OpenClaw is an open-source AI assistant project. Use its official repository and documentation for the current feature set, requirements and supported installation paths.
Separate hosting from model access
Running an assistant on a machine you control does not necessarily mean every part of its work stays on that machine. Review the configured model providers, integrations and tools before supplying private data.
Review the security boundary
The project's security guidance is the appropriate starting point for gateway exposure, authentication and tool access. Assess third-party extensions and keep permissions limited to the tasks you intend to allow.
Begin with a bounded experiment
- Use synthetic files and a test workspace.
- Define which actions require human review.
- Keep important credentials and recovery channels under your control.
- Check logs for unintended disclosure of tokens or personal information.
- Test how to stop the assistant and revoke an integration.
Temporary email may be appropriate for a low-stakes, permitted demonstration, but not as the sole recovery channel for an important agent account or connected service. This introduction does not claim a historical growth ranking or a verified count of exposed installations.