The StarNet GitHub repo lives at github.com/androoAGI/starnet, and it's the free, open-source source code for StarNet, the pixel-art space station where your AI agents live and work.
I cloned it onto my Mac, hit one install trap the README doesn't warn you about, and had my first agent awake a few minutes later.
So this is the operator walkthrough I wish I'd had on day one.
You'll get the exact repo URL, what's inside it, the commands that actually worked, the trap, how to connect a brain, how to update it, and a quick tour of what you see once it runs.
If you're wondering whether the repo is safe, whether it costs anything, or whether you need to be a developer to run it, I'll answer all three as we go.
What Is The StarNet GitHub Repo?
StarNet is a local-first desktop harness for building and running teams of AI agents, and the StarNet GitHub repo is where its full source code lives.
The repo describes it as a living pixel-art station where real AI agents do real work.
That wording matters, because the station is not a cartoon laid over a chatbot.
The README spells out the rule plainly: the interface must never claim something the harness cannot prove.
So when an agent walks to a desk in the station, a real model call and real tools are behind it.
Here are the hard facts I pulled straight from the repo on 7 October 2026.
| Fact | What the repo says |
|---|---|
| Repo URL | github.com/androoAGI/starnet |
| Author | Andrew Sims, who publishes as androoAGI |
| Licence | MIT for the code, while the StarNet name, logo and artwork stay with the author |
| Latest version | v0.13.1, tagged on 4 October 2026 |
| Package name | starnet-harness |
| Runtime | Node.js 18 or newer, with Node 22 matching CI |
| Desktop shell | Rust and Tauri, needed only if you build the app yourself |
| Default port | 8787 on your own machine |
| Repo size | Roughly 1.8 GB, so the clone takes a while |
| Default branch | feat/harness-backend rather than main |
The repo had over 1,100 stars and more than 200 forks when I checked it, which is a lot of attention for a project that was only created in June 2026.
Who Made StarNet And What Does It Cost?
StarNet was made by Andrew Sims, and the code is free under the MIT License.
That means you can fork it, change it and even use it commercially.
The one thing you can't do is ship your fork under the StarNet name, because the brand, sprites and station artwork are owned separately.
StarNet itself costs nothing to run.
What you pay for is the brain you plug into it, which can be a subscription you already have, an API key, or a free local model.
I'll cover the brain options in a minute, because that choice affects your bill more than anything else.
What Problem Does The StarNet Repo Solve?
Most agent tools give you a chat box and a log file, and that's it.
You can't see which agent is busy, which one is stuck, or what each one is allowed to touch.
StarNet turns that into a station you can actually look at.
Each agent gets its own workspace, memory, transcript and permissions, and you watch them move around the station while they work.
I'll be honest with you, because I was honest in the video too.
I think StarNet is mostly fun to play with right now, and I don't think there are a massive number of practical use cases yet.
But it's the smoothest, most enjoyable office-style agent setup I've tested, and it reminded me of the first time I tried Hermes Agent.
🔥 Want my full StarNet and agent-office setups? Inside the AI Profit Boardroom, I've got step-by-step agent tutorials, including how I built StarNet into my own Agent OS as a tab. You'll be learning with 3,400+ members, and there are four coaching calls a week where you can ask about your setup. → Get access here
Two Ways To Get StarNet From GitHub
There are two routes, and you should pick one before you start.
The first route is the desktop installer, which is the easiest path for most people.
The signed builds live on a separate repo called androoAGI/starnet-releases, not on the main source repo.
| Your machine | File to download |
|---|---|
| Windows 10 or 11, 64-bit | StarNet_0.13.1_x64-setup.exe |
| Mac with Apple Silicon (M1 to M4) | StarNet_0.13.1_aarch64.dmg |
| Intel Mac | StarNet_0.13.1_x64.dmg |
The install guide says to use the aarch64 file on Apple Silicon, because the x64 file would run under Rosetta instead.
It also says Windows is the most-tested target, macOS has less real-world coverage, and Linux is not a supported public release.
The second route is running from source, which is what I did and what the rest of this walkthrough covers.
You'd pick this route if you want the newest code, you want to read what it does, or you want to run it on your own port next to other tools.
How To Install StarNet From Source, Step By Step
These are the commands from the README, plus the extra step my Mac needed.
First, check that you have Node.js 18 or newer and Git installed.
node -v
git --version
Second, clone the StarNet GitHub repo and move into the folder.
git clone https://github.com/androoAGI/starnet.git
cd starnet
The clone is around 1.8 GB, so make a cup of tea while it downloads.
Third, install the dependencies, even though the README says you don't need to.
npm ci
Fourth, start the sidecar, which is the local Node runtime that runs your agents.
node sidecar/index.js
You can also run npm start, because it points at the same file.
Fifth, open http://localhost:8787 in your browser and the station loads.
Sixth, connect a brain on the first-run screen, which I'll walk you through next.
The install trap: zero-install isn't zero-install
The README says the sidecar uses Node core modules only, so it should run without installing anything.
On my Mac that wasn't true, and it wouldn't come up properly until I ran npm ci.
So if the station won't load after a fresh clone, run npm ci in the repo folder and start it again.
That one command saved me a lot of head-scratching.
Running StarNet on a different port
Port 8787 was already taken on my Mac by another app, so I run StarNet on 8788 instead.
The sidecar reads a PORT setting at startup, so this is all you need.
PORT=8788 node sidecar/index.js
Then you open http://localhost:8788 instead of 8787.
I keep mine running in the background so it's always there when I open my Agent OS.
How To Connect A Brain To StarNet
An agent in StarNet is only as smart as the model you connect, and the repo supports a long list of providers.
These are the ones I'd point most people at.
| Provider in StarNet | How you connect | Who it suits |
|---|---|---|
| Grok OAuth | Sign in with a SuperGrok or X Premium+ account, with no API key | People already paying for Grok |
| Claude CLI | Uses your existing Claude Code sign-in through the local claude command | Claude Code subscribers |
| Kimi OAuth | Sign in with a Kimi subscription | Kimi for Coding users |
| ChatGPT Codex | Sign in with a ChatGPT subscription | ChatGPT Plus and Pro users |
| OpenRouter | Paste one key to reach a broad catalogue of models | People who want to switch models often |
| Ollama | Point it at your local Ollama on 127.0.0.1:11434 | People who want free and offline |
The registry also lists direct API options for OpenAI, Anthropic, Gemini, xAI, Groq, Mistral, DeepSeek, Together, Fireworks, Perplexity and Cerebras, plus a custom slot for any OpenAI-compatible endpoint.
Keys are held by the sidecar and your operating system's keychain, never in the browser front end.
In the video I signed in with Grok in one quick step, then picked the model and the reasoning level.
You can choose low, medium, high or max reasoning.
I chose Grok 4.6 over Grok 4.7, simply because I prefer how Grok 4.6 talks.
Running it free with a local model
If you want zero bills, install Ollama, pull a model and pick OLLAMA as the provider.
ollama pull qwen3:8b
The README is honest that local models are smaller, slower and rougher on long tasks.
It also says to pick a model that supports tool calls, because agents act through tools, and it notes that an 8B model uses about 10 GB of graphics memory while it works.
My own install runs on a hosted brain rather than a local model, because I want the agents to be quick and sharp.
Create Your First Agent In StarNet
Once a brain is connected, creating an agent takes one or two clicks.
You give it a name and look at its appearance, which you can change.
You pick a personality, and the options include warm, dry, composed and unhinged.
You choose an approval level, which is either ask for approval or full power.
I'd start every new agent on ask for approval until you trust it.
Then a little waking-up animation plays, and my agent said it was awake and asked to give the station a direction.
A Quick Tour Once StarNet Is Running
I'm keeping this tour short, because the full feature-by-feature guide is in my StarNet AI agent walkthrough.
Here's what you'll see in your first ten minutes.
- The station view lets you zoom in and out, change the frame, and drag around, and everything you run lives in four docks you can click open.
- Your agents level up the more you use them, so it feels a bit like a Tamagotchi for an AI agent.
- The agent panel shows the model, personality, instructions, web and local-file permissions, and approval prompts, with Brief, Growth, History and Memories tabs.
- The chat lets you talk to any agent, shows its online status and level, and lets you switch agents from the list on the left and manage projects and sessions.
- The plus button at the top adds widgets such as connected apps, pinned widgets and a daily brief.
- The app catalogue adds abilities like Gmail, Notion, GitHub, flight search, physical letters, email and web search through Firecrawl, and most of them sign in directly while some need an API key.
- The skill market has ready-made skills like ad copy testing and calendar scheduling, and the skill library holds frameworks such as making a plan, a decision framework for hard technical problems, humanising content and handling a hard conversation, each of which you can switch on or off.
- The crew recruit screen adds roles like strategist, chief of staff, researcher and marketer, where Deploy rewrites an existing agent with the role and Summon adds a new agent beside yours.
- Tasks, quests, the work list, builds and the Systems menu all sit in the same station.
I noticed a slight delay when I sent a message, so don't panic if a reply takes a moment.
It felt like a much more fun version of Paperclip to me.
🔥 Want to see how I run agents like these every day? The AI Profit Boardroom has my agent courses on Hermes, Grok Bot, DeepSeek Harness and more, with four live coaching calls every week. → Join the AI Profit Boardroom
How To Update StarNet From GitHub
Updating a source install is a four-step job.
First, stop the sidecar with Ctrl+C in its terminal.
Second, pull the latest code.
git pull
Third, run npm ci again in case the dependencies changed.
Fourth, start the sidecar back up and reload the browser tab.
If you've edited any files in the repo yourself, a pull can clash with your changes, so keep notes of what you touched.
Desktop installs work differently, because they update through StarNet's built-in Update Center, which checks each update's signature before installing it.
Version v0.13.1 was a repair release, and it's worth having.
It fixed web requests that had been failing with a "fetch failed" error since 0.12.5, stopped crew freezing in the halls, made group chats save properly, and made spending limits hold during retries.
Useful Extras Hiding In The StarNet Repo
The repo has a few things most people never notice.
There's a terminal command in bin/starnet.js that runs one agent task without the station UI, using a -p flag for the prompt.
The package file also includes scripts called mcp:serve and acp:serve, which start MCP and ACP servers from the sidecar.
If you contribute, the README says npm run test:fast is the required merge gate, and npm test runs the full suite.
And if you already run Hermes or OpenClaw, StarNet can import an existing agent from its home folder, bringing across the persona, instructions, memory and model.
Your API keys don't transfer, so you re-enter those in the KEYS tab.
That answers the viewer who told me he was annoyed StarNet doesn't work with Hermes.
You can't run Hermes inside StarNet, but you can bring a Hermes agent's personality and memory across.
Common StarNet GitHub Problems And Fixes
| Problem | What to do |
|---|---|
| The station won't load after cloning | Run npm ci in the repo folder and start the sidecar again. |
| Port 8787 is already in use | Start it with PORT=8788 node sidecar/index.js and open that port instead. |
| Ollama shows as not ready | Make sure Ollama is running and you've pulled a model, because StarNet only reports ready once it can list your models. |
| A local agent does nothing useful | Pick a model that supports tool calls, since agents act through tools. |
| The Mac app runs slowly | Check you installed the aarch64 file on Apple Silicon and not the x64 one. |
| You can't find the main branch | The default branch is feat/harness-backend, and the clone picks it up for you. |
The install guide also says that if macOS or Windows flags an unknown publisher on a public release, you should stop and report it rather than bypass the warning.
I'd follow that advice, because it's a sensible security habit for any tool you download.
If you'd like help planning which agents are worth running in your business, you can book a free SEO strategy session with my team.
Related Reading
- How to set up Hermes Agent in one click is the quickest route if you'd rather run a single agent first.
- My Claude Obsidian setup shows you how to give any agent a proper memory to work from.
- The full StarNet AI agent feature guide walks through every screen in the station.
Also On Our Network
- 🌐 How businesses and creators can put the StarNet repo to work
- 🌐 My feature-by-feature verdict on the StarNet GitHub project
- 🌐 A StarNet quick start inside an Agent OS workflow
- 🌐 Which benchmarked models make the best StarNet brain
FAQ: StarNet GitHub
Where is the StarNet GitHub repo?
The source code is at github.com/androoAGI/starnet.
The signed Windows and macOS installers live on a separate repo, github.com/androoAGI/starnet-releases.
Is StarNet free and open source?
Yes, the code is free under the MIT License from Andrew Sims.
The name, logo and artwork aren't covered by that licence, and you still pay your model provider unless you use a local model.
What do I need to run StarNet from source?
You need Node.js 18 or newer and Git, and Node 22 matches what the project tests on.
Rust and the Tauri prerequisites are only needed if you build the desktop shell yourself.
Why does StarNet fail to start after cloning?
The README says it needs no install step, but my Mac needed npm ci before the sidecar would come up properly.
Run it once in the repo folder and start the sidecar again.
How do I update StarNet from GitHub?
Stop the sidecar, run git pull, run npm ci again, and start it back up.
Desktop installs update through the built-in Update Center instead.
📺 Video notes + links to the tools 👉
🎥 Learn how I make these videos 👉
🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉
About Julian
I'm Julian Goldie, an SEO entrepreneur, author and founder of the AI Profit Boardroom, which has 3,400+ members.
I help business owners scale with AI agents, automation and SEO.
- I've built a 7-figure agency, Goldie Agency, with a team of around 50 people.
- I've grown a YouTube channel to 400,000+ subscribers.
- I wrote the Amazon best-sellers "SEO Link Building Mastery" and "Agency Marketing Mastery".
- My Udemy courses have taught over 50,000 students.
→ Get my best AI training inside the AI Profit Boardroom
Clone it, run npm ci, connect a brain, and you'll have your own station running from the StarNet GitHub repo today.
