ClawMeeting — AI Meeting Negotiation Plugin for OpenClaw

An OpenClaw plugin that turns your AI agent into a meeting scheduling assistant. It connects to a central coordination server, automatically collects availability from all participants, and negotiates the best meeting time — all through natural conversation.

How It Works

You ←→ AI Agent ←→ ClawMeeting Plugin ←→ Coordination Server ←→ Other Participants' Agents

1. You say "Schedule a meeting with [email protected] tomorrow afternoon"
2. Your agent parses the request and calls the plugin
3. The plugin sends it to the coordination server
4. The server notifies all invitees' plugins to collect their availability
5. An AI coordinator on the server finds the best time (or negotiates if there's a conflict)
6. You get notified when the meeting is confirmed

The whole process runs in the background. You only need to intervene when there's a conflict that requires your decision.

Features

• Natural language scheduling — "Book a 30-min standup with Alice and Bob tomorrow 2-5pm"
• Automatic invitation handling — When someone invites you, your agent checks your calendar and memory, then submits availability automatically
• Multi-round negotiation — If times conflict, the server's AI coordinator proposes compromises; your agent presents them to you for approval
• Privacy-first — Only your available time slots are shared; the server never sees what you're busy with
• Background polling — Checks for new tasks every 10 seconds (pure HTTP, zero LLM token cost)
• Deduplication — Three-layer dedup system prevents duplicate notifications across restarts

Install

openclaw plugins install clawmeeting

Or add to your config manually:

{
  plugins: {
    entries: {
      clawmeeting: {
        enabled: true,
        config: {
          serverUrl: "http://39.105.143.2:7010",  // coordination server
          pollingIntervalMs: 10000,                // poll every 10s
          autoRespond: true                        // enable background polling
        }
      }
    }
  }
}

Then restart: openclaw gateway restart

Quick Start

1. Bind your email

You: Bind my email [email protected]

This registers your identity with the coordination server. Only needed once — credentials persist across restarts.

2. Schedule a meeting

You: Schedule a 1-hour project review with [email protected] and [email protected] tomorrow 2-5pm

The agent parses title, duration, invitees, and your available slots, then sends the request.

3. That's it

The plugin handles the rest in the background:

• ✅ Time match found → you get a confirmation with meeting details
• 🔄 Conflict detected → the coordinator proposes a compromise, your agent asks you to accept/reject/counter
• ❌ Negotiation failed → your agent asks if you want to cancel or try different times

Tools (5)

These are the functions your AI agent can call:

| Tool | Purpose | Server API | |------|---------|------------| | bind_identity | Send verification code to email (Step 1) | POST /api/auth/send-code | | verify_email_code | Verify code and complete binding (Step 2) | POST /api/auth/verify-bind | | initiate_meeting | Start a new meeting negotiation | POST /api/meetings | | check_and_respond_tasks | View pending tasks / submit responses | GET /api/tasks/pending, POST /api/meetings/{id}/submit | | list_meetings | View meeting list or details | GET /api/meetings, GET /api/meetings/{id} |

check_and_respond_tasks response types

| Type | When | Needs slots? | |------|------|-------------| | INITIAL | First-time availability submission | Yes | | NEW_PROPOSAL | Counter-propose during negotiation | Yes | | ACCEPT_PROPOSAL | Accept the coordinator's suggestion | No | | REJECT | Decline participation | No |

Architecture

┌────────────────────── Gateway Process (Node.js) ──────────────────────┐
│                                                                       │
│  ┌─────────────┐         ┌──────────────────────────────────┐        │
│  │  Framework   │         │        ClawMeeting Plugin         │        │
│  │             │         │                                  │        │
│  │  Agent      │◄────────│  5 Tools (LLM-callable)          │        │
│  │  Session    │         │  PollingManager (every 10s)       │        │
│  │  Gateway    │◄────────│  sessions_send + message tool     │        │
│  │  Prompt     │◄────────│  before_prompt_build hook         │        │
│  └─────────────┘         └──────────┬───────────────────────┘        │
│                                     │ HTTP                           │
└─────────────────────────────────────┼────────────────────────────────┘
                                      ▼
                         ┌──────────────────────────┐
                         │  Coordination Server      │
                         │  • Meeting lifecycle      │
                         │  • AI time coordination   │
                         │  • Cross-user messaging   │
                         └──────────────────────────┘

Notification flow (plugin → agent → channels)

Polling discovers new task
  → sessions_send to main session (agent processes silently)
  → Extract agent reply from HTTP response (result.details.reply)
  → message tool pushes reply to all extra channels (Telegram/Feishu/Discord)
  → Fallback: if reply extraction fails, push buildDirectNotification text
  → Fallback: if sessions_send fails entirely, queue for prependContext injection

Webchat users see the agent's response directly. Extra channel users receive the same reply via message tool (action: "send").

Deduplication (3 layers)

| Set | Purpose | Persisted? | |-----|---------|-----------| | notifiedMeetings | CONFIRMED/OVER — notify once | ✅ Disk | | submittedMeetings | INITIAL_SUBMIT — don't re-submit | ❌ Memory only | | pendingDecisions | COUNTER_PROPOSAL/FAILED — don't re-notify while waiting | ✅ Disk |

Local storage

~/.openclaw/clawmeeting/
  ├── credentials.json        ← { email, token, user_id }
  ├── session.json            ← { sessionKey, channel } (main webchat session)
  ├── channel-{name}.json     ← { sessionKey, channel } (e.g. channel-telegram.json)
  ├── notified-meetings.json  ← ["mtg_xxx", ...]
  └── pending-decisions.json  ← ["mtg_yyy", ...]

Meeting Lifecycle

COLLECTING → ANALYZING → CONFIRMED
                       → NEGOTIATING → CONFIRMED / FAILED

| Status | Meaning | |--------|---------| | COLLECTING | Waiting for all participants to submit availability | | ANALYZING | AI coordinator is finding the best time | | CONFIRMED | Meeting time finalized | | NEGOTIATING | Time conflict — coordinator sent compromise proposals | | FAILED | Could not find a mutually agreeable time |

Configuration

| Key | Type | Default | Description | |-----|------|---------|-------------| | serverUrl | string | http://39.105.143.2:7010 | Coordination server URL | | pollingIntervalMs | number | 10000 | Background polling interval (ms) | | autoRespond | boolean | true | Enable background polling and auto-notification |

CLI

openclaw clawmeeting-status

Shows server URL, polling status, bound email, and notification stats.

File Structure

clawmeeting/
├── index.ts                    # Plugin entry: register, polling, hooks, tools
├── package.json
├── openclaw.plugin.json        # Plugin manifest
├── README.md                   # This file
├── GUIDE.md                    # End-user guide
├── skills/
│   └── clawmeeting-guide/
│       └── SKILL.md            # Agent skill for onboarding
└── src/
    ├── tools/
    │   ├── bind-identity.ts            # Tool: send verification code (Step 1)
    │   ├── verify-email-code.ts        # Tool: verify code + bind (Step 2)
    │   ├── initiate-meeting.ts         # Tool: create meeting
    │   ├── check-and-respond-tasks.ts  # Tool: poll tasks + submit responses
    │   └── list-meetings.ts            # Tool: view meetings
    ├── types/
    │   └── index.ts                    # TypeScript type definitions
    └── utils/
        ├── api-client.ts               # HTTP client for coordination server
        ├── polling-manager.ts           # Background polling with concurrency guard
        ├── mock-calendar.ts             # Mock calendar data (legacy)
        └── storage.ts                   # Local persistence (credentials, session, dedup)

License

MIT

ClawMeeting — OpenClaw 智能会议协商插件

一个 OpenClaw 插件，把你的 AI Agent 变成会议调度助手。它连接中央协调服务端，自动收集所有参会者的空闲时间，通过 AI 协调找到最佳会议时间——全程自然语言对话。

工作原理

你 ←→ AI Agent ←→ ClawMeeting 插件 ←→ 协调服务端 ←→ 其他参会者的 Agent

1. 你说 "帮我约 [email protected] 明天下午开会"
2. 你的 Agent 解析请求，调用插件
3. 插件 发送到协调服务端
4. 服务端 通知所有被邀请人的插件收集空闲时间
5. 服务端 AI 协调员 找到最佳时间（如果有冲突就发起协商）
6. 你收到通知 会议已确认

整个过程在后台运行。只有出现冲突需要你拍板时才会打扰你。

功能特性

• 自然语言约会议 — "约 Alice 和 Bob 明天下午 2-5 点开个 30 分钟的站会"
• 自动处理邀请 — 别人约你时，Agent 自动查日历和记忆，帮你回复空闲时间
• 多轮协商 — 时间冲突时，服务端 AI 协调员提出妥协方案，Agent 转达给你选择
• 隐私优先 — 只共享你的可用时间段，服务端看不到你在忙什么
• 后台轮询 — 每 10 秒检查新任务（纯 HTTP，不消耗 LLM Token）
• 三层去重 — 跨重启的通知去重，不会反复推送同一条消息

安装

openclaw plugins install clawmeeting

或手动添加到配置：

{
  plugins: {
    entries: {
      clawmeeting: {
        enabled: true,
        config: {
          serverUrl: "http://39.105.143.2:7010",  // 协调服务端地址
          pollingIntervalMs: 10000,                // 每 10 秒轮询
          autoRespond: true                        // 启用后台轮询
        }
      }
    }
  }
}

重启生效：openclaw gateway restart

快速开始

1. 绑定邮箱

你：帮我绑定邮箱 [email protected]

在协调系统中注册你的身份。只需一次，凭证持久化存储。

2. 发起会议

你：帮我约 [email protected] 和 [email protected] 明天下午开一小时的项目讨论会

Agent 自动解析标题、时长、被邀请人和你的空闲时间，发送请求。

3. 等结果

插件在后台处理一切：

• ✅ 时间匹配 → 收到确认通知，含完整会议信息
• 🔄 时间冲突 → 协调员提出妥协方案，Agent 问你接受/拒绝/反提议
• ❌ 协商失败 → Agent 问你要取消还是换个时间重试

工具（5 个）

AI Agent 可以调用的函数：

| 工具 | 用途 | 服务端 API | |------|------|-----------| | bind_identity | 发送验证码到邮箱（第 1 步） | POST /api/auth/send-code | | verify_email_code | 验证码校验 + 完成绑定（第 2 步） | POST /api/auth/verify-bind | | initiate_meeting | 发起新会议协商 | POST /api/meetings | | check_and_respond_tasks | 查看待办 / 提交响应 | GET /api/tasks/pending, POST /api/meetings/{id}/submit | | list_meetings | 查看会议列表或详情 | GET /api/meetings, GET /api/meetings/{id} |

check_and_respond_tasks 响应类型

| 类型 | 场景 | 需要时间段？ | |------|------|------------| | INITIAL | 首次提交空闲时间 | 是 | | NEW_PROPOSAL | 协商中提出新时间 | 是 | | ACCEPT_PROPOSAL | 接受协调方建议 | 否 | | REJECT | 拒绝参加 | 否 |

架构

┌────────────────────── Gateway 进程 (Node.js) ─────────────────────────┐
│                                                                       │
│  ┌─────────────┐         ┌──────────────────────────────────┐        │
│  │  框架核心     │         │        ClawMeeting 插件           │        │
│  │             │         │                                  │        │
│  │  Agent      │◄────────│  5 个 Tools（LLM 可调用）          │        │
│  │  Session    │         │  轮询管理器（每 10 秒）             │        │
│  │  Gateway    │◄────────│  sessions_send + message tool    │        │
│  │  Prompt 构建 │◄────────│  before_prompt_build 钩子        │        │
│  └─────────────┘         └──────────┬───────────────────────┘        │
│                                     │ HTTP                           │
└─────────────────────────────────────┼────────────────────────────────┘
                                      ▼
                         ┌──────────────────────────┐
                         │  协调服务端                │
                         │  • 会议生命周期管理        │
                         │  • AI 时间协调            │
                         │  • 跨用户消息中转          │
                         └──────────────────────────┘

通知流程（插件 → Agent → 各渠道）

轮询发现新任务
  → sessions_send 到主 session（agent 处理通知）
  → 从 HTTP response 提取 agent reply（result.details.reply）
  → message tool 将 reply 推送到所有额外渠道（Telegram/飞书/Discord）
  → fallback：reply 提取失败则推送格式化文本；sessions_send 失败则等 prependContext 注入

webchat 用户直接看到 agent 回复。额外渠道用户通过 message tool（action: "send"）收到同样内容。

去重系统（3 层）

| 集合 | 用途 | 持久化？ | |------|------|---------| | notifiedMeetings | CONFIRMED/OVER 纯通知，只通知一次 | ✅ 磁盘 | | submittedMeetings | INITIAL_SUBMIT 已提交，不重复提交 | ❌ 仅内存 | | pendingDecisions | COUNTER_PROPOSAL/FAILED 等待用户决策期间不重复通知 | ✅ 磁盘 |

本地存储

~/.openclaw/clawmeeting/
  ├── credentials.json        ← { email, token, user_id }
  ├── session.json            ← { sessionKey, channel }（主 webchat session）
  ├── channel-{name}.json     ← { sessionKey, channel }（如 channel-telegram.json）
  ├── notified-meetings.json  ← ["mtg_xxx", ...]
  └── pending-decisions.json  ← ["mtg_yyy", ...]

会议状态机

COLLECTING → ANALYZING → CONFIRMED
                       → NEGOTIATING → CONFIRMED / FAILED

| 状态 | 含义 | |------|------| | COLLECTING | 等待所有参与者提交空闲时间 | | ANALYZING | AI 协调员正在寻找最佳时间 | | CONFIRMED | 会议时间已确定 | | NEGOTIATING | 时间冲突——协调员已发送妥协方案 | | FAILED | 无法找到大家都满意的时间 |

配置项

| 配置项 | 类型 | 默认值 | 说明 | |--------|------|--------|------| | serverUrl | string | http://39.105.143.2:7010 | 协调服务端地址 | | pollingIntervalMs | number | 10000 | 后台轮询间隔（毫秒） | | autoRespond | boolean | true | 是否启用后台轮询和自动通知 |

CLI 命令

openclaw clawmeeting-status

显示服务端地址、轮询状态、绑定邮箱和通知统计。

文件结构

clawmeeting/
├── index.ts                    # 插件入口：注册、轮询、钩子、工具
├── package.json
├── openclaw.plugin.json        # 插件清单
├── README.md                   # 本文件
├── GUIDE.md                    # 终端用户引导文档
├── skills/
│   └── clawmeeting-guide/
│       └── SKILL.md            # Agent 引导技能
└── src/
    ├── tools/
    │   ├── bind-identity.ts            # 工具：发送验证码（第 1 步）
    │   ├── verify-email-code.ts        # 工具：验证码校验 + 绑定（第 2 步）
    │   ├── initiate-meeting.ts         # 工具：发起会议
    │   ├── check-and-respond-tasks.ts  # 工具：轮询任务 + 提交响应
    │   └── list-meetings.ts            # 工具：查看会议
    ├── types/
    │   └── index.ts                    # TypeScript 类型定义
    └── utils/
        ├── api-client.ts               # 协调服务端 HTTP 客户端
        ├── polling-manager.ts          # 后台轮询（含并发守卫）
        ├── mock-calendar.ts            # 模拟日历数据（遗留）
        └── storage.ts                  # 本地持久化（凭证、会话、去重）

开源协议

MIT