Writing code, managing finances, running marketing campaigns, generating daily reports, handling DevOps — all you need to do is send a message in Discord.

📌 About Originality — This project was first committed on 2026-02-22 (commit history) and represents the original implementation of the concept "modeling AI multi-agent collaboration after China's imperial court system." We noticed that cft0808/edict (first committed 2026-02-23, approximately 21 hours later) shares striking similarities with this project in framework selection, SOUL.md persona files, install.sh deployment approach, and competitor comparison tables — see Issue #55 for details.

Reposts welcome — please credit the source.

📕 Original Xiaohongshu series: Day 3 of Being an AI Emperor — I'm Already Hooked | Life of a Cyber Emperor: Give Orders Before Bed, AI Cranks Out Code Overnight

Why This Setup?

	ChatGPT & Web UIs	AutoGPT / CrewAI / MetaGPT	AI Court (This Project)
Multi-agent collaboration	❌ Single generalist	✅ Requires Python orchestration	✅ Config files only, zero code
Persistent memory	⚠️ Single shared memory	⚠️ BYO vector database	✅ Each agent has its own workspace + memory files
Tool integrations	⚠️ Limited plugins	⚠️ Build your own	✅ 60+ built-in Skills (GitHub / Notion / Browser / Cron …)
Interface	Web	CLI / Self-hosted UI	✅ Native Discord (works on phone & desktop)
Deployment difficulty	No deployment needed	Docker + coding required	✅ One-line script, up in 5 minutes
24h availability	❌ Manual conversations only	✅	✅ Cron jobs + heartbeat self-checks

The key advantage: it's not a framework — it's a finished product. Run a script, start chatting in Discord by @mentioning your agents.

Architecture

Discord Message
    ↓
Clawdbot Gateway (Node.js daemon)
    ├── Message routing: @mention → match binding → dispatch to agent
    ├── Session isolation: each agent has its own session & workspace
    ├── Auto-threading: large tasks automatically spawn Threads
    └── Cron scheduler: time-triggered agent tasks
         ↓
    ┌──────────────┐  ┌──────────────┐  ┌──────────────┐
    │ 兵部 bingbu  │  │ 户部 hubu    │  │ 礼部 libu    │  ... (extensible)
    │ Power Model  │  │ Power Model  │  │ Fast Model   │
    │ Code Expert  │  │ Finance Pro  │  │ Marketing Pro│
    └──────┬───────┘  └──────┬───────┘  └──────┬───────┘
           │                 │                  │
    ┌──────┴─────────────────┴──────────────────┴──────┐
    │              Skill Layer (60+)                    │
    │  GitHub · Notion · Browser · Cron                │
    │  TTS · Weather · Screenshots · Video …           │
    └──────────────────────────────────────────────────┘

Each agent is a standalone Discord Bot bound to its own AI identity:

Independent memory: Every agent has its own memory/ directory — the more you use it, the better it knows your projects
Independent models: Use power models for heavy lifting, fast models for light work — save money without sacrificing quality
Sandboxed execution: Agent code runs in isolation, no cross-contamination
Identity injection: Clawdbot automatically assembles SOUL.md + IDENTITY.md + workspace files into the system prompt

Quick Start

Step 1: One-Line Deployment (5 minutes)

Grab a free cloud server (ARM 4-core, 24 GB RAM — free tier), SSH in, and run:

bash <(curl -fsSL https://raw.githubusercontent.com/wanikua/boluobobo-ai-court-tutorial/main/install.sh)

The script automatically handles:

✅ System updates + cloud provider firewall configuration
✅ 4 GB Swap (prevents OOM kills)
✅ Node.js 22 + GitHub CLI + Chromium
✅ Global Clawdbot installation
✅ Workspace initialization (SOUL.md / IDENTITY.md / USER.md / clawdbot.json multi-agent template)
✅ Gateway systemd service (auto-starts on boot)

The install script features color-coded output with progress indicators and ✓ success markers for each step.

Step 2: Add Your Keys (10 minutes)

After the script finishes, you just need two things:

LLM API Key → From your LLM provider's dashboard
Discord Bot Token (one per department) → discord.com/developers

# Edit config — fill in API Key and Bot Tokens
nano ~/.clawdbot/clawdbot.json

# Start the court
systemctl --user start clawdbot-gateway

# Verify
systemctl --user status clawdbot-gateway

@mention your Bot in Discord and say something. If it replies, you're live.

Step 3: Full Six Ministries + Automation (15 minutes)

@bingbu Write me a user login API
→ 兵部 bingbu (Power Model): Complete code + architecture suggestions, large tasks auto-spawn Threads

@hubu How much did we spend on APIs this month?
→ 户部 hubu (Power Model): Cost breakdown + optimization recommendations

@libu Write a Xiaohongshu post about AI tool recommendations
→ 礼部 libu (Fast Model): Copy + hashtag suggestions

@everyone Tomorrow's meeting at 3 PM — all departments prepare weekly reports
→ All agents acknowledge individually

Set up automated daily reports:

# Get the Gateway Token
clawdbot gateway token

# Auto-generate daily report at 22:00 Beijing time
clawdbot cron add \
  --name "Daily Report" --agent main \
  --cron "0 22 * * *" --tz "Asia/Shanghai" \
  --message "Generate today's daily report, write to Notion and post to Discord" \
  --session isolated --token <your-token>

Court Architecture — The Six Ministries

Inspired by the Ming Dynasty's six-ministry system, each "department" is an independent AI Agent + Discord Bot:

Ministry	Role	Recommended Model	Typical Use Cases
司礼监 siliijian — Chief Steward	Central coordination	Fast Model	Daily chat, task delegation, automated reporting
兵部 bingbu — Engineering	Software engineering	Power Model	Coding, architecture design, code review, debugging
户部 hubu — Finance	Finance & operations	Power Model	Cost analysis, budgeting, e-commerce operations
礼部 libu — Marketing	Brand & marketing	Fast Model	Copywriting, social media, content strategy
工部 gongbu — DevOps	Infrastructure	Fast Model	DevOps, CI/CD, server management
吏部 libu — Projects	Project management	Fast Model	Startup incubation, task tracking, team coordination
刑部 xingbu — Legal	Legal & compliance	Fast Model	Contract review, intellectual property, compliance checks

💡 Model tiering strategy: Use power models for heavy tasks (coding/analysis), fast models for light tasks (copywriting/management) — saves roughly 5× on costs. You can also plug in economy models for further savings.

Core Capabilities

🤖 Multi-Agent Collaboration

Each department is its own Bot. @mention one and it responds; @everyone triggers all of them. Large tasks automatically spawn Threads to keep channels tidy.

⚠️ To enable Bot-to-Bot interactions (e.g., word chain games, multi-Bot discussions), add "allowBots": true in the channels.discord section of clawdbot.json. Without this, Bots ignore messages from other Bots by default. Each account also needs "groupPolicy": "open", otherwise group messages will be silently dropped.

🧠 Independent Memory System

Each agent has its own workspace and memory/ directory. Project knowledge accumulated through conversations is persisted to files and retained across sessions. The more you use an agent, the better it understands your project.

🛠️ 60+ Built-in Skills

It's not just a chatbot — the built-in toolset covers the entire development lifecycle:

Category	Skills
Development	GitHub (Issues/PRs/CI), Coding Agent
Documentation	Notion (databases/pages/automated reporting)
Information	Browser automation, Web search, Web scraping
Automation	Cron scheduled tasks, Heartbeat self-checks
Media	TTS voice, Screenshots, Video frame extraction
Operations	tmux remote control, Shell command execution
Communication	Discord, Slack, Telegram, WhatsApp, Signal…

⏰ Scheduled Tasks (Cron)

Built-in Cron scheduler lets agents run tasks on autopilot:

Auto-generate daily reports, post to Discord + save to Notion
Weekly summary roll-ups
Scheduled health checks and code backups
Any custom recurring task you can think of

👥 Team Collaboration

Invite friends to your Discord server — everyone can @mention department Bots to issue commands. No interference between users, and results are visible to all.

🔒 Sandbox Isolation

Agents can run inside Docker sandboxes with isolated code execution. Supports configurable isolation levels for networking, filesystem, and environment variables.

Detailed Tutorials

For the basics (server provisioning → installation → configuration → first run) and advanced topics (tmux, GitHub, Notion, Cron, Discord, prompt engineering tips), check out the Xiaohongshu tutorial series.

FAQ

Basics

Q: Do I need to know how to code? No. The one-line install script handles setup, and configuration is just filling in a few keys. All interaction happens through natural language in Discord.

Q: Is the server really free? Cloud providers' free tiers (where available).

Q: How is this different from just using ChatGPT? ChatGPT is a single generalist that forgets everything when you close the tab. This system gives you multiple specialists — each agent has its own domain expertise, persistent memory, and tool permissions. They can automatically commit code to GitHub, write docs to Notion, and run tasks on a schedule.

Q: Can I use other models? Yes. Clawdbot supports various LLM providers, OpenAI, Google Gemini, and others (including economy models). Just change the model config in clawdbot.json. Different departments can use different models.

Q: How much does the API cost per month? Depends on usage intensity. Light usage: $10–15/month, moderate: $20–30/month. Cost-saving tip: use power models for heavy tasks, fast models for light tasks (roughly 5× cheaper), and plug in economy models for simple tasks to save even more.

Technical

Q: @everyone doesn't trigger agent responses? In the Discord Developer Portal, each Bot needs Message Content Intent and Server Members Intent enabled. The Bot's role in the server also needs View Channels permission. Clawdbot treats @everyone as an explicit mention for every Bot — once permissions are set correctly, it will trigger responses.

Q: Agent reports permission errors after enabling sandbox? Sandbox mode set to all runs the agent inside a Docker container with a read-only filesystem, no network access, and no inherited environment variables by default. Fix:

"sandbox": {
  "mode": "all",
  "workspaceAccess": "rw",
  "docker": {
    "network": "bridge",
    "env": { "LLM_API_KEY": "your-LLM-API-KEY" }
  }
}

workspaceAccess: "rw" — Allows the sandbox to read/write the working directory
docker.network: "bridge" — Enables network access
docker.env — Passes in API keys (sandbox doesn't inherit host environment variables)

Q: Will multiple users @mentioning the same agent cause conflicts? No. Clawdbot maintains independent sessions for each user × agent combination. Multiple people can @mention 兵部 bingbu simultaneously — their conversations won't interfere with each other.

Q: Can agents call each other? Yes. Agents can use sessions_spawn to delegate sub-tasks to other agents, or sessions_send to message another agent's session. For example, 司礼监 (Chief Steward) can assign coding tasks to 兵部 bingbu.

Q: How do I create custom Skills? Clawdbot has a built-in Skill Creator tool for creating custom Skills. Each Skill is a directory containing SKILL.md (instructions) + scripts + resources. Place it in the skills/ directory of your workspace, and agents can use it immediately.

Q: How do I connect private models (Ollama, etc.)? Add an OpenAI API-compatible provider in the models.providers section of clawdbot.json, pointing baseUrl to your Ollama address. Local Ollama models have zero API costs.

Q: How do I troubleshoot Gateway startup failures?

# Check detailed logs
journalctl --user -u clawdbot-gateway --since today --no-pager

# Run diagnostics
clawdbot doctor

# Common causes: missing API Key, malformed JSON, invalid Bot Token

🏛️ Join the Court

Xiaohongshu (小红书)	WeChat Official: 菠言菠语	WeChat Group: OpenClaw Emperors

Profile	Follow for tutorials & updates	If QR expired, follow the official account for latest link

⚠️ 免责声明 / Disclaimer

本项目按"原样"提供，不承担任何直接或间接责任。/ This project is provided "as is" without any warranties.

使用前请注意 / Please note:

AI 生成内容仅供参考 / AI-generated content is for reference only
- AI 生成的代码、文案、建议等可能存在错误或不准确之处
- 使用前请自行审核，确认无风险后再实际应用
- Code, suggestions, etc. may contain errors. Please review before using in production.
代码安全 / Code Security
- 自动生成的代码建议在合并前进行 code review
- 涉及财务、安全敏感的操作请务必人工复核
- Review AI-generated code before merging. Human review required for financial/sensitive operations.
API 密钥安全 / API Key Security
- 请妥善保管您的 API 密钥 / Keep your API keys safe
- 不要将包含密钥的配置文件提交到公开仓库 / Don't commit config files with keys to public repos
服务器费用 / Server Costs
- 免费服务器（云服务商等）有一定使用限额 / Free servers have usage limits
- 超出限额后可能产生费用，请留意账单 / Excess usage may incur charges
数据备份 / Data Backup
- 建议定期备份您的工作区和数据 / Regularly backup your workspace
- 本项目不提供任何数据保证 / This project provides no data guarantees

v3.4 | MIT License

📜 This project is licensed under MIT. If you create derivative works or projects inspired by this architecture, please credit the original: boluobobo-ai-court-tutorial by @wanikua

Languages

TypeScript 34.4%

JavaScript 29.5%

Shell 29.2%

Python 2.9%

PowerShell 2.6%

Other 1.4%

README_EN.md Unescape Escape

🏛️ Build Your AI Court in 30 Minutes