GoldieBench Blog · 9 min read
Julian Goldie Agent OS GitHub: Where The Agent OS Really Lives, And The Models Inside It
Julian Goldie Agent OS GitHub answer: no public repo. What the Agent OS is, how members install it, and the real GoldieBench scores of the models it runs.
The Julian Goldie Agent OS GitHub repo does not exist, because my Agent OS is not on public GitHub and it is only available inside the AI Profit Boardroom as an installable download with step-by-step guides.
I built the Agent OS myself, and I run my own business on it every day.
Members get it as a dated zip pack with the dashboard, install docs, a video walkthrough, a 30-day roadmap and updates.
In this post I will explain what it is, what is inside, how members install it, why it is not a public repo and how to avoid fake copies.
After that, I will show you the AI models the Agent OS runs and how each one scores on this leaderboard.
Is the Julian Goldie Agent OS on GitHub?
No, it is not on GitHub, and I have not published its source code anywhere public.
The only official place to download it is the Agent OS section inside the AI Profit Boardroom classroom.
If a GitHub page calls itself "the Julian Goldie Agent OS", it is not from me.
What I do share for free is the method, and my AI Agent OS build guide shows the exact prompts I used to build mine.
What the Agent OS is
The Agent OS is my AI agent operating system, which means one dashboard where every AI agent lives, shares one memory and picks up work from one task board.
I built it in mid-2025, when I moved my focus from SEO tactics to AI agents that do the SEO for me.
It runs on macOS and Linux and through a web browser, and I open it from my phone through a private Tailscale network.
The problem it solves is scattered AI, where you have Claude in one tab, Hermes in a terminal and your notes somewhere else, and none of them remember you.
The Agent OS swaps that for three pieces that work together.
| Piece | What it does |
|---|---|
| One dashboard | Every agent sits on one screen with a chat panel, a file workspace and a control room. |
| One shared memory | An Obsidian vault that every agent reads before it works and writes to after it finishes. |
| One task board | A Kanban board where you drop task cards and agents pick them up in parallel. |
What is inside the Agent OS
| Feature | What it does |
|---|---|
| Mission Control | It shows each agent's live status, latency, version and the model it is using. |
| Unified chat | It gives every agent the same chat box, so you switch agents with one click. |
| Hermes desk | One Hermes agent covers Chat, Apollo voice, Oracle, Astros, Studio, Sessions, Outreach, Mixture, Workspace, MCPs, Manage, Control Room and Goal Mode. |
| Voice | You say "hey Hermes" and get a spoken answer in about five seconds, saved to the chat thread. |
| Workspace | Every HTML app, image, video and audio file an agent builds previews inline. |
| Kanban board | Agents pick up cards, work in parallel and write a summary of the job into the vault. |
| Free Claude Code | It runs Claude Code tooling through OpenRouter, so routine jobs cost very little. |
| Studio and research | It makes images, voice and video, and pulls NotebookLM assets into the dashboard. |
| Mobile access | Tailscale opens the whole dashboard on your phone without exposing your computer to the internet. |
My vault passed 1,200 memories in its first month, and my Hermes desk has saved over 1,000 sessions.
That memory is why the system gets more useful every week.
How I use the Agent OS in a normal day
In the morning, Mission Control shows me what every agent did while I was away.
Oracle, a named Hermes agent, has already checked my competitors and the news and written a short report into the vault.
Then I drop a batch of cards on the Kanban board, such as "draft the outline", "check indexing" and "pull this week's numbers".
I go to the gym, the agents work through the cards in parallel, and I come back to finished work in Done with notes explaining what happened.
My SEO content publishes across five websites through Claude Code, with schema and indexing handled in the same run.
During the day, I ask Hermes quick questions out loud and get spoken answers that land in the same chat thread.
When I travel, I open the same dashboard on my phone, with the same cards, chats and memory.
The rule that makes it work is manual, then assisted, then automated.
You do a task by hand with the agent until you would sign your name to the output, then you save it as a workflow, and only then do you hand it to an agent.
People who skip straight to full automation get poor results and blame the AI, when the real problem was a missing process.
How members get and install the Agent OS
The install is built for people who do not code.
| Step | What you do |
|---|---|
| 1 | Join the AI Profit Boardroom and check the address is skool.com/ai-profit-lab-7462. |
| 2 | Download the dated Agent OS pack from the classroom and unzip it. |
| 3 | Read the README and the disclaimer, because agents can touch your files. |
| 4 | Install Node.js if you do not have it, then double-click Start Agent OS on a Mac. |
| 5 | Or open Claude Code or Codex in the folder and ask it to follow the Setup With AI playbook. |
| 6 | Run Check My Setup, which shows which features are ready and never prints your keys. |
| 7 | Connect your Obsidian vault, add one agent at a time and follow the 30-day roadmap. |
The Setup With AI playbook tells the AI to stop and ask before anything that needs a paid account or an API key.
Updates are just as simple, because Update Agent OS backs up your old code and never touches your config, keys or vault.
Windows and Linux users get an Update With AI guide that Claude, Codex or Cursor can follow.
Why the Agent OS is not a public repo
It is the members' version.
I share it inside the AI Profit Boardroom together with the docs, videos, updates and four coaching calls a week that help people get it running.
I have not published the source on GitHub, and the free guides on agentos.guide teach the method instead.
Beware of fake Agent OS repos
When I searched GitHub on 9 October 2026, I found a few community repos that mention my name or the Agent OS.
None of them is published by me, and none of them is the members' pack.
Do not run install scripts from any repo that claims to be my Agent OS, and never paste your API keys into one.
Free open-source alternatives on GitHub
If you want something free from GitHub today, these are the three I have tested and written about.
| Project | GitHub repo | What it is | Stars on 9 Oct 2026 |
|---|---|---|---|
| Hermes Agent | NousResearch/hermes-agent | The open-source agent from Nous Research, under the MIT licence. | More than 250,000 |
| Herald OS | iamlukethedev/Herald-OS | An alpha operating system with Hermes Agent as its interface, under the MIT licence. | 329 |
| StarNet | androoAGI/starnet | A local-first pixel-art agent station, under the MIT licence. | 1,272 |
Herald OS is an independent project by Luke The Dev, and my Herald OS breakdown covers which brains to connect to it.
StarNet is the most fun agent add-on I have tested, and I built it into my own Agent OS as a tab, which my StarNet GitHub post explains.
Does the Agent OS ship its own model?
No, the Agent OS does not ship its own AI model.
It is the dashboard, memory and task board, and the brains plug into it.
That is why the models matter, because they decide how good the work is and what it costs you.
To compare them fairly, I use GoldieBench, where every model gets the same one-shot prompts to build games, pages, simulations and visuals, scored from 0 to 10.
A GoldieBench score measures raw building ability, not how well a model behaves inside the Agent OS.
The models the Agent OS runs, with real GoldieBench scores
These are the engines wired into my own Agent OS build, matched to their live scores on this board.
| Where it runs in the Agent OS | Model on GoldieBench | Avg score | Tasks scored |
|---|---|---|---|
| The Mixture tab, Hermes Mixture of Agents | Hermes MoA | 8.17 | 47 |
| Claude Code and a Hermes profile, on my Claude subscription | Claude Opus 5.5 | 7.57 | 50 |
| The Codex engine | GPT-5.6 Sol | 8.16 | 50 |
| A Kimi Hermes profile | Kimi K3 | 7.89 | 50 |
| A GLM Hermes profile | GLM-5.2 | 7.77 | 47 |
| The Muse Code tab | Muse Spark 1.2 | 7.55 | 50 |
| The Grok tab and a Hermes profile | Grok 4.7 | 7.15 | 20 |
| The Hy3 coder tab | Hy3 | 6.76 (provisional) | 7 |
| The MiMo Code tab | MiMo-V2.6 Pro | 6.35 | 50 |
Hermes MoA is the one model on this table that is an Agent OS feature in its own right.
It runs from the Mixture tab, where a panel of models drafts each build and an aggregator merges the best of every draft.
It scored 8.17 across 47 tasks, which puts it right next to the strongest single models on the board.
Claude Code can also run Claude Opus 5, which scores 8.27 across all 50 tasks and is the strongest single model I would put in the builder seat.
The Claude Opus 5 vs GPT-5.6 Sol head-to-head is worth reading if you are choosing between the two biggest names.
Models the Agent OS runs that are not scored yet
I would rather leave a gap than invent a number, so here are the honest gaps.
The DeepSeek Coder tab runs DeepSeek V4 Flash and DeepSeek V4 Pro, and both pages are on the board but not yet scored.
I also run GLM 5.3 through Hermes, and GLM 5.3 is not on the board yet, so the closest benched model is GLM-5.2 at 7.77.
The Agent OS has a Gemini agent too, and the newest Gemini on the board is Gemini 3.6 Flash at 7.08.
The free local engines inside the Agent OS
The Agent OS has a Local engine for free, offline work, and I have swapped its model several times.
| Local model | Avg score | Tasks scored | Note |
|---|---|---|---|
| Qwable 5 27B Coder | 7.14 | 41 | It runs locally through MLX only. |
| Gemma-4 12B Coder | 4.25 (provisional) | 6 | It is wired into the Local chat, the Local Hermes engine and the Agent Kanban. |
| Qwythos 9B | 2.98 | 42 | It is a small local model with a low build score. |
The gap between local and cloud is the real story.
The best local builder scores 7.14, while the cloud leaders sit above 8.
That is why I keep the local engine for private grunt work and send the heavy builds to stronger models.
The full ranking is on the local models board.
Which model I would plug in for each Agent OS job
| Agent OS job | Model I would try first | Why |
|---|---|---|
| Building apps and pages in Claude Code | Claude Opus 5 (8.27) | It is the strongest single model on the board for one-shot builds. |
| Long goals in Codex Goal Mode | GPT-5.6 Sol (8.16) | It sits a hair behind Opus 5 across all 50 tasks. |
| Hard builds where quality matters most | Hermes MoA (8.17) | The Mixture tab combines several models into one answer. |
| Reading long documents and vault notes | Kimi K3 (7.89) | It has a 1M-token context window. |
| Private, offline grunt work | Qwable 5 27B Coder (7.14) | It is free and runs on your own machine. |
If you already pay for Claude or ChatGPT, connect those first, because the Agent OS plugs straight into the CLIs you already have.
My verdict
The Agent OS is not on GitHub, so the only official download is inside the Boardroom.
Its value is the shared memory and the task board, and the brains are swappable parts you can upgrade whenever a better model lands.
On this board, the strongest brains you can plug into it are Claude Opus 5 at 8.27, Hermes MoA at 8.17 and GPT-5.6 Sol at 8.16.
Also On Our Network
- Agent Operatorsthe step-by-step Agent OS install walkthrough
- AI Income Deskhow businesses put the Agent OS to work
- AI Tool Verdictevery Agent OS feature scored against the GitHub alternatives
- agentos.guidethe Agent OS quick start and daily workflow
- aiprofitboardroom.comwhere to download an agentic OS safely
Download the real pack inside the Boardroom, plug in the brains the board backs, and you will have the only genuine answer to the Julian Goldie Agent OS GitHub search.
FAQ
Is the Julian Goldie Agent OS on GitHub?
No. Julian Goldie's Agent OS is not published as a public GitHub repo. It is only available inside the AI Profit Boardroom on Skool, as an installable download with step-by-step guides and video tutorials.
How do members install the Agent OS?
They download the dated pack from the AI Profit Boardroom classroom, install Node.js and double-click Start Agent OS on a Mac. Claude Code or Codex can also install it by following the Setup With AI playbook in the pack.
Does the Agent OS ship its own AI model?
No. The Agent OS is the dashboard, shared memory and task board, and models plug into it. Julian runs Claude, GPT-5.6, Kimi, GLM, Grok, Muse Spark, MiMo, Hy3, DeepSeek and local models inside his own build.
Which model scores highest of the ones the Agent OS runs?
Of the engines wired into Julian's build, Hermes MoA from the Mixture tab scores highest on GoldieBench at 8.17 across 47 tasks, followed by GPT-5.6 Sol at 8.16 across 50. Claude Opus 5, which you can run in Claude Code, averages 8.27 across 50.
Are there free alternatives to the Agent OS on GitHub?
Yes. Hermes Agent (github.com/NousResearch/hermes-agent), Herald OS (github.com/iamlukethedev/Herald-OS) and StarNet (github.com/androoAGI/starnet) are free under the MIT licence. They are not Julian's Agent OS.
Is a GitHub repo called the Julian Goldie Agent OS safe to install?
It is not official, because Julian has not published the Agent OS on GitHub. Do not run its install scripts or paste API keys into it.
Keep reading
Hermes Herald AI Agent: Which Brain To Run It On, With Real GoldieBench Scores
Herald OS: What It Is, How To Set It Up, And Which Benchmarked Models To Run Under It
Hark AI Agent: What It Is, The Model Behind It, And How Those Brains Score On GoldieBench🌐 Sister-site take: read this on agentos.guide