Hermes3D: Repo, Setup And 2D Pixel Mode (2026)

Hermes3D is a free, open-source 3D office for AI agents, and this page is the setup guide: what is in the repo, which backend mode to pick, and how to get it running on your own machine in minutes. If you want the quick "what is this thing" overview first, read my Hermes 3D office explainer. This page is the hands-on part.

📺 Watch: Hermes3D Is INSANE: AI Agents Now Have a 3D Office

🔥 Get the Agent OS as a free bonus: AI Profit Boardroom members get the full Agent OS zip, prompt libraries, daily tutorials and weekly live coaching calls. → Get inside

The repo is iamlukethedev/Hermes3D. Per the README, it is "a 3D virtual office for AI agents you run on your own infrastructure." It is built by LukeTheDev, it is MIT licensed, and it is unofficial and community-driven. It is not affiliated with or endorsed by the teams behind the agent backends it connects to.

If you want to go deeper on Hermes Agent itself, the full Hermes courses live inside the AI Profit Boardroom ($69/mo), along with the agent operating system I use every day.

What Is in the Hermes3D Repo

The line from the README that sums it up best is this: "The Gateway is just the control plane — the product is the office." In plain English, the boring plumbing that talks to your agent is there so you can see your agents working in a shared space.

Per the README, the repo sits at about 575 stars, 123 forks and 5 contributors, with a release tagged Hermes3D 1.0.0. That is a real project with real people using it, but still a small one. Treat it like a fun, capable side project, not a product with a support desk.

Here is what is under the hood, per the README:

Next.js, React and TypeScript for the app itself.
Three.js, React Three Fiber and Drei for the 3D office.
Phaser for the 2D pixel office.
ElevenLabs keys if you want voice replies from your agents.
An /office/builder for editing and publishing office layouts.
An optional office bridge plugin that makes the office react to chats you have in Hermes Desktop, the TUI or the CLI.

What it does not do: install Hermes for you. Hermes3D is the office. You bring the agent.

The Three Backend Modes

This is the bit most people get stuck on, so here it is in one table. Pick the mode that matches where you are right now.

Mode	What it connects to	Command (per the README)	Best for
Demo gateway	A built-in fake gateway, no agent framework needed	npm run demo-gateway, then npm run dev	Seeing the office in minutes
Hermes adapter	The Hermes HTTP API at localhost:8642 by default, through an adapter on ws://localhost:18789	npm run hermes-adapter, then npm run dev	Running your real Hermes agents
Custom runtime	A direct HTTP custom runtime provider	Set in Studio	Builders wiring in their own agent runtime

My advice: always start with the demo gateway. If the office loads and looks right, you know Node, npm and the app are fine. Then you swap in real Hermes. If something breaks at that point, you know it is the connection, not the install.

Before You Start: Requirements

Per the README you need Node.js 20 or newer and npm 10 or newer. That is it for the office. For the Hermes mode you also need Hermes Agent already installed and running, because Hermes3D does not install it.

If you have never set up Hermes before, get that working first. Get a normal chat going in the terminal, then come back here.

Step 1: Run Demo Mode in Minutes

This is the fastest way to see the office. Per the README, running from source looks like this:

Clone the Hermes3D repo to your machine.
Run npm install inside the folder.
Copy the .env.example file to a new file called .env.
Run npm run demo-gateway to start the built-in demo gateway.
In a second terminal, run npm run dev.
Open localhost:3000 in your browser.
In Studio, set the gateway URL and token so the app talks to the demo gateway.

That is it. You now have a 3D office with demo agents moving around. No API keys, no agent framework, nothing to break. This is the version to show a friend or a client who asks what "agents in an office" even means.

Quick tip: if you do not want to type commands yourself, paste the steps above into your AI agent and ask it to do them.

Step 2: Connect Your Real Hermes Agent

Once the demo works, stop the demo gateway and switch to the Hermes adapter. Per the README:

Make sure Hermes Agent is running and its HTTP API is up. The adapter looks for it at localhost:8642 by default.
Run npm run hermes-adapter. The adapter listens on ws://localhost:18789.
In a second terminal, run npm run dev.
Open localhost:3000 and point Studio at the adapter's gateway URL and token.

Now the agents in the office are your actual Hermes agents. If you also install the optional office bridge plugin, the office reacts when you chat with Hermes in Hermes Desktop, the TUI or the CLI. So you can keep working the way you normally do, and the office shows what is going on.

If you would rather have someone map out how agents like this fit into your business, book a free AI strategy session and we will go through it with you.

Step 3: Connect to a Remote Hermes Over Tailscale or SSH

A lot of people run Hermes on a VPS or a home server, not on their laptop. Hermes3D handles that. Per the README you have two routes:

Tailscale serve. Expose the service over your private Tailscale network so only your own devices can reach it. The repo includes a doc for connecting to a remote hermes-agent over Tailscale with no adapter at all.
SSH tunnel. Forward the remote port to your local machine and point Studio at the local end of the tunnel.

Either way, the office runs where you are and the agent runs where it lives. Tailscale is the easier one if you already use it. SSH is the one that works everywhere with nothing extra installed.

The New 2D Pixel Office Mode

Not every machine wants to render a 3D scene all day. Per the README, Hermes3D now has a 2D pixel office mode, Gather-style, built for low-power machines. You switch to it in Settings, under Office renderer, or with the 2D button.

It runs on Phaser instead of Three.js, so it is lighter. Same agents, same idea, just top-down pixel art instead of a 3D room. If you like the pixel look, I also covered a separate project in my Hermes pixel office write-up.

Features: Standups, PR Review, Gym, Janitor and Builder

This is where Hermes3D goes beyond "cute room with agents in it." Per the README you can:

Run standups with agents connected to GitHub and Jira, so they report on real work.
Review pull requests inside the office.
Monitor QA pipelines and logs without leaving the office view.
Train agents in the gym to develop skills.
Reset sessions with the janitor system when an agent's context gets messy.
Edit and publish layouts with the /office/builder.
Get voice replies if you add ElevenLabs keys.

The standup and PR review pieces are the ones I would test first if you run a dev team. The gym and janitor are the fun ones that also happen to be useful.

Security Settings You Should Not Skip

This is the boring section that saves you a headache. Per the README:

Set STUDIO_ACCESS_TOKEN before you expose Studio beyond localhost. Without it, anyone who can reach the page can reach your office.
Set UPSTREAM_ALLOWLIST in production, so the app only talks to the upstream hosts you approve.
Know where your token lives. The browser holds the gateway URL and token in memory at runtime.

Simple rule: keep it on localhost or behind Tailscale until you have set those two values. Your agent can run tools on your machine. Do not hand that to the open internet.

Limitations to Know Up Front

Per the README, a few things are not finished or work differently than you might expect:

/office and the builder are separate stacks. They do not share everything.
Voice transcription is not bundled. The endpoint returns 501, so speaking to agents needs your own setup.
The Spotify jukebox (SOUNDHERMES) stores only an access token.
It does not install Hermes. You need a working agent first.

None of that is a deal breaker. It is a version 1.0 community project. Go in knowing that and you will have a good time with it.

Where Hermes3D Fits With Other Hermes Front Ends

Hermes3D is a visual layer. It does not replace how you actually run agents day to day. LukeTheDev also built Herald OS, which is a full agent-native OS with Hermes as the interface. Different job: Herald OS is where you work, Hermes3D is where you watch.

For my own daily work I use the custom agent operating system we built inside the AI Profit Boardroom. It runs Hermes Agent and other agents like Claude Code and Codex in one dashboard. The Agent OS guide walks through it, and the Goldie Bench is where I compare how models actually perform on real tasks.

I covered Hermes3D on YouTube in "Hermes3D Is INSANE: AI Agents Now Have a 3D Office" on 24 August 2026 if you want to see it moving.

If you want the full Hermes training plus the Agent OS zip and live calls, join the AI Profit Boardroom. If you want help planning agents for your business, grab a free AI strategy session.

FAQ

Is Hermes3D free?

Yes. Per the README it is MIT licensed and open source. You run it on your own machine or server. Any costs come from the agent and model you connect, not from Hermes3D.

Do I need Hermes Agent to try Hermes3D?

No. The built-in demo gateway runs the office with no agent framework at all. You only need Hermes when you want your real agents in the office, and Hermes3D does not install it for you.