A foundational framework for multi-agent consensus and AI–human agreement models.
Consensus Architecture is the world’s first framework that defines
how multiple intelligent agents — LLMs, humans, or hybrid networks —
negotiate, align values, and form shared decisions using natural language.
While Natural-Language OS (NL-OS) defines the internal cognition of a single LLM,
Consensus Architecture defines the interaction layer, focusing on:
- LLM ↔ LLM consensus protocols
- Human ↔ LLM negotiation structures
- Multi-agent agreement and decision fusion
- Conflict resolution and alignment strategies
This repository provides the initial specification, early models,
and conceptual foundations for future AI societies and cooperative intelligence.
A full formal version will be published in the upcoming
Consensus Architecture Whitepaper (Ochiai, 2025) on Zenodo.
Consensus Architecture(コンセンサス・アーキテクチャ) は、
複数の知的エージェント(LLM同士・人間×LLM・混合ネットワーク)が
自然言語を用いて 合意形成・価値整合・意思統合 を行う仕組みを定義する、
世界初の理論フレームワークです。
Natural-Language OS(NL-OS)が「1つのLLMの内的認知構造」を扱うのに対し、
Consensus Architecture は エージェント間の相互作用レイヤー を扱います。
- LLM ↔ LLM の合意形成プロトコル
- 人間 ↔ LLM の交渉構造
- マルチエージェントの意思統合アルゴリズム
- 衝突解消と整合性戦略
将来の AI社会・協調知性(Cooperative Intelligence) の基盤となる理論を
本レポジトリで公開します。
正式な理論・数理モデル・全体系は
Consensus Architecture Whitepaper(Ochiai, 2025)
として Zenodo にて公開予定です。
Consensus-Architecture/ ├─ README.md ├─ core/ ← 基本概念・モデル ├─ protocols/ ← 合意形成プロトコル ├─ examples/ ← 事例・ケーススタディ └─ papers/ ← Whitepaper(Zenodo版)置き場
- This README is the short public version (English + Japanese).
- After publishing the Zenodo Whitepaper,
replace this README with a longer English-only version.