Support 24/7
Client areaGet started
Shared hostingManaged WordPressReseller hostingOpenClaw hostingVPS hostingDedicated serversManaged serversAbout usHow we compareRegionsKnowledge baseSupportDomainsSSLBlogClient area
OpenClaw hosting
Knowledge base / OpenClaw hosting

Run the security audit and restore a backup

What `openclaw security audit` checks, which findings matter on a managed container, and how to roll the workspace back after an agent does the wrong thing.

Last updated: September 16, 2026

An agent that can act for you can also act wrongly for you. Two habits cover most of it: run the audit after any configuration change, and know how to put the workspace back.

  1. 1Run the auditopenclaw security audit reports what is exposed.
  2. 2Keep the agent currentA self-hosted agent is a normal software supply chain: the version you installed is the version you are running.
  3. 3Restore the workspaceMemory, files and configuration live in ~/.openclaw/workspace.
  4. 4Re-run the auditThe restored configuration has not been checked yet.
Audit, update, restore — then audit again. A restore brings back the configuration that existed before, including whatever was wrong with it.

The audit

openclaw security audit

It compares your configuration against the secure defaults and reports drift, with a severity per finding and an auto-fix where one exists safely. Run it after you connect a channel, add a plugin, or change anything under gateway.auth.

The findings worth stopping for:

  • The gateway is bound beyond loopback without auth. A regular install binds to loopback on purpose. gateway.auth.mode: "none" is for private ingress only — never on anything the internet can reach.
  • A channel with an open DM policy. dmPolicy: "open" means anyone who finds the bot is talking to your agent.
  • A plugin outside the official registry. Plugins run with real permissions — control-plane tools, node execution via system.run, dynamic skills. Install from ClawHub, and read what a plugin asks for before you approve it.
  • One gateway, several trust boundaries. A gateway is meant for one operator or a group who trust each other. If two people should not read each other’s conversations, they need two gateways, not one with careful settings.

Keep the agent current

Self-hosted agents are a normal software supply chain: the version you installed is the version that is vulnerable six months later. Updates matter more here than on a static site, because the thing being exploited can act on your accounts.

On inSave containers the platform keeps OpenClaw and the Node.js runtime updated, so the security patches arrive without you tracking releases. What stays yours: the channel allowlists, the plugins you install and the provider keys.

Restore the workspace

The agent’s memory, files and configuration live in the workspace — ~/.openclaw/workspace by default. That is what you restore when an agent deletes something, follows a bad instruction, or a plugin turns out to be a mistake.

The order that avoids making it worse:

  1. Stop the agent so it cannot act while you work.
  2. Restore the workspace from the backup taken before the change — files and memory together, not one without the other.
  3. Rotate anything that might have leaked in the meantime: device tokens (device.token.rotate), the provider key, and any channel token that was in a message.
  4. Start the agent and send one test message before you tell anyone it is fixed.

Point 3 is the one people skip: if the reason you are restoring is that something got in, the credentials it saw are still valid until you change them.

After the restore

Re-run the audit. A restore brings back the configuration that existed before — including, if the backup is old enough, the setting that caused the problem.

openclaw security audit
openclaw doctor --fix