StartClaw

OpenClaw is a conversation-first AI agent that runs on your own devices or servers, connecting to messaging platforms for chatting, performing tasks like file creation or summarization, web searches, and using large language models.

OpenClaw installation requires a Telegram account or similar, terminal comfort, and Node.js on macOS, Linux/WSL2, or servers like Raspberry Pi or Mac Mini. The setup process involves model and authentication setup, workspace bootstrap, gateway settings, provider selection, and daemon installation. For quick deployment, one-click options are available via platforms like Railway for cloud setup, or local install with guided prompts. Testing can be done using the Control UI browser dashboard or TUI, and API keys like Brave Web Search can be added via chat for enhanced capabilities.

To troubleshoot and diagnose issues, run specific OpenClaw commands for quick checks. The openclaw status command shows gateway and agent health and a config snapshot. openclaw models status verifies provider authentication and model availability. The openclaw doctor command validates and repairs configuration and state issues. Other useful commands include openclaw status --all, openclaw logs --follow, and openclaw gateway status.

OpenClaw supports various messaging apps, including WhatsApp, Telegram, Slack, Discord, Mattermost (with a plugin), Google Chat, Signal, iMessage, and WebChat, with some supporting voice and Canvas features. For hardware, it is advisable to start on secondary devices such as a Mac Mini, Raspberry Pi, an old laptop, or a cloud server before installing on a main computer to build trust.

Key features of OpenClaw include persistent memory, which allows it to learn user details like name, timezone, preferences, and patterns over time. It offers autonomy, monitoring inboxes, sending daily briefings, organizing files across drives and cloud services, and handling rate limits. Customization is also a feature, with an adjustable personality, support for slash commands for new bots, and tool integration. For model recommendations, it is suggested to use large models for tools and to enable sandboxing or allowlists for small models.

OpenClaw can run either fully locally or cloud-connected, for example, via Telegram.

OpenClaw was formerly known as Clawbot, ClawdBot, or Maltbot; check the installation documentation for details on how to get updates.