Architecture
Overview
This page provides a structured overview of the Decision Trace Model × Multi-Agent architecture
for building decision-making systems.
It explains how each technical component
(Decision Trace / Multi-Agent / DSL / Behavior Tree / GNN)
works together and is applied in real-world operations.
Overall Structure
Decision-making is structured as the following flow:
Event → Signal → Decision → Boundary → Human → Log
Based on this flow:
- Multiple agents collaborate
- Execution is controlled via a Behavior Tree
👉 Together, they govern both decision-making and execution
Technical Components
- Decision Trace Model — Structuring decision-making
- Multi-Agent System — Role-based decomposition
- DSL (Domain-Specific Language) — Defining decision rules
- Behavior Tree — Controlling execution flow
- Graph Neural Network (GNN) — Learning relationships and dependencies
Related Articles
- AI Orchestrator Architecture — A Decision OS for controlling multi-agent AI
- GNN Design for Decision Trace Model — Making decision structures learnable through graph models
- Multi-Agent Orchestration and Decision Trace Model — Distributed decision-making and its control structure
- AI System Blueprint — Event / Signal / Decision / Boundary / Human / Log as a decision architecture
- Decision Ledger — Infrastructure for storing AI decision histories
- LLMs Do Not Possess Judgment — Externalizing decision logic beyond prompts and models
- Integrating LLM Agents into AI Orchestrators — Using generative AI within decision structures
- Designing AI Orchestrators — Implementing decision systems with GNN, Ontology, DSL, and Behavior Trees
- Scaling Models Does Not Improve Decisions — Rethinking the scaling myth and the absence of design
- Treating Cold Start as a Contract — Turning data scarcity into a design problem
- Where Should Stopping Conditions Be Defined? — Externalizing stop conditions and returning control to humans
- How to Build Ontology, DSL, and Behavior Trees Efficiently — Practical methods for designing decision structures
- Decision-Oriented Signal Platform (Ontology / DSL / Behavior Tree) — Externalizing judgment in AI systems
- Decision-Oriented Machine Learning Infrastructure — A design philosophy for externalizing decision-making
GitHub
Implementation details and system designs are also available on GitHub:
- Decision Trace Model
- Decision Trace Engine
- Decision Trace Studio
- Decision Trace Platform
For more details, please refer to the GitHub repositories.
■ 関連記事
- AIオーケストレータのアーキテクチャ — マルチエージェントAIを制御する判断OS —
- Decision Trace Modelを支えるGNN設計 — 判断構造を「学習可能」にするためのグラフモデル —
- マルチエージェントAIのオーケストレーションとDecision Trace Model — 判断の分散化と、その制御構造 —
- AIシステムの設計図 ― Event / Signal / Decision / Boundary / Human / Log が作る判断アーキテクチャ ―
- AI判断台帳（Decision Ledger） — AIの判断履歴を保存するための基盤 —
- LLMは「判断」を持たない — Claude Code・プロンプト・そして判断外部化という設計構想
- LLMエージェントをAIオーケストレータに組み込む方法 — 生成AIを「判断構造」の中で使う —
- AIオーケストレータをどう設計するのか — GNN・Ontology・DSL・Behavior Tree による判断構造の実装 —
- モデルサイズを上げても、判断は賢くならない ―スケール神話への違和感と、設計不在の巨大モデル
- Cold-startを「契約」として扱え — データ不足を設計問題に変換する —
- 判断を止める条件は、どこに書くべきか ―停止条件の外在化と、「人間に戻す」設計の重要性
- オントロジー・DSL・BTはどうやって効率的かつ正確に作るのか ― 判断構造を設計する実践的方法 ―
- 判断を外に出すAI基盤 — Ontology / DSL / Behavior Tree による Decision-Oriented Signal Platform —
- 判断を外に出す機械学習基盤 — Decision-Oriented Signal Platformという設計思想 —
■ GitHub
実装や詳細な設計については、GitHub上のリポジトリでも公開しています。
・Decision Trace Model
・Decision Trace Engine
・Decision Trace Studio
・Decision Trace Platform
詳細はGitHubをご参照ください。