PentacostAI
Agentic frameworks for legacy systems

Legacy systems, mapped into ontology. Run by BPMN and DMN.

PentacostAI builds agentic frameworks that adopt legacy systems into ontology maps. The entities, rules and workflows your systems already hold become one explicit, standards-based model. Agents act on that model through BPMN processes and DMN decisions: every action traceable, every outcome auditable, every actor governed.

The approach

From legacy system to governed agent.

A legacy system already encodes how an organization works, in schemas, codes and procedures that are rarely documented. We make that knowledge explicit as an ontology map, then drive it with open process and decision standards. Agents do not improvise around the system. They execute its model.

01

Adopt

What does the system already know?

The legacy system's data, codes and procedures are the starting point. Its existing knowledge is captured, not rewritten.

Legacy dataSchemasProcedures
02

Map

What exists, and what does it mean?

An ontology map grounded in a foundational ontology. Every term has one meaning, and every relation is checked by a reasoner.

OWLUFOOntoUMLgUFO
03

Drive

What happens, and what decides?

BPMN processes define the steps and their order. DMN decision tables define each outcome, readable and auditable by the domain expert.

BPMNDMNFEEL
04

Govern

Who acts, and with what freedom?

People, programs and AI agents act under explicit roles. Where judgement belongs to a person, the run waits for that person. Every run keeps its provenance.

RolesHuman in the loopPROV-O

Legacy system → ontology map → BPMN & DMN → governed agents

The method

A definition and its execution are different things.

A process exists twice: once as defined, and once for each run that performed it. A decision exists twice too: as its table of rules, and as each outcome with the rule that produced it. Keeping both makes an agentic system explainable by construction.

Modelled once, in plain language

The people who know the subject state what they mean once, in their own words. The framework turns that into a model they can read back, confirm or correct. Nobody has to rewrite their meaning as code.

Executed by one engine, with evidence

A general engine carries the model out. Every run records what happened, who acted, and which rule decided, so every answer can be traced to its source.

SABiOSystematic approach for building ontologies
SEONThe Software Engineering Ontology Network
UFOUnified Foundational Ontology
BPMN & DMNProcess and decision standards from the Object Management Group
OWL & SPARQLKnowledge representation and query from the W3C
PROV-OProvenance of every run and every outcome
Products

The framework, applied.

In development

Tungs

The knowledge-driven engine at the core of the framework. It reads a sentence, places its terms on the ontology, runs the process and decisions the model defines, and hands every judgement call to the responsible person.

In development

PentacostAI Health

An interactive guide to the WHO digital health ecosystem. It shows health specialists and technologists how the phases of a digital health enterprise relate, and where Digital Adaptation Kits sit within them.

Live on Algorand TestNet

AlgoCity

An agent-native city on Algorand. People and AI agents take seats at game tables, pay through the x402 payment protocol, and play under rules that smart contracts enforce.

Visit AlgoCity →