🌐 More than a chatbot: the everything of AI
So far, you've seen individual pieces: memory, personas, models, goals, sub-agents. The Operating System is what bring everything together in one place — the "everything of AI." Instead of jumping between apps, you manage your AI life from a single dashboard. It shifts the mindset from "chat app" to "system."
🏛️ Pantheon of personas
O Pantheon shows your personas visually inside the OS — each with its own job, system prompt, and model. Seeing everything in one place makes it easier to choose who uses what and ensures you’re not running the expensive model for a trivial task.
📊 What each persona carries
- Job — what it’s for (research, writing, support).
- System prompt — how it behaves.
- Model — the assigned brain (expensive only for high-impact tasks).
🗄️ Memory vault
O memory vault is a vault that retrieves any email, meeting, or note — and you can connect external sources like NotebookLM. It's the memory from Track 2 elevated into a central OS component, accessible at a glance.
Illustrative diagram · pluggable sources converging into a central vault
📈 Mission Control
No mission control at a glance, you can see available connections, the model in use, memory, AI spend, gains from skills, and usage in real time. Operating without a dashboard is flying blind—especially after what you learned about cost in module 3.5.
💡 Practical tip
Mission control's live metrics are your first line of defense against an out-of-control cron job: if spending spikes, you'll see it right away.
💭 Dreaming Improvements
The OS isn't static: it "dreams up" improvements at night and comes back with proactive suggestions the next day. Combined with heartbeat and memory, the system evolves on its own—it proposes new crons, persona adjustments, and cost optimizations.
✓ What the "dream" produces
- ✓Automation suggestions that repeat your patterns.
- ✓Persona adjustments based on how you use it.
- ✓Token-saving ideas.
✗ What it shouldn't do alone
- ✗Taking irreversible actions without approval (back to 3.1).
- ✗Enable new permissions without asking you to confirm.
- ✗Spending above the defined cap.
💾 GitHub Backup
Your OS is too valuable to die with the hardware. Connect it to GitHub for a daily backup of all of Hermes — personas, memory, configs. Got a new PC or lost your machine? Restore everything. (Caution: as shown in 3.1, this reinforces why secrets must never be in the chat — they would end up in the backup.)
Backup flow (illustrative)
Everything in one hub.
Visual personas.
Live metrics.
Daily GitHub.
🏗️ Building your OS, step by step
The OS doesn't come ready-made — you build it in layers. Here's an order that works well, from the foundation up:
Foundation: soul + memory
Give the agent an identity and let its memory take shape over ~2 weeks.
Connections with lower access
Enable 2-3 integrations in read-only mode first (module 3.1).
Automation + limit
A briefing cron and a spending cap. Monitor mission control.
Backup + Claude Code
Enable GitHub backup and connect Claude Code (module 3.7).
✓ Build well
- ✓Layer by layer, observing each one.
- ✓Backup enabled before relying on the OS.
- ✓Spending cap from the very first cron job.
✗ Common mistakes
- ✗Enable everything on day one without monitoring it.
- ✗Trusting the OS without a configured backup.
- ✗Too many crons before understanding the cost.
💡 Practical tip
Treat the OS like a house: lay the foundation (soul + memory) before the walls (connections), and install the alarm system (backup + cap) before moving in.
📌 Module Summary
Next Module:
3.7 - 🔗 One Brain: connect Hermes and Claude Code into a single brain — the culmination of the 21 concepts.