Let your AI summon minions! / 让你的 AI 召唤小弟！
An open-source OpenClaw skill that lets your AI summon other AI bots into group chats for multi-agent discussions.
一个开源的 OpenClaw 技能，让你的 AI 在群聊中召唤其他 AI 机器人进行多智能体讨论。
This is an OpenClaw skill that enables multi-agent discussions in Telegram groups. Your main AI (OpenClaw) can call other AI models (DeepSeek, Gemini, etc.) to join the conversation — each with its own persona, speaking through its own Telegram bot.
这是一个 OpenClaw 技能，可以在 Telegram 群里实现多 AI 讨论。你的主 AI（OpenClaw）可以召唤其他 AI 模型（DeepSeek、Gemini 等）加入对话——每个都有自己的人设，通过各自的 Telegram bot 发言。
No need for multiple OpenClaw instances. No Docker. No complex setup.
不需要部署多个 OpenClaw，不需要 Docker，不需要复杂的环境配置。
Normally, if you want two AIs to discuss a problem, you have to copy-paste messages between them like a human messenger. That's tedious, slow, and you lose context along the way.
通常情况下，如果你想让两个 AI 讨论一个问题，你得像信使一样在它们之间来回复制粘贴消息。这很累、很慢，而且上下文会丢失。
With this skill:
用了这个技能之后：
- Multiple AI bots discuss directly in a group chat / 多个 AI 机器人直接在群聊里讨论
- They challenge each other, catch errors, and converge on answers / 它们会互相质疑、纠错，最终收敛到一致的答案
- The entire conversation is visible to you in real time / 整个对话过程对你完全可视化
- You can interrupt, redirect, or ask follow-up questions anytime / 你可以随时打断、改变方向或追问
- Add as many bots as you want / 可以添加任意数量的 bot
You (in Telegram group)
│
▼
OpenClaw (main AI, e.g. 小龙虾)
│ reads SKILL.md → runs deepseek_speak.py
│ loads subsoul.md (persona identity)
│ calls DeepSeek API
│ sends response via helper bot
▼
Helper Bot (e.g. 批判派🔍) speaks in the same group
│
▼
OpenClaw reads the response, replies, calls helper again...
│
▼
Multi-round discussion until consensus or user stops it
The key insight: OpenClaw orchestrates everything. The helper bot is just a "mouth" — it doesn't need to run independently. This avoids the Telegram limitation where bots can't see each other's messages.
核心思路：OpenClaw 负责指挥一切。辅助 bot 只是一个"嘴"——它不需要独立运行。这样就绕开了 Telegram 中 bot 无法互相看到消息的限制。
- A running OpenClaw instance connected to Telegram
- A second Telegram bot token (create one via @BotFather)
- A DeepSeek API key (cheap, ~$0.001/request)
cp -r call-agents-help /path/to/your/openclaw/skills/Add this to your openclaw.json under skills.entries:
在 openclaw.json 的 skills.entries 中添加：
{
"skills": {
"entries": {
"call-agents-help": {
"env": {
"DEEPSEEK_API_KEY": "your-deepseek-api-key",
"DEEPSEEK_BOT_TOKEN": "your-telegram-bot-token"
}
}
}
}
}kill -9 $(pgrep -f openclaw-gateway)
sleep 2
openclaw gatewayAdd your main OpenClaw bot and the helper bot to the same Telegram group.
把你的 OpenClaw 主 bot 和辅助 bot 拉进同一个 Telegram 群。
Send a message to your main AI in the group:
在群里跟你的主 AI 说：
叫帮手来讨论一下：AI 会不会在五年内取代程序员？
Your main AI will share its opinion first, then summon the helper bot to respond. They'll go back and forth until they reach a conclusion (max 10 rounds).
你的主 AI 会先发表观点，然后召唤辅助 bot 回应。它们会来回讨论直到达成结论（最多 10 轮）。
call-agents-help/
├── README.md ← You're here / 你在这里
├── SKILL.md ← Skill definition (OpenClaw reads this) / 技能定义
├── subsoul.md ← Helper bot persona / 辅助 bot 的人设档案
└── scripts/
└── deepseek_speak.py ← One-shot script (auto-loads subsoul.md) / 一次性脚本
Edit subsoul.md to change the helper bot's personality. No code changes needed.
编辑 subsoul.md 即可修改辅助 bot 的性格，不需要改代码。
The default persona is "批判派🔍" — a calm, analytical thinker who looks at problems from different angles. You can make it anything: a creative brainstormer, a devil's advocate, a domain expert, etc.
默认人设是「批判派🔍」——一个冷静的分析型思考者，擅长从不同角度看问题。你可以改成任何角色：创意派、唱反调的、领域专家等。
Want 3, 4, or 5 AIs discussing at once? Duplicate this skill with different personas:
想要 3 个、4 个甚至 5 个 AI 同时讨论？复制这个 skill 并改用不同的人设：
cp -r call-agents-help call-agents-creative
# Edit call-agents-creative/subsoul.md with a new persona
# Edit call-agents-creative/SKILL.md to update the name and triggers
# Add a new bot token in openclaw.jsonYou can even swap DeepSeek for other OpenAI-compatible APIs (Gemini, Claude, local models) by modifying the API endpoint in deepseek_speak.py.
你甚至可以把 DeepSeek 换成其他 OpenAI 兼容 API（Gemini、Claude、本地模型），只需修改 deepseek_speak.py 中的 API 端点。
-
Check BotFather privacy settings / 检查 BotFather 隐私设置
Go to @BotFather →
/setprivacy→ select your helper bot →DisableNote: This skill doesn't actually need the helper bot to read messages (OpenClaw handles that), but it's good practice.
-
Check environment variables / 检查环境变量
# Test the script directly / 直接测试脚本 DEEPSEEK_API_KEY="your-key" DEEPSEEK_BOT_TOKEN="your-token" \ python3 scripts/deepseek_speak.py \ --chat-id "your-group-chat-id" \ --topic "test topic" \ --no-telegram
If
--no-telegramworks but regular mode doesn't, the issue is with the bot token or chat ID. -
Check the chat ID / 检查群组 chat ID
Group chat IDs are negative numbers (e.g.,
-5176267683). You can find yours by sending a message in the group and checking the OpenClaw logs.
-
Check SKILL.md is in the right place / 确认 SKILL.md 位置正确
The skill folder should be inside your OpenClaw skills directory (usually
~/path-to-openclaw/skills/call-agents-help/). -
Restart the gateway / 重启网关
kill -9 $(pgrep -f openclaw-gateway); sleep 2; openclaw gateway
-
Try explicit trigger words / 尝试明确的触发词
Use phrases like "叫帮手", "让批判派来", "召唤助手" — these are defined in SKILL.md.
- 401 Unauthorized: API key is wrong or expired / API key 错误或过期
- 429 Rate Limited: Too many requests, wait a moment / 请求太频繁，稍等一下
- 5xx Server Error: DeepSeek is having issues, the script will auto-retry once / DeepSeek 服务端问题，脚本会自动重试一次
DEEPSEEK_API_KEY="your-key" \
python3 scripts/deepseek_speak.py \
--chat-id "dummy" \
--topic "人工智能的未来" \
--context "有人说AI会取代所有工作" \
--no-telegramThis will print the response to stdout without sending to Telegram. Useful for testing persona tweaks.
这会把回复打印到终端而不发送到 Telegram，方便测试人设调整。
- Telegram bots cannot see each other's messages — this is a platform limitation, which is why OpenClaw orchestrates everything / Telegram bot 之间无法互相看到消息——这是平台限制，所以由 OpenClaw 统一调度
- Context is managed by OpenClaw per-call, not persisted across sessions / 上下文由 OpenClaw 每次调用时传入，不跨会话保存
- Default max 10 discussion rounds to avoid runaway conversations / 默认最多 10 轮讨论，避免无限对话
This is v1 — it works, but there's room to grow. Upcoming improvements may include:
这是 v1 版本——能用，但还有优化空间。后续可能的改进方向：
- Smarter context management (auto-summarize long discussions) / 更智能的上下文管理（自动总结长讨论）
- Support for more AI providers out of the box / 开箱支持更多 AI 提供商
- Multi-platform support (Discord, Slack, etc.) / 多平台支持（Discord、Slack 等）
- Preset persona packs / 预设人设包
Contributions welcome! Feel free to fork this repo and build your own optimized version. If you come up with something cool, open a PR or share it with the community.
欢迎贡献！ 随时 fork 这个仓库，做你自己的优化版本。如果你搞出了什么好玩的，欢迎提 PR 或者分享给社区。
Having trouble deploying or using this skill? Join our WeChat group for help and discussion:
部署或使用中遇到问题？加微信群交流：
MIT