GoldieBench Blog · 1 min read
How To Build An Agentic OS (Dashboard, Memory, Models, Workflows)
How to build an agentic operating system: one dashboard for every AI agent, custom workflows, Obsidian memory, and free local models — the exact architecture behind the Agent OS, benchmarked here.
An agentic OS is one dashboard where all your agents live: Claude, Hermes, Codex, your custom workflows, your memory — everything one click away.
Julian published a full one-hour build course on YouTube; this is the condensed architecture with the model data to back each choice.
The four layers
1.
Dashboard. A local Next.js (or any) app on localhost — local for security: an OS that can edit files must not be an open website.
Start from a template (minutes) or have GPT-5.6/Claude scaffold from scratch (hours).
2.
Agents. Wire in the CLIs you already pay for (Claude, Codex) plus a free local lane through Ollama.
Which local model?
That's literally what our local leaderboard answers — currently Qwable for quality, Agents-A1 for agentic speed.
3.
Memory. Obsidian as the shared vault: every agent reads and writes the same markdown, so Claude knows what Hermes did.
Add automatic capture (meeting/screen notes) and a weekly agent-written review.
4.
Workflows. Audit where your time goes, then build one tab per repeated job: SEO pipeline, video agent, lead gen, image studio.
Rule of thumb from the course: if it's daily and repeatable, it becomes a one-click tab.
Principles that keep it alive
System over models (models swap, the OS stays).
One builder-agent at a time (two agents editing one repo overwrite each other).
Test everything by using it, not by trusting "done".
Keep features hideable — you'll only use a handful daily.
Get the working template
The maintained Agent OS (zip + video tutorial, updated daily) ships inside the AI Profit Boardroom; guides for each subsystem are free at agentos.guide.
FAQ
What is an agentic OS?
A single dashboard that hosts all your AI agents, custom automation workflows, and a shared memory — so instead of hopping between chat apps, everything runs and persists in one place.
Do I need to be a coder to build one?
No. The one-hour course builds it live with GPT-5.6 doing the coding; members typically install the template zip in ~10 minutes and customize by asking their agent for changes.
Should it run locally or on a VPS?
Locally by default — an agent OS can edit files and run commands, so exposing it publicly is a security risk. VPS is viable if you strictly limit its tools.
Which model should power it?
Route: your existing Claude/GPT subscription as the main brain, plus a free local model for sub-agent work — see our local leaderboard for the current best picks with real scores.
Keep reading
Hermes Herald AI Agent: Which Brain To Run It On, With Real GoldieBench Scores
GEO Examples: Four Real AI Overview Wins, The Three Moves Behind Them, And The Models Writing The Answers
Best Generative Engine Optimization Agency: Seven Picks By Region, And How The Models Behind AI Answers Score🌐 Sister-site take: read this on agentos.guide