Olympus
Model the institution.Then run it.
A working model of what your organisation is, what it may do, and the rules that govern it, authored by your people and our agents together.
- CLASS
- Decision infrastructure
- DEPLOYMENT
- On-prem · Sovereign · Air-gapped
- AUTHORING
- Agent-assisted, human-ratified
- RECORD
- Immutable by construction
Proposes relation Sortie → Part [consumes] · inferred from 2,481 closed work orders
RatifyRejectEvery institution alreadyhas a model of itself.
It is split across a scheduling system, a spreadsheet, a regulation, and four people who know the exceptions. None of it is in a form anything can check, so the rule that matters is applied by whoever happens to be on shift.
Entities
What exists
The nouns the organisation actually works in: an airframe, a ward bed, a substation, a case file, a counterparty, a shift. Not tables, and not microservices, but the things people name out loud when they describe their day. Each is declared once, and every system that touches it refers to the same declaration.
The view at the top of this page follows this list. Selecting a part changes what the interface is showing.
A model nobody maintains is worse than no model, because it is trusted for a while before it is found out. Olympus is built around the only maintenance pattern that survives contact with a real organisation: the people who do the work own the model, and agents do the labour of keeping it honest.
Agents read what is already in place — the schemas, the logs, the forms, the exception handling, the way a rule is actually applied rather than how it is written down — and put forward entities, relations and rules they can evidence. Each proposal arrives with what it was inferred from and what it would change.
Nothing an agent proposes takes effect on its own. A human with the authority accepts it, rejects it, or corrects it. The model is the institution's statement about itself, and only the institution can make it.
Because actions are typed and rules are checked at the point of invocation, the audit trail is not a feature somebody had to build. It is a by-product of the model being the only route by which anything changes. Every state the organisation has been in is reconstructable, along with who moved it and under which rule.
That matters most on the day it is questioned: an incident review, an inquiry, a regulator asking why a thing was released. The answer is not a reconstruction assembled from logs after the fact. It is the record.
- 01Declarative ontology: entities, relations, states, actions and rules in one model
- 02Agent-assisted authoring with human ratification on every change
- 03State machines enforced at the point of action, not validated after the fact
- 04Policy written against the model rather than embedded in applications
- 05Complete, immutable decision history as a property of the design
- 06Role and clearance aware: the authority to act is part of the model
- 07Runs on-premise, in a sovereign region, or fully air-gapped
Olympus does not ask an organisation to replace what it runs on. Connectors read from the systems of record already in place — ERP, EHR, asset management, case management, scheduling, SCADA, ticketing — and write back through the same interfaces, so the model is a layer over the estate rather than a migration of it.
Where a decision has to be executed by a system that cannot be changed, Olympus holds the model and hands the step over with its evidence attached.
Tell us the decision that is going wrong.
Not the roadmap. Not the stack. The call. We answer within 24 hours, including when the answer is that this is not our work.
hello@rynth.ai