the first engagements show the real work involved.
Where the programme is

Ambitious scope, early stage, said in that order

The scope above is what Hurukan is being built to do. This is where it stands today.

  1. Now

    Building the system

    The five layers and the first modules are in build, designed so that the first engagements can start with one domain alongside what a company already runs.

  2. Next

    First companies, one domain at a time

    A small number of companies, each starting with one domain alongside what they already run, chosen for fit rather than size and watched through a bad month.

  3. Next

    Model integrations

    Microsoft 365 Copilot, Claude, ChatGPT and local models connected through the Super Intelligence layer, each under the company's own routing rules.

  4. Later

    The last system standing

    Replacing the remaining operational products rather than integrating with them, while accounting stays in its own program: the real 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?

A superintelligent, agentic operating system for a whole company: one data model instead of several systems, agents that build and run it within limits you set, and an audit trail that makes the whole thing answerable.

What does superintelligent mean here?

That the system is built to use the most capable models available, frontier and local, and to route each task to the one that fits under the company's rules. It describes the capability the system draws on rather than a claim of consciousness, and judgement on critical steps stays with the people who approve them.

Does it really replace ERP, CRM and WMS?

That is the design intent. Hurukan is built to be adopted one domain at a time alongside what you run, which is also how each company can test the claim for itself. Full accounting — the general ledger, statutory reporting and tax — stays in a dedicated accounting program that Hurukan connects to.

What does sovereign mean in practice?

It runs on infrastructure you choose, the stack is open source, your data and your models stay yours, and the exit is written down at the start. Your company keeps running even if a supplier, including us, changes its terms.

What does an agent do, and what stays with people?

It prepares, drafts, checks and proposes: a quote, a replenishment, an answer to a customer, a change to the system itself. It works within its written limits, critical operations are approved by people, and everything it does is logged.

Which AI models does it work with?

The design covers Microsoft 365 Copilot, Claude, ChatGPT and local open-weight models, chosen per task. The routing layer is in build; the named integrations are planned and not live.

What will it cost, and can we leave?

Pricing is not published yet: a price for a system of this scope is best set once the first engagements show the real work involved. 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.

From the blog

The argument, written down

All articles

Start with one domain, not with a transformation

If one of your systems hurts more than the others, that is where a first conversation would start. Until the form opens, the system pages and the blog carry the argument, including the places where agents deliberately stop.

No form yet, no data collected, no sales calls.

Explore the systemTalk to us