OpenClaw

What it does

OpenClaw is an open-source AI agent that you run on your own computer or server. You message it through chat apps such as WhatsApp, Telegram, Slack, Signal or Microsoft Teams, and it carries out tasks: sorting and sending emails, managing your calendar, researching on the web, editing files. It remembers context across conversations and can be extended with skills and plugins, for example for email, GitHub or Obsidian. You pick the language model yourself, such as Claude, GPT or a local model.

Who it suits

Technically skilled teams who want an assistant with access to their own systems and handle setup and maintenance (Node.js, configuration files, permissions) themselves.

Data protection for businesses

Software and data stay on your hardware, so you choose the location. The US non-profit foundation runs no hosted service and so offers no data processing agreement; if needed, sign one with your model provider. Training use depends on that provider’s terms – with a local model, data never leaves your network. Telemetry is limited to a daily version check, which you can switch off. There are no paid business plans; an open-source enterprise edition with multi-tenancy is in development.

How we use it

Agent platforms like OpenClaw are exciting for businesses but need a concept for security, data protection and permissions before they go into production. We advise on this and always build agents for clients with as few permissions as possible, approval steps and traceable logs – with the tool that fits the task.

Limits

The agent can operate with far-reaching system permissions, and prompt injection via external content is a real risk. Restrict its rights with access profiles and the sandbox. There is no official support with contractual commitments.