Published May 5, 2026 | Version v2
Preprint Open

Active Things Modelling Methodology for Agent-Native Knowledgebases

Authors/Creators

Description

 Abstract

Modern artificial intelligence (`AI`) agents increasingly use tools, retrieval, workflows, and long-context reasoning, yet their operational worlds are often exposed as fragments: rows, documents, messages, statuses, dashboards, and hidden process logic. Relational systems gained a modelling grammar through the `entity-relationship diagram` (`ERD`). Enterprise AI still lacks an equivalent practical standard for agent-operable knowledge. This paper proposes the `Active Things Modelling Methodology` (`ATMM`) for modelling knowledge as a world of explicit, stateful, active things.
ATMM begins from the claim that an active thing becomes knowable through boundary and that its boundary becomes intelligible through lifecycle. A Knowledgebase is therefore treated as an authored operational knowledge world for agents. ATMM defines `Knowledgebase`, `Active Thing Type`, `Active Thing Instance`, `Identity Index`, `Lifecycle Memory`, and `Canonical Events` as core structures. Identity Index exposes the active skeleton of a thing before action begins; Lifecycle Memory records how the thing has lived rather than merely what changed; Canonical Events preserve shared occurrences and their per-thing consequences.

The paper advances two bounded claims. First, ATMM is universal methodologically: primitives of boundary, lifecycle, state, event, transition, relation, and purpose recur where knowledge concerns active things; this is not empirical exhaustion. Second, `Agent-Native` means that the Knowledgebase already exposes the identity, lived path, lawful movement, relation, impact, and monitoring surfaces an agent needs. The empirical program includes one full order-processing proof of concept and two lighter transfer PoCs. Current results support ATMM as a serious candidate foundation, not an industrially validated standard.

Keywords

knowledge engineering; agent-native knowledgebase; lifecycle modelling; active things; case modelling; enterprise AI; knowledge representation
 

Highlights

- Active Things Modelling Methodology composes knowledge around active things.
- Identity Index exposes bounded active identity before action begins.
- Lifecycle Memory records lived movement rather than generic history.
- Canonical Events preserve shared occurrences and per-thing consequences.
- Evidence is bounded methodology evidence, not industrial validation.

Files

ATMM-paper-preprint-v3.pdf

Files (713.1 kB)

Name Size Download all
md5:2ab60f661bd2e18e20bb5eb49e9b17ef
713.1 kB Preview Download

Additional details

Software

Programming language
Python