hermes - 💡(How to fix) Fix [Bug] Dashboard consumes 20-50% CPU constantly when idle

Bug Description

Hermes Dashboard consumes abnormally high CPU (20-50%) constantly, even when idle with no active connections or browser sessions. The process shows sustained high CPU usage that does not drop after initialization completes.

Steps to Reproduce

1. Start dashboard on a clean system:

hermes dashboard --host 0.0.0.0 --port 9119 --insecure --no-open

1. Monitor CPU usage after 1-2 minutes (after initialization):

top -p $(pgrep -f "hermes dashboard")

1. Observe sustained high CPU usage (20-50%) with no browser connections

Expected Behavior

Dashboard should be near-idle (<5% CPU) when no active connections or requests are being processed.

Actual Behavior

• Fresh start: 46% CPU
• After 30-60 seconds: 20-25% CPU (does not drop further)
• After 25+ hours uptime: 9-10% CPU (still elevated, accumulated 143+ minutes of CPU time)

CPU usage remains elevated indefinitely, regardless of whether anyone is accessing the dashboard.

Environment

• Hermes Agent: v0.16.0
• Python: 3.12.3
• OS: Linux (Armbian/Debian)
• Hardware: ARM64, 1.8GB RAM
• Dashboard flags: --host 0.0.0.0 --port 9119 --insecure --no-open

Additional Context

Observed patterns:

1. Consistent across restarts — behavior reproducible on fresh dashboard start
2. No correlation with access — CPU high even when no browser sessions active
3. Memory stable — RAM usage normal (~90-110MB), only CPU affected
4. No errors in logs — no visible exceptions or error messages

System impact:

On resource-constrained systems (1-2GB RAM, low-power ARM), this constant CPU drain:

• Increases system load average
• Reduces available CPU for gateway and other services
• Potentially increases power consumption and heat

Possible causes to investigate:

• Polling loop without proper delays
• WebSocket keepalive/heartbeat running too frequently
• Database/session cleanup running continuously instead of periodic
• Reactive UI framework re-rendering without actual state changes
• Hot-reload / file watcher running in production mode

Workaround

Disable dashboard when not actively needed to reclaim CPU resources.

Additional Information

Tested with both fresh install and long-running instance. Behavior is consistent across multiple restarts and different uptime durations.