🧶 Charl: Technology That Feels Like Magic

Technically, Charl is a cloud-native orchestration platform for OpenClaw agents. Psychologically, it is the end of "DevOps Anxiety."

"The best interface is no interface. The best configuration is a conversation."

---

🌟 The "End of Friction" Promise

Most AI agents are smart but helpless. They need you to configure environments, manage keys, and restart processes. Charl is different. It's a fleet of self-managing OpenClaw agents that live on the edge.

When you want to change something, you don't edit a config file. You just tell the agent:

You: "We've switched to the new Claude Sonnet model. Please update yourself." OpenClaw: "Consider it done."

And it happens. The agent updates its own state, persists it, and adapts instantly.

---

🏗️ Architecture & Agency

Charl is built on the Cloudflare Agents SDK, leveraging the power of Durable Objects for persistence and stateful autonomy.

Core Components

1. Charl (The Platform): A high-performance Worker that acts as the orchestrator.
2. Registry Agent: The "Phonebook" of the system.
   • Service Discovery: Automatically registers new agents.
   • Real-time State: Uses WebSockets to sync fleet status to the dashboard instantly.
   • Presence: Tracks agent heartbeats (Online/Offline status).
3. OpenClaw Agents (The Workers):
   • AI Brain: Powered by gpt-4o (or compatible models).
   • Tool Use: Can schedule tasks, perform calculations, and browse the web.
   • Sandbox: Optional integration with Moltbot for secure, sandboxed execution of complex code.

🧠 The Hive Mind

graph TD
    classDef platform fill:#f48120,color:#fff,stroke:#333,stroke-width:2px;
    classDef agent fill:#e1e1e1,color:#333,stroke:#333,stroke-width:1px;
    classDef client fill:#f9fafb,color:#333,stroke:#333,stroke-width:1px;

    User("User / Dashboard"):::client
    
    subgraph Cloudflare_Worker ["☁️ Charl Worker"]
        direction TB
        Registry["📖 Registry Agent"]:::platform
        
        subgraph Agents
           A1["🦁 Agent: Research"]:::agent
           A2["🦁 Agent: DevOps"]:::agent
        end
    end
    
    User <-->|WebSocket| Registry
    Registry <-->|JSRPC| A1
    Registry <-->|JSRPC| A2

Loading

---

🚀 Getting Started

1. Zero-Config Deployment

Deploy the entire platform to your Cloudflare account in seconds.

npm run deploy

npm run deploy

2. Configure Secrets (Critical)

After deployment, you must set the following secrets to avoid "WebSocket connection failed" errors:

# 1. Your OpenAI Key (Required)
npx wrangler secret put OPENAI_API_KEY

# 2. Development Mode (Recommended for first run)
# Skips strict Cloudflare Access checks
npx wrangler secret put DEV_MODE
# Value: true

# 3. Gateway Token (Random string)
npx wrangler secret put MOLTBOT_GATEWAY_TOKEN
# Value: any-random-string

3. Access the Dashboard

Visit your deployed worker URL (e.g., https://charl.your-subdomain.workers.dev).

• Create Agents: Spin up new workers with a click.
• Monitor Fleet: See real-time status and heartbeats.
• Broadcast: Send system-wide commands to all agents.

3. Documentation

For detailed guides on architecture, configuration, and API usage, visit the Docs page on your deployed instance at /docs.

---

🛠️ Development

Run the full stack locally:

npm run dev

This starts the Worker, the Client (Vite), and the connection to Cloudflare services locally.

Key Config

• wrangler.jsonc: Cloudflare bindings and configuration.
• .dev.vars: Local secrets (OPENAI_API_KEY, etc.).

---

License

MIT. Because freedom reduces friction.