The Swif Agent Inventory Policy finds AI agents installed on your devices, and reports how each one is set up. These agents can run in the background, act on their own, and run commands on the device, often without IT knowing they're there.

Swif gives each agent a lasting identity, so you can see exactly which AI agent is running on which device, for which employee.

One policy covers every supported agent:

For each user on the device, Swif reports:

For OpenClaw, the policy can also upload its activity, audit, and security logs. See Log Agents below.

This policy only collects information. It doesn't install, change, or stop any agent.

Agents a platform doesn't support are skipped. For example, Grok Bot isn't collected on Linux.

A shorter interval keeps the inventory more current. A longer one uses fewer device resources. The 60-minute default suits most organizations.

When you create or change the policy, the device collects right away once it receives the policy, without waiting for the interval. Devices pick up policy changes within about 20 minutes.

Log Agents turns on log collection for an agent. Currently, OpenClaw supports it: its activity, audit, and security logs are uploaded to Swif. An agent must also be selected in Agents before you can add it here.

Logs appear in the device's Agentic Security view, under Logs.

A policy you create yourself starts with Log Agents empty. Policies created during Agentic Security onboarding already have openclaw in Log Agents.

Watch for an exposed gateway with no authentication. Together, they can let anyone on the network control OpenClaw.

Swif reads OpenClaw's launch settings from the gateway's service settings, not from the running process, because OpenClaw changes how its running processes appear.

Swif never stores secrets or tokens from OpenClaw.

Watch for yolo. It lets Hermes run actions without asking the user first.

On Windows, Swif reports only whether Hermes is installed, its version, and its settings. Background service, running process, and launch setting data aren't available there.

Swif finds Hermes only in its default location: ~/.hermes on macOS and Linux, and %LOCALAPPDATA%\hermes on Windows. If a user moved it, Hermes shows as not installed. To control what Hermes can do, use the Swif Hermes Access Policy.

Watch for always. It lets Grok Bot run commands on the device without asking.

These are the values Grok Bot last recorded on the device. Swif reports them, but doesn't enforce them.

Swif reports whether Grok Bot is running on both macOS and Windows.

For each user on the device, Swif reports:

Swif gives each agent one identity per user, per device. The identity stays the same when the device restarts, and when the agent is updated or reinstalled, so you can follow one agent over time.

Agents are shown by employee, then OS user, then agent. For each agent, you see:

Agent activity across Agentic Security is linked to the agent's identity, including:

So you can trace any activity back to a specific agent, on a specific device, for a specific employee.

You can also create and assign it from a device:

Assign to all devices to find where these agents are installed, with minimal device overhead.

Inventory OpenClaw, flag risky gateway settings, and upload its activity, audit, and security logs.

Check developer devices every 30 minutes for installs, background services, and risky settings, such as an exposed OpenClaw gateway or Hermes started with yolo.

Check whether OpenClaw is flagged as running in dev mode. Dev mode uses a separate state folder, so the settings Swif reports may not apply to it.

A policy you create yourself starts with Log Agents empty. Add openclaw to Log Agents.

This is expected. On Windows, Swif reports that the OpenClaw gateway's background service is installed, but not whether it's running. Running status and process details are available on macOS and Linux only.

Swif finds Hermes only in its default location: ~/.hermes on macOS and Linux, and %LOCALAPPDATA%\hermes on Windows. If the user installed or moved it somewhere else, Swif can't find it.

This is expected. On Windows, Swif reports only whether Hermes is installed, its version, and its settings.

The inventory reports installs separately from running processes. An installed agent may not be running at the time of collection.

It shouldn't. Each agent keeps one identity per user, per device, across restarts, updates, and reinstalls. The reinstall appears in the agent's history.

This is expected. Linux has no built-in code signing, so Swif reports the file hash, but not the signer.

This is expected. Swif detects it on macOS and Linux only.

No. It only reports what's installed. To block agents or commands, use the Swif AI Agent Activity Monitoring Policy (Agent Access) or the Swif AI Agent Command Protection Policy.

It's been replaced by this policy. Use Log Agents to upload OpenClaw logs.

Swif

Agents