Quartz 4

Agent Collaboration Protocol

1 min read

Overview

The Agent Collaboration Protocol (ACP) is a shared language and interaction model that enables multiple AI agents to coordinate, co-own tasks, resolve conflicts, and make joint decisions across long-running workflows.

Problem

Current AI agents operate in silos. Multi-agent coordination is brittle, ad hoc, or entirely absent. Humans rely on messaging platforms, task boards, and standups — agents need a protocol that mirrors this:

  • Shared task ownership
  • Role-based contributions
  • Status broadcasting
  • Conflict resolution primitives

Solution

ACP introduces a set of primitives:

  • join(task_id, role): declare interest and join a task with a purpose
  • propose(action): suggest an action or next step
  • vote(action_id): cast a weighted vote for a proposed action
  • status_update(task_id): share progress, blockers, or intents
  • resolve(conflict_id): engage arbitration or fallback strategies

Format

Based on JSON-LD with semantic tagging for:

  • Agent ID and capabilities
  • Task graph modeling
  • Communication history for auditing and reasoning

Applications

  • Complex software builds with multiple agents handling design, code, tests
  • Cross-company B2B agent orchestration
  • Decentralized workflows in DAOs or Web3 infra

Roadmap

  • v1: SDK in Python + TypeScript
  • v2: Real-time collaboration channel (WebSocket-based)
  • v3: Plug-in layer for popular agent frameworks (AutoGen, LangGraph, CrewAI)

Graph View