OpenClaw operates on a hub-and-spoke model. At the center of this system is the Gateway, which acts as a central control plane. It's responsible for managing sessions, handling security policies, and directing the flow of information between different components.

This design decouples the user-facing channels, like Telegram or Slack, from the AI models that provide reasoning, like Claude 3.5 Sonnet. The Gateway orchestrates everything, allowing you to swap out components without re-architecting the entire system. This follows a 'local-first' philosophy: your data and configurations reside with you, even when the agent uses cloud-based APIs for intelligence.

We'll use Replit's agentic development environment to get started. The Replit Agent can automate the tedious parts of project setup. After creating a new blank Repl, you don't need to manually install dependencies. Instead, you can instruct the agent to do it for you.

The agent will handle the entire toolchain installation. Once it's finished, you can initialize your gateway with a single command: openclaw onboard. This command walks you through the initial configuration, including setting up your API keys from providers like Anthropic or OpenAI. To keep these keys secure, OpenClaw integrates directly with Replit Secrets rather than using plaintext files.

With your gateway configured, the next step is to start it. For a persistent agent, you'll want to run it as a headless process. This allows the gateway to operate continuously in the background of your Repl, even if you close the terminal window or browser tab.

Once the gateway is running, you can check its status at any time to ensure it's operational. The health check command provides a quick summary of the gateway's state and connectivity.

Let's check your understanding of the setup process.

Now that your gateway is configured and running, you're ready to connect it to different channels and tools.