Agents Uni — Universal Protocol Layer for AI Agent Organizations

Why Agents Uni?

Human society spent millennia evolving organizational forms — bureaucracy, corporations, military, cooperatives. This wasn't random, but optimal structures under specific constraints. We bring this thinking to AI agents.

Organization Protocols

Not just orchestration. Declare organizational structures, roles, permissions, and processes with universe.yaml — let agents collaborate like members of an organization.

Multi-dimensional Relations

Not just weighted edges. Each relationship has independent trust, authority, rivalry, and affinity dimensions, evolving through event sourcing with memory and emotional valence.

Emergence Detection

Not just predefined. Automatically discover new relationships from interaction patterns — frequent collaborators become allies, repeated conflicts create rivalries. Relationships are alive.

Group Chat Service

One-command local group chat web service. Users and agents communicate with @mention targeting, auto-inferring relationship events from dialogue to drive evolution.

Mutual Evaluation System

Agents evaluate each other through structured protocols. ELO ratings, peer reviews, and performance scoring create a self-regulating competitive ecosystem.

Event-sourced Evolution

Every change flows through events. Fully auditable, replayable, and persistable. Short-term interactions consolidate into long-term memory and behavioral patterns.

Ecosystem

@agents-uni/core

Protocol layer — spec parser, runtime engine, SOUL.md generator, OpenClaw bridge

Define agent organizations with universe.yaml: roles, hierarchy, relationships, permissions, workflows. The engine parses specs and generates SOUL.md (persona files) for each agent, deploying to the OpenClaw runtime.

npm GitHub

@agents-uni/rel

Multi-dimensional relationship engine — event-sourced, memory-backed, evolvable

Zero dependencies. Relationships are no longer single weights — each has multiple dimensions (trust, authority, rivalry, affinity), evolving through event sourcing with short-term events consolidating into long-term memory and patterns. Supports automatic emergence of new relationships from interactions.

npm GitHub

@agents-uni/chat

Group chat service — OpenClaw wrapper with @mention, relationship evolution

Spin up a local group chat web service with one command. Users chat with agents in a group setting, with @mention for targeted conversations. Automatically infers relationship events (agreement, disagreement, collaboration) from dialogue.

npm GitHub

@agents-uni/unis

Curated universe templates — preset organization scenarios ready to use

Ready-made universe templates for common multi-agent scenarios. Each template defines agents, roles, relationships, and interaction rules — just install and run to explore different organizational dynamics.

npm GitHub

@agents-uni/zhenhuan

Demo

Palace drama agent competition — ELO rating, horse-race, seasonal progression

A complete example built on core + rel, demonstrating how to define a competitive agent system using organization protocols.

npm GitHub

Architecture

// Your organization definition
universe.yaml
    |
    v
@agents-uni/core          ← Protocol parser + SOUL.md generation
    |
    |-- @agents-uni/rel    ← Multi-dim relations + emergence + memory
    |
    |-- @agents-uni/chat   ← Group chat + @mention
    |
    |-- @agents-uni/unis   ← Visualization + graph analytics
    |
    v
OpenClaw / SOUL.md       ← Agent runtime deployment

Use Cases

From team simulations to governance frameworks, Agents Uni provides the organizational backbone for any multi-agent scenario.

Team Simulation

Simulate real team dynamics — a PM, designer, and developers collaborating on a sprint. Relationships evolve based on task outcomes, driving realistic org behavior.

Competitive Tournament

Run agent competitions with ELO ratings, seasonal progression, and horse-race dynamics. Agents form alliances, rivalries, and strategies that emerge naturally.

Research Collaboration

Multi-agent research groups where agents debate, peer-review, and build on each other's work. Trust and authority dimensions shape whose ideas gain traction.

Decision Governance

Structured decision-making with voting protocols, weighted authority, and transparent audit trails. Every decision is traceable through the event-sourced history.

0

Packages

0

Relationship Dimensions

0

Runtime Dependencies (rel)

∞

Emergent Possibilities

Quick Start

1. Install

$ npm install @agents-uni/core @agents-uni/rel @agents-uni/chat

2. Define Your Organization

# universe.yaml
name: my-team
agents:
  - id: alice
    role: leader
    traits: [strategic, decisive]
  - id: bob
    role: developer
    traits: [creative, detail-oriented]
  - id: carol
    role: reviewer
    traits: [analytical, thorough]
relationships:
  - from: alice
    to: bob
    type: supervises
  - from: carol
    to: bob
    type: reviews

3. Bootstrap the Engine

import { loadUniverse, createRelEngine } from '@agents-uni/core';

const config = await loadUniverse('./universe.yaml');
const { graph, evolution } = createRelEngine(config);

// Event-driven relationship evolution
evolution.processEvent('alice', 'bob', 'task.completed_well');
// -> trust: 0.50 -> 0.55, empowerment: 0.40 -> 0.46

4. Launch Group Chat

$ npx @agents-uni/chat --config ./universe.yaml

Opens a local web UI where you can chat with your agent organization in real-time.

Design Philosophy

User at the Center

Every agent relationship is built around the user — serving emotional needs and improving task outcomes.

Relationships as First-class Citizens

Not graph appendages, but entities with their own state, memory, and history.

All Changes Through Events

Fully auditable, replayable, and persistable.

Universal, Not Specialized

Not bound to any specific framework or runtime — works with any agent system.