Checking host health (operator)

For whoever runs the box. Casual users don’t need to worry about system metrics.

When you run a Lowkey box, you want a simple answer to one question: Is my box OK? Because Lowkey runs as a self-contained appliance, you shouldn’t have to SSH into a terminal or run diagnostic commands just to check if your agents have enough breathing room.

Where this lives

Open Settings → System in the web app. This dashboard polls the daemon every five seconds to show you real-time metrics on memory, disk, CPU load, and network health.

Memory and the systemd safety cap

Lowkey shows your total host memory usage alongside the daemon’s internal swap space. However, as an operator, the most critical markers are the systemd memory caps:

Soft cap (14 GB): The threshold where systemd begins warning of high resource usage.
Hard cap (18 GB): The point at which the daemon faces memory restrictions.

If your agents are working through massive codebases or running heavy concurrent tasks, you might see the memory indicator bar slide from emerald green into amber or red. Watching this line tells you when you’re approaching an Out-of-Memory (OOM) threshold, giving you time to prune active sessions before the daemon terminates.

Disk space and data footprint

The Disk card displays the total capacity of the volume backing your data. Below the system-wide disk bar, you will see the exact size of your .lowkey data directory.

Because calculating the size of thousands of session artifacts and logs can be slow, the daemon measures this in the background with a 60-second cache. It uses a safety timeout of four seconds so that a massive data directory will never freeze the settings page. If you notice your .lowkey data footprint ballooning, it is usually a sign that you should delete old, unused conversations.

Load, daemon, and live work

The dashboard displays three core performance areas:

Load: The standard 1-minute, 5-minute, and 15-minute CPU load averages. If a load number exceeds your CPU core count (colored red in the UI), it means your agents are waiting for CPU cycles.
Daemon: Details about the running process, including its uptime, PID, Resident Set Size (RSS), and active Javascript heap size.
Live work: A real-time count of active sessions, turns, agent runs, command runs, and automation runs. If your box feels sluggish, you can check this card to see exactly how many tasks are running.

Provider reachability and releases

Your box relies on external services to drive its intelligence and host the web interface:

Providers: Shows which AI providers are reachable, and whether their API keys are valid (authed or needs login). Note: Certain providers like Gemini do not expose an authentication check, so they will display as unknown.
Reachability: Probes loopback (http://127.0.0.1:3000), the daemon itself, and your configured public URL (set via LOWKEY_PUBLIC_URL). It flags any endpoint that times out or returns HTTP errors.
Release status: Compares the running build commit SHA against the main branch tip. If they diverge, a drift warning appears, indicating that the daemon is either behind the latest code or running an uncommitted custom build.