hermes-performance

Diagnose and reduce Hermes multi-profile RAM bloat using a PowerShell serve watchdog.

Updated Aug 21, 2026
One-click install
npx skills add https://github.com/TylerSimons1127/vibe --skill hermes-performance-tylersimons1127
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: hermes-performance
Source: https://github.com/TylerSimons1127/vibe/tree/main/skills/devops/hermes-performance
Command: npx skills add https://github.com/TylerSimons1127/vibe --skill hermes-performance-tylersimons1127

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve? Running multiple Hermes profiles spawns a separate serve backend per profile, each launching its full MCP server list — roughly 60 node.exe processes consuming ~3.8 GB of RAM on a loaded Windows machine, causing system lag and stutter. ## Core Features & Use Cases - RAM Diagnosis: PowerShell commands to measure top memory consumers, count MCP node processes, and identify which profile serves are live before making changes. - Active-Profile Watchdog: A scheduled-task PowerShell script that kills serve backends for inactive profiles every 30 seconds while protecting the active profile, the default gateway, and any profile with a live session lease. - Config-Based MCP Scoping: An alternative fix that adds an empty mcp_servers: {} block to inactive profiles so they spawn zero node processes without killing anything. - Use Case: A user with 7 Hermes profiles notices their laptop lagging on launch; after setting the active profile and running the watchdog, free RAM rises from 3.4 GB to 8.5 GB while active sessions stay untouched. ## Quick Start Ask the agent to diagnose Hermes RAM usage on this Windows machine and set up the profile watchdog with 'vibe' as the active profile.

Frequently Asked Questions about hermes-performance

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I reduce RAM usage from multiple Hermes profiles?▼

Run the hermes-profile-watchdog.ps1 script, which kills serve backends for inactive profiles every 30 seconds while protecting the active profile and default gateway. Alternatively, add an empty mcp_servers: {} block to inactive profiles' config.yaml so they spawn zero node processes.

Why does Hermes use so much memory on Windows?▼

Hermes launches a serve backend per profile, and each backend starts that profile's entire mcp_servers list. With 7 profiles and ~15 MCP servers each, that is roughly 60 node.exe processes consuming about 3.8 GB of RAM.

Is it safe to kill Hermes profile serve processes?▼

Killing a --profile X serve process is safe because the desktop does not auto-respawn it and state.db persists all session data. Never kill the bare default serves, which are the machine gateway the desktop app depends on.

How does the watchdog avoid killing a profile mid-session?▼

The watchdog reads each profile's state.db and checks the session_turn_leases table for non-expired rows. Any profile with a live lease is protected from being killed, even if it is not the currently active profile.

Why does Register-ScheduledTask fail with Access is denied?▼

Register-ScheduledTask requires an elevated PowerShell session on Windows. Launch the installer with Start-Process powershell.exe -Verb RunAs, which triggers a UAC prompt the user must approve.