Published: 2026-07-31

Hermes "agent OS": the brain, mission control and memory-vault layers

Chapters / key moments (click to jump — plays here on the page)

Julian Goldie tours what he calls an "agent operating system" built around Hermes, organised in three layers: the brain (Hermes plus swappable models, including free ones via OpenRouter or a local engine), mission control (one dashboard with chat, voice, agent teams and a workspace, so you stop driving from the terminal), and a memory vault (a plain-text Obsidian folder the agents read before working and write to afterwards). His argument is that the memory layer is the one people skip and the reason agents feel generic — models become swappable parts once context, buttons and history live outside them. The video is a walkthrough of his own build, distributed as a zip to his paid community, and it shows no commands or configuration files.

What this video is

A product tour with a substantial paid-community pitch — testimonials, pricing and community access take up a large share of the runtime. The three-layer framing is a useful mental model, but there are no reproducible steps here. For hands-on setup, see our Hermes setup guide, Hermes memory guide, Hermes web dashboard and best free models for Hermes.

Watch on YouTube → · ← Back to News

What you can actually set up from this

Extracted from the video's own transcript — the specifics the original summary left out.

Reproducible steps

Layer 1: the brain
Hermes plus the models it thinks with. The speaker says free models can be used, naming an OpenRouter-style router (transcribed "Omniroot" and "n router") and local models, and says you can plug in your own CLI so existing subscriptions are reused at no extra cost.
Layer 2: mission control
A single dashboard with a button per agent job: chat with profiles for different jobs, voice (talk instead of type, e.g. ask it to open Obsidian), agent teams (he shows Paperclip with sub-agents managing each other), a kanban board where you assign a task and agents triage and complete it, and a workspace where everything each agent builds is saved. He says to build the dashboard by asking Claude or Hermes to build it for you.
Layer 3: the memory vault
A folder of plain text notes, kept in Obsidian (free and open source), that agents read before working and write to after finishing. Contents he lists: who you are, decision logs, daily logs, information about clients, agents and projects. The agents organise the vault, not the user. A graph view shows the notes as a "memory galaxy".
Why three layers
A model change then becomes a drop-in swap that inherits memory, dashboard and running jobs; he says he plugged in a new model the same day in about two minutes.

Gotchas

Promotional avatar video: the finished Agent OS is a zip sold in his paid community, and no commands or config are shown, only the architecture.
Scheduled jobs (his Hermes Oracle news radar that publishes to WordPress) run through the kanban and Paperclip setup rather than needing you to watch.
Claims about token cost being avoidable rest on free models and local engines whose quality is not shown.