Spin up a fully-deployed expert AI agent for any domain in under 2 minutes.

Built on OpenClaw — an open-source AI agent framework with persistent identity, memory, and multi-channel support.

One command creates everything needed for a public-facing AI expert:

You: "Create a new expert agent for Thai cooking"

Agent Factory asks 10 questions:
  1. Domain → Thai cooking and recipes
  2. Name → Chef AI
  3. ID → chef-ai
  4. Emoji → 👨‍🍳
  5. Personality → Friendly
  6. Languages → Thai + English
  7. Audience → Home cooks
  8. RAG data → /path/to/recipes.csv
  9. Self-expand? → No
  10. Disclaimer → No

Then deploys everything automatically.
→ https://chef-ai.yourdomain.com is live!

User → cloudflared tunnel → FastAPI (port 808x) → OpenClaw Gateway → Claude API
                                                        ↓
                                                   Agent Workspace
                                                   (SOUL + RAG + Skills)

The web app proxies chat requests through the OpenClaw Gateway's OpenAI-compatible API (/v1/chat/completions), keeping the gateway token server-side. The frontend uses Server-Sent Events (SSE) for streaming responses.

• OpenClaw installed and running
• Python 3.10+
• cloudflared tunnel configured
• systemd (Linux user services)
• A Cloudflare-managed domain

Copy the SKILL.md to your Claude Code skills directory:

mkdir -p ~/.claude/skills/agent-factory
cp SKILL.md ~/.claude/skills/agent-factory/SKILL.md

Then copy the full skill to your OpenClaw workspace:

cp -r . ~/.openclaw/workspace/skills/agent-factory/

Now you can use /agent-factory in Claude Code or just say "create a new agent".

Copy the entire directory to your OpenClaw workspace skills:

cp -r . ~/.openclaw/workspace/skills/agent-factory/

The skill triggers on keywords like "create agent", "new bot", "deploy agent", etc.

You can also run the deploy script directly:

# 1. Create a config file
cat > /tmp/my-agent-config.json << 'EOF'
{
  "agent_id": "cooking-expert",
  "agent_name": "Cooking Expert AI",
  "agent_name_local": "",
  "emoji": "👨‍🍳",
  "domain": "Thai cooking and recipes",
  "language": "Thai",
  "language_secondary": "English",
  "vibe": "Friendly",
  "audience": "Home cooks wanting authentic Thai recipes",
  "disclaimer": "",
  "self_expand": false,
  "rag_files": ["/path/to/recipes.csv"]
}
EOF

# 2. Run the deploy script
python3 scripts/deploy.py --config /tmp/my-agent-config.json

The script runs 10 steps automatically:

1. Create agent workspace (~/.openclaw/workspace-{id}/)
2. Create domain skill (~/.openclaw/workspace/skills/{id}/)
3. Register in openclaw.json
4. Restart OpenClaw gateway
5. Create web app (~/{id}-app/)
6. Create venv + install dependencies
7. Create + enable systemd service
8. Configure cloudflared tunnel + DNS
9. Ingest RAG data (if provided)
10. Health check

Thai, English, Japanese, Chinese, Korean, Spanish, French, German, Portuguese, Vietnamese, Indonesian, Malay, Hindi, Arabic, Russian.

The web UI (headers, placeholders, welcome message) is automatically localized.

# Check status
systemctl --user status {agent-id}-app

# View logs
journalctl --user -u {agent-id}-app -f

# Restart
systemctl --user restart {agent-id}-app

# Stop
systemctl --user stop {agent-id}-app

# Add more RAG data
python3 ~/.openclaw/workspace/skills/rag/rag_tool.py ingest /path/to/file \
  --collection {agent-id}

agent-factory/
├── SKILL.md                          # Skill definition (Q&A workflow)
├── scripts/
│   └── deploy.py                     # Deployment pipeline (10 steps)
├── templates/
│   ├── workspace/                    # Agent identity templates
│   │   ├── SOUL.md.tmpl
│   │   ├── AGENTS.md.tmpl
│   │   ├── IDENTITY.md.tmpl
│   │   ├── USER.md.tmpl
│   │   └── MEMORY.md.tmpl
│   ├── skill/
│   │   └── SKILL.md.tmpl            # Domain skill template
│   ├── webapp/
│   │   ├── env.tmpl                  # Environment variables
│   │   ├── server/
│   │   │   ├── app.py.tmpl           # FastAPI app with SSE streaming
│   │   │   ├── chat.py.tmpl          # OpenClaw gateway proxy
│   │   │   ├── requirements.txt
│   │   │   └── __init__.py
│   │   └── frontend/
│   │       ├── index.html.tmpl       # Chat UI
│   │       ├── app.js                # SSE stream handler
│   │       └── style.css             # Responsive styles
│   └── systemd/
│       └── service.tmpl              # Systemd user service
└── references/
    └── usage-guide.md                # Quick reference

All templates use {{PLACEHOLDER}} syntax. The deploy script replaces them with values from the config JSON. You can customize any template to change the agent's behavior, UI, or infrastructure.

By default, agents are assigned ports from 8080-8099. Change PORT_RANGE in deploy.py to adjust.

Change TUNNEL_DOMAIN in deploy.py to use your own domain.

MIT