OpenClaw

OpenClaw is a personal AI assistant you run as a daemon on your own hardware. Rather than opening a chat app, you message it through WhatsApp, Telegram, Slack, Discord, or any of 20+ other channels it connects to. The assistant has persistent memory, access to your files and tools, and can run background tasks while you are away from your desk. Your data stays on your machine because that is where the gateway runs.

The architecture is straightforward. OpenClaw’s gateway runs as a systemd or launchd service on your machine or server. It connects to the messaging channels you configure and routes incoming messages to an AI model of your choice. You supply the model credentials — Claude, OpenAI, or others — and OpenClaw handles the session management, tool execution, memory, and channel routing.

Skills are the extension mechanism. The assistant can install new capabilities through conversation. Ask it to build a skill that queries your calendar, pulls data from an API, or triggers a Claude Code session on your machine, and it writes and registers the skill itself. The community registry at ClawHub provides pre-built skills for common integrations.

Voice support works on macOS, iOS, and Android. A macOS menu bar companion app provides a Live Canvas — a visual workspace the agent can render content into when a chat message is not the right output format.

OpenClaw reached 375k GitHub stars within six months of its November 2025 launch, with sponsors including OpenAI, GitHub, and NVIDIA. It was created by Peter Steinberger (@steipete), the developer behind PSPDFKit. The pace of development is fast, which means frequent releases but also rough edges in newer features.

The security model requires attention. The gateway runs with full host access for your main session by default. Before connecting any channel accessible to people other than yourself, configure dmPolicy='pairing', set up sandboxing for non-main sessions, and run openclaw doctor to surface any misconfigurations.

OpenClaw: Pros & Cons