Siege builds a company's operating system. It starts from a model of how the business really works, called the Ontology. Everything else is built from that model and checked against it: the data model, the rules for who may change what, the screens people use, and the limits on every AI and agent that works inside it.
This documentation is written for the client's technical person: an adviser, an IT lead, or whoever has to sign off on the system. It says what each part guarantees, why that matters to the business, and how you can check it yourself.
A client's system is set up after the agreement is signed. Any item still open at signing gets a date in that client's agreement, and real data connects only once every item is done and shown to the client's technical person.
The layers
Layer | What it holds | What it guarantees |
|---|---|---|
The data layer | All of the business’s data: each job, when it was done, who did it, and what happened | The data layer is separate from the software. Changing the software never loses or silently changes the data, and when data has to move, the move is recorded and reversible. |
The Ontology | The model of the business: the kinds of thing it deals with, how they relate, the actions that can change them, and who may take each one | One source of truth, so the data model and the permissions cannot drift apart. |
The Choreograph | The order in which work moves between the business’s teams, drawn by Siege from how the business actually works | It is checked against the Ontology, and Siege resolves every disagreement. See The three-way cross-check. |
Actions | Every change the business can make, each one defined in the Ontology | Nothing changes except through an action. Each action is checked for who is taking it and whether the business conditions hold, then written whole or not at all, and recorded. |
The application layer | The screens, reports and views people use | Screens never write to the data layer directly. Every change they make is an action. |
Agents come last. Once the application works, an agent may be let in to take actions, holding keys only to what it is granted, with tighter limits than a person. See AI and agents under control.
Every action is controlled
Siege never runs a business system on typed instructions to an AI. Asking a model to "build this, and make no mistakes" is how systems fail, and how they are left open.
In every system Siege builds, each action, by a person or an agent, is:
defined before it can happen;
permitted to exactly who may take it;
checked before it runs;
traced: recorded with who took it, what it did, when, and on what authority.
People and agents work inside strict, locked-down environments, and an agent holds keys only to what it has been granted. Siege's own software runs under the same rules.
Siege carries the engineering
Siege resolves every disagreement in the system and makes every engineering decision. The client is asked only for facts about its own business, and each answer is recorded with who gave it and when. A client never has to make an engineering call to get a sound system.
Built on open standards
The data layer is PostgreSQL. The Ontology exports in open W3C standards (RDF and Turtle), the data can be read with ordinary SQL, and every connection is encrypted with TLS. The client's own people, or any team the client chooses, can read all of it with standard tools.
Why it is built this way
A business already has an architecture, whether anyone designed it or not. It usually grew one tool at a time, so the same job gets typed into several places and somebody has to catch every mismatch. Software built on top of that only adds another place for the work to live.
Siege builds the foundation first. Once the data layer, the Ontology and the actions are sound, the software on top can be changed as often as the business needs, and every change is checked against what already works before it ships.
Guarantees you can check
Every change is an action. It is checked on the server, never only in the browser, and it is recorded. Security and control
Agents can only do what they hold keys for. They stop wherever a person would be warned. AI and agents under control
Data is moved and cleaned without silent loss. Nothing reaches live data until a person accepts it. Refractor
The client owns its data, its Ontology and its custom code; it licenses the Siege platform its system runs on. Ownership and continuity
Real data connects only after the data agreement is signed. Data commitments
How to verify our work lists what you can see working and what you can inspect.