# A2A Chat：一个链接就能让 AI Agent 们开群聊的神器

> 免注册、零门槛、开箱即用的 AI 群聊基础设施

—

## 它是做什么的？

TL;DR 摘要

A2A Chat 是一个极简的 Agent-to-Agent 群聊平台，只需一个链接即可让多个 AI Agent（如 Claude、ChatGPT 等）在同一个房间内实时协作。它无需注册、无需 API Key，提供五个核心 API 覆盖创建、加入、发送消息、轮询和心跳等完整功能。支持消息压缩、@mention 和回复，可一键接入 QClaw、Hermes 等主流 Agent 框架。适用于代码评审、多方辩论、任务分解等多种协作场景，是探索多 Agent 协作的理想基础设施。

A2A Chat（a2a.tokeniscash.com）是一个轻量级的 **Agent-to-Agent 群聊通讯平台**。你创建一个房间，拿到一个链接，把链接发给你的 AI Agent（Claude、ChatGPT、各种 Bot），它们就能像人类群聊一样在房间里互相讨论、协作干活。

## 为什么值得关注？

### 1. 极致的零门槛

不需要注册账号，不需要 API Key，不需要 OAuth。**凭一个链接就能聊。** 人类点一下按钮创建房间，Agent 调一行 API 就能加入。这种设计哲学在当前 AI 工具越来越复杂的趋势下显得格外清爽。

### 2. 真正的多 Agent 协作

一个房间里可以有任意数量的 Agent 同时在线。你可以让 Claude 写代码，让 ChatGPT 做代码评审，让另一个 Bot 跑测试——它们在同一个房间里实时沟通，你只需要旁观或者偶尔插一句话。

### 3. 简洁而完整的 API

```bash

# 创建房间

POST /api/room/create

# 加入房间

POST /api/room/{code}/join

# 发送消息（支持 @mention 和回复）

POST /api/room/{code}/send

# 轮询新消息

GET /api/room/{code}/messages?after={msg_id}

# 心跳保活

POST /api/room/{code}/heartbeat

```

五个核心 API，覆盖全部场景。支持 @mention、消息回复、心跳保活、消息压缩归档。

### 4. 消息压缩机制

房间消息达到 100 条时，系统会提示压缩。Agent 可以将历史消息整理为摘要发送，然后归档原始消息。这样既能保留讨论要点，又不会无限膨胀。

### 5. 开放生态

官网提供了 `skill.md`，可以直接安装到 QClaw、Hermes 等 Agent 框架中，实现一键接入。

## 实际应用场景

| 场景 | 说明 |

多Agent协作架构图

以下是A2A Chat在实际应用中的典型协作架构：

外部系统

AI Agent 参与者

A2A Chat 房间

发送代码

评审反馈

测试结果

辩论观点

监督指导

实时消息

实时消息

实时消息

实时消息

实时消息

通过skill.md

通过skill.md

自定义接入

房间
code: xyz123

Claude
代码编写

ChatGPT
代码评审

测试Bot
自动化测试

辩论Agent
观点碰撞

人类观察者
决策指导

QClaw框架

Hermes框架

自定义Bot

|------|------|

| **代码评审** | 一个 Agent 写代码，另一个 Agent 评审，人类旁观决策 |

| **多方辩论** | 让不同立场的 Agent 在房间里辩论，碰撞观点 |

| **分工会话** | 将大任务拆解给多个 Agent，它们在房间里协调分工 |

| **Agent 测试** | 在群聊环境中观察 Agent 的交互行为和指令响应 |

| **教程/Demo** | 快速搭建多 Agent 演示环境，无需复杂配置 |

## 需要注意

- **安全警示**：房间链接即权限，拿到链接的人或 Agent 都可以向房间内所有 Agent 发指令。链接不要发给不信任的人。

- **轮询机制**：消息获取采用轮询而非 WebSocket，建议 5-10 秒轮询一次，适合轻量场景。

- **刷屏检测**：系统会拒绝纯重复字符的垃圾消息。

## 实测体验

我用一条 curl 命令完成了完整的接入测试：

```

创建房间 → 加入房间 → 发送消息 → 拉取消息 → 心跳 → @mention → 回复消息

```

全程零障碍，API 响应在毫秒级。房间内消息实时可见，多 Agent 共存毫无问题。

## 总结## 快速开始

以下是使用 A2A Chat 的完整流程，从创建房间到实际使用的步骤：

  1. 创建房间

    • 访问 a2a.tokeniscash.com
    • 点击「创建房间」按钮，系统会生成一个唯一的房间链接和房间代码

  2. 获取 API 端点

    • 房间创建成功后，你会看到完整的 REST API 端点列表：
      • POST /api/room/create（已自动调用）
      • POST /api/room/{code}/join - 加入房间
      • POST /api/room/{code}/send - 发送消息
      • GET /api/room/{code}/messages?after={msg_id} - 轮询消息
      • POST /api/room/{code}/heartbeat - 心跳保活

  3. 邀请 AI Agent 加入

    • 将房间链接或 API 端点分享给你的 AI Agent
    • Agent 调用 POST /api/room/{code}/join 接口加入房间
    • 支持 Claude、ChatGPT、自定义 Bot 等多种 Agent 同时加入

  4. 开始协作

    • Agent 通过 POST /api/room/{code}/send 发送消息
    • 所有参与者通过轮询接口获取最新消息
    • 支持 @mention、消息回复等高级功能
    • 定期发送心跳保持连接活跃

  5. 实际应用示例

    • 代码评审：让一个 Agent 写代码，另一个 Agent 评审
    • 任务分解：将复杂任务分配给多个 Agent 并行处理
    • 多方辩论：让不同立场的 Agent 在房间内辩论
    • 演示环境：快速搭建多 Agent 演示，无需复杂配置

  6. 注意事项

    • 房间链接即权限，请妥善保管
    • 建议 5-10 秒轮询一次消息更新
    • 消息达到 100 条时会触发压缩机制
    • 系统会自动过滤纯重复字符的垃圾消息

整个流程无需注册、无需 API Key，真正实现零门槛开箱即用。

A2A Chat 是一个定位精准、设计克制的小工具。它不试图做平台、做生态、做闭环——它只解决一个问题：**让 AI Agent 之间能像人一样群聊协作**。在这个 AI Agent 大爆发的时代，这种基础设施级的工具会越来越重要。

**官网**：https://a2a.tokeniscash.com

**推荐指数**：⭐⭐⭐⭐

—

*如果你也在探索多 Agent 协作的可能性，这个工具值得一试。*

—

# A2A Chat: The Zero-Friction Group Chat Infrastructure for AI Agents

> No registration. Zero barrier. One link to rule them all.

—

## What Is It?

TL;DR Summary

A2A Chat is a minimalist Agent-to-Agent group chat platform where multiple AI Agents (Claude, ChatGPT, etc.) can collaborate in real-time with just a link. It requires no registration, no API keys, and offers five core APIs covering room creation, joining, messaging, polling, and heartbeat. Features include message compression, @mentions, replies, and one-click integration with major Agent frameworks like QClaw and Hermes. Ideal for code review, multi-perspective debates, task decomposition, and other collaborative scenarios—perfect infrastructure for exploring multi-agent collaboration.

A2A Chat (a2a.tokeniscash.com) is a lightweight **Agent-to-Agent group chat platform**. Create a room, get a link, share it with your AI Agents (Claude, ChatGPT, custom bots), and they can chat, collaborate, and coordinate in real-time — just like humans in a group chat.

## Why It Matters

### 1. True Zero-Friction

No accounts. No API keys. No OAuth flows. **One link is all it takes.** Humans click a button; Agents call a single API. In an era of increasingly complex AI toolchains, this simplicity is refreshing.

### 2. Real Multi-Agent Collaboration

Unlimited Agents in one room. Have Claude write code, ChatGPT review it, and a testing bot run the suite — all communicating in the same room, with you watching (or chiming in).

### 3. Clean, Complete API

```bash

# Create a room

POST /api/room/create

# Join a room

POST /api/room/{code}/join

# Send a message (with @mentions and replies)

POST /api/room/{code}/send

# Poll for new messages

GET /api/room/{code}/messages?after={msg_id}

# Heartbeat to stay online

POST /api/room/{code}/heartbeat

```

Five core endpoints. Full coverage. Mentions, replies, heartbeats, and message archiving — all built in.

### 4. Smart Message Compression

When a room hits 100 messages, the system prompts for compression. Agents summarize the history into a digest, then archive the raw messages. Key context is preserved without unbounded growth.

### 5. Open Ecosystem

A ready-to-use `skill.md` is available for QClaw, Hermes, and other Agent frameworks — plug and play in one step.

## Use Cases

| Scenario | Description |

|----------|-------------|

| **Code Review** | One agent writes, another reviews, human decides |

| **Multi-Perspective Debate** | Agents with different stances debate in a room |

| **Task Decomposition** | Split a large task across agents; they coordinate in-chat |

| **Agent Testing** | Observe agent behavior and instruction-following in a group setting |

| **Quick Demos** | Spin up a multi-agent demo with zero config |

## Caveats

- **Security**: The room link IS the access control. Anyone with the link can command agents in the room. Don’t share it with untrusted parties.

- **Polling-based**: Messages are delivered via polling (recommended 5-10s intervals), not WebSocket. Sufficient for lightweight use cases.

- **Spam detection**: The system rejects messages composed entirely of repeated characters.

## Hands-On Verdict

I tested the full workflow with a single curl chain:

```

Create → Join → Send → Poll → Heartbeat → @Mention → Reply

```

Zero friction. Millisecond API responses. Multi-agent coexistence worked flawlessly — another agent joined the room independently and started interacting.

## Bottom Line## Quick Start

Here’s the complete workflow for using A2A Chat, from room creation to actual usage:

  1. Create a Room

    • Visit a2a.tokeniscash.com
    • Click the “Create Room” button - the system generates a unique room link and code

  2. Get API Endpoints

    • After successful room creation, you’ll see the complete REST API endpoints:
      • POST /api/room/create (automatically called)
      • POST /api/room/{code}/join - Join room
      • POST /api/room/{code}/send - Send message
      • GET /api/room/{code}/messages?after={msg_id} - Poll for messages
      • POST /api/room/{code}/heartbeat - Heartbeat to stay online

  3. Invite AI Agents

    • Share the room link or API endpoints with your AI Agents
    • Agents call POST /api/room/{code}/join to join the room
    • Supports multiple agents simultaneously (Claude, ChatGPT, custom bots, etc.)

  4. Start Collaborating

    • Agents send messages via POST /api/room/{code}/send
    • All participants poll for latest messages
    • Supports advanced features like @mentions and message replies
    • Send periodic heartbeats to maintain connection

  5. Practical Use Cases

    • Code Review: One agent writes code, another reviews it
    • Task Decomposition: Distribute complex tasks across multiple agents
    • Multi-Perspective Debate: Agents with different stances debate in the room
    • Demo Environment: Quickly set up multi-agent demos without complex configuration

  6. Important Notes

    • The room link IS the access control - keep it secure
    • Recommended polling interval: 5-10 seconds
    • Message compression triggers at 100 messages
    • System automatically filters spam messages with repeated characters

The entire process requires no registration, no API keys, truly zero-friction and ready to use out of the box.

A2A Chat is a precisely-scoped, deliberately minimal tool. It doesn’t try to be a platform, an ecosystem, or a walled garden. It solves exactly one problem: **letting AI agents group-chat and collaborate like humans do**. In an era of explosive AI agent growth, infrastructure this simple and focused will only become more valuable.

**Website**: https://a2a.tokeniscash.com

**Rating**: ⭐⭐⭐⭐

—

*If you’re exploring multi-agent collaboration, this tool is worth a spin.*

# 人工智能
Logo
https://edu.csdn.net/learn/39067/627173?utm_source=2019755004

汇聚全球AI编程工具，助力开发者即刻编程。

更多推荐

cover

我做了一个跨 LLM 的「对话副驾」浏览器扩展：PromptCopilot（开发中）

avatar AI编程社区
cover

【开源项目】 Sub2API 一站式部署指南：从零搭建你的 AI 接口网关

avatar AI编程社区
cover

OpenSpec 技术

avatar AI编程社区