OpenClaw can be accessed remotely, but it is not recommended to expose the Gateway or Dashboard directly to the public network. OpenClaw is a control plane plus tool execution plane, and once it is accessed without authorization, the risk is much higher than that of ordinary web applications. The most stable route is local-first, remotely using Tailscale, SSH tunnels, or trusted reverse proxies, and enforcing authentication.
Ask yourself three things before remote access
- Who can access this address? Is it just me, or the entire public network?
- What to do after the visit? Just chat, or can you change the configuration, run commands, and operate the browser?
- How to stop loss if something goes wrong? Are there separate accounts, separate machines, logs, and backups?
Security configuration focus
- Gateway should not be bound to a non-loopback address without authentication.
- Firewalls release only necessary sources.
- Dashboard uses HTTPS or trusted tunnels instead of temporary weak passwords.
- Personal and team assistants separate Gateway, separate credentials, and separate OS users.
- Run
openclaw security audit --deepregularly.
If you just want to use it on your phone, give priority to accessing through channels such as Telegram, WhatsApp, and Slack instead of opening the dashboard. The channel layer is re-equipped with allowlist and pairing, and the risk is easier to control.
Official open source address: https://github.com/openclaw/openclaw.
After the remote access is configured, it is recommended to perform a reverse check from the public network: confirm that the access is denied without logging in, the error token will not enter the console, and the failure reason can be seen in the logs. Security is not "open", but "those who can't open can't open it".
