StarNet AI agent is a free, open-source harness that lets you create real AI agents and watch them work inside a pixel-art space station on your own computer.
I tested it on my own machine, signed it into Grok, and had my first agent awake and talking to me in about two clicks.
This is the operator walkthrough I wish I'd had on day one.
I'll show you what StarNet is, how to create your first agent, how to connect a brain, how to set permissions, and how to recruit a whole crew.
I'll also be honest about where it falls short, because it isn't the right tool for every job.
What The StarNet AI Agent Actually Is
StarNet is a local-first desktop harness for building and running AI agent teams.
It was built by Andrew Sims, and the code lives on GitHub under the androoAGI account.
The version I ran was v0.13.1, and it's released under the MIT License, so the code is free to use.
The big idea is that your agents live in a pixel-art space station instead of a boring chat window.
That station isn't decoration.
The README describes it as a projection of live runtime state, which means what you see on screen is what the agents are actually doing.
A room is a team with certain capabilities, a hallway is a handoff lane, and an object you place is a real capability grant.
The agents make real model calls, use real tools and run up a real bill if you connect a paid model.
There's one rule in the README that I love as an operator.
It says the interface must never show a state the harness cannot prove.
So when an agent looks busy, it really is busy.
Why Operators Should Care About A Pixel-Art Agent Station
I'll be straight with you on this one.
In my video I said I think StarNet is fun to play with, and I don't think there's a massive amount of practical use cases yet.
That's still my honest view.
But it's the smoothest office-style agent add-on I've tested, and it made me feel the way I did the first time I tried Hermes Agent.
It also feels like a much more fun version of Paperclip.
For an operator, the value is visibility.
You can see every agent, its level, what it's working on and what it's allowed to touch, all in one picture.
That makes it a great way to learn how agent teams work before you build serious automations elsewhere.
Step 1: Get StarNet Running On Your Machine
The quick version is that you either download the desktop app or run it from source.
Desktop builds are published for Windows and for macOS on both Apple Silicon and Intel.
The README says Windows is the most-tested target and macOS has had less real-world use so far.
If you run it from source, you need Node.js 18 or newer and Git.
Once it's running, it opens as a local web app in your browser.
One honest note from my own install is that the README pitches the source route as zero-install, but my setup still needed an npm ci before it ran properly.
I've written the full install walkthrough, including the GitHub repo, requirements, licence and updates, in my StarNet GitHub setup guide.
Step 2: Create Your First StarNet AI Agent
When StarNet opens for the first time, it walks you straight into creating an agent.
You give the agent a name first.
Then you can see its pixel-art appearance and change it if you want a different look.
Next you pick a personality, which sets how the agent talks to you.
| Personality | What it feels like in chat |
|---|---|
| Composed | It stays calm, measured and professional. |
| Warm | It's friendly and encouraging. |
| Blunt | It gets to the point without padding. |
| Dry | It answers with a deadpan, understated tone. |
| Unhinged | It's chaotic and colourful, and the code even asks you to confirm twice before you pick it. |
| Upbeat | It brings high energy to every reply. |
In the video I talked through warm, dry, composed and unhinged, and the source code adds blunt and upbeat as well.
You can change the personality at any time later, so don't overthink it.
Step 3: Choose The Approval Level
This is the most important setting for any operator.
StarNet gives you two broad modes when you create an agent.
The first mode is "ask for approval", where the agent checks with you before it does something.
The second mode is "full power", where the agent just gets on with the job.
My advice is to start every new agent on ask for approval.
Watch what it tries to do for a few sessions, then loosen the leash once you trust it.
That's exactly how I'd onboard a new team member, and it's the same principle with agents.
Step 4: Connect A Brain And Pick The Model
An agent needs a brain, which means a model.
StarNet gives you two ways to connect one.
You can sign in with an account you already pay for, or you can paste an API key.
I signed in with Grok in one quick step, and it was the smoothest part of the whole setup.
| Brain option | How you connect it |
|---|---|
| Grok | Sign in with your Grok account. |
| Kimi | Sign in with your Kimi account. |
| Claude | Use your Claude Code subscription through the Claude CLI, or paste an Anthropic key. |
| ChatGPT | Sign in through Codex. |
| OpenRouter | Paste one OpenRouter key and pick from many models. |
| Other APIs | Paste a key for OpenAI, Gemini, xAI, Groq, Mistral, DeepSeek, Together, Fireworks, Perplexity or Cerebras. |
| Local and free | Pick Ollama and run a model on your own machine. |
After you connect, you pick the model and the reasoning level.
The reasoning levels run from low through medium and high up to max.
I chose Grok 4.6 over Grok 4.7, because I actually prefer Grok 4.6 for this kind of chat.
The README says your keys live in the operating system keychain and never in the front end, which is how it should be.
If you want to go fully free, the README suggests installing Ollama and pulling a model such as qwen3:8b.
Pick a model that supports tools, because StarNet agents act through tool calls.
The README is honest that local models are slower and rougher on long tasks, and that an 8B model needs about 10 GB of graphics memory while it works.
Step 5: Wake The Agent Up
Once the brain is connected, the agent is created in one or two clicks.
A "waking up" animation plays, and then the agent introduces itself.
Mine said "I'm awake. Let's give the station a direction."
That little moment is why I keep calling StarNet fun.
🔥 Want my full agent setups, not just the fun ones? Inside the AI Profit Boardroom, I've got step-by-step agent training, including Hermes Agent, Grok Bot, DeepSeek Harness and Claude Opus 5.5 courses. You'll be learning alongside 3,400+ members, with four coaching calls a week. → Get access here
Step 6: Learn The Station View Controls
The station view is where you'll spend most of your time.
You can zoom in and out, change the frame and go to a full zoom.
Everything you run lives in four docks around the station.
You click a dock to open it, you can show everything at once, and you can drag things around the screen.
Your agents also level up the more you use them.
It honestly feels like a Tamagotchi for an AI agent, and it turns using your agents into a game.
Step 7: Set The StarNet AI Agent's Permissions
Click any agent and you get its detail panel.
This is where an operator should spend real time.
- The model setting shows which brain the agent is using, and you can swap it.
- The personality setting can be changed at any time.
- The instructions field holds the agent's purpose, so write a clear job description here.
- The access permissions control what it can reach, such as the web and your local files, and you can allow or remove each one.
- The approval prompts setting controls when the agent has to ask you first.
The panel also has four tabs.
The Brief tab shows what the agent is focused on, the Growth tab tracks its levelling, the History tab shows what it has done, and the Memories tab shows what it has remembered.
My rule is simple.
Give an agent only the access its job needs, and nothing more.
Step 8: Chat With Your Agent And Switch Between Agents
The chat is where you actually talk to your agent.
My agent asked me, "What made you want to set up an AI agent?"
I told it SEO and gave it aiprofitboardroom.com, and it picked up from there.
The chat shows whether the agent is online and what level it's at.
The list on the left lets you switch between all your agents and see which are idle and which are online.
You can also manage projects and sessions from here.
There's a slight delay when you send a message, so don't panic if a reply takes a moment.
Step 9: Add Widgets And Apps
Press the plus button at the top to add widgets.
You'll see connected apps, pinned widgets, a chooser for apps and a daily brief.
You decide what information each widget shows.
The app catalog is labelled "add an ability", and it's where your agents get new powers.
It includes Gmail, Notion and GitHub, and most of them sign in directly while some need an API key.
There are real-world categories too.
You'll find things like finding flights, which turns an agent into a travel helper, plus sending physical letters, email and web search through Firecrawl.
The README adds that you can attach MCP connectors and wire agents to Telegram, Discord, Slack, Signal or Matrix so you can talk to your station away from your desk.
Step 10: Install Skills And Turn On Frameworks
The skill market has standard originals, such as ad copy testing.
Some skills come built in, some you install later, and the community can push updates to them.
Calendar scheduling is another example I spotted.
The skill library is different, because it holds frameworks, which are the procedures the AI follows.
| Framework example | What it helps the agent do |
|---|---|
| Decision framework | It works through hard technical problems step by step. |
| Make a plan | It breaks a goal into an ordered plan before acting. |
| Humanising content | It rewrites output so it sounds less robotic. |
| Hard conversation | It helps prepare for a difficult conversation. |
You can enable or disable each framework, so only switch on the ones the agent really needs.
Step 11: Recruit, Deploy And Summon A StarNet Crew
One agent is fun, but a crew is where StarNet starts to feel like a real team.
Open Crew and then Recruit.
You'll see roles such as strategist, chief of staff, researcher and marketer.
You give each recruit a name and an appearance, just like your first agent.
Then you have two choices, and operators mix these up all the time.
- Deploy overwrites an existing agent's setup with the role you picked.
- Summon adds a brand new agent with that role, and it appears standing next to your others in the station.
If you want to keep your first agent as it is, always choose Summon.
The README says each agent gets its own workspace, transcript, memory and permissions, and several can run at the same time.
Step 12: Run Tasks, Quests And Systems
The last area pulls everything together.
You get tasks, quests, a work list, a place to build new stuff, a Systems menu and builds, all in one system.
The README also lists recipes, schedules on cron and a Night Shift mode.
Night Shift lets agents keep working while you're away, inside a leash you set, and every away action is logged for you to review.
Finished work lands in an OUTBOX as real files, so you're not digging through chat to find a deliverable.
Spend, budgets and run history are saved on disk and shown as they are.
Tips And Limits Every Operator Should Know
Start with one agent and one clear job before you build a crew.
Keep approval on until you've watched the agent work.
Remember that a paid model connected to StarNet still costs real money, because the calls are real.
Expect a short delay in chat replies.
Treat it as early software, because the README itself calls this an early release.
If you came from Hermes, you can't use Hermes as the brain, but the README says StarNet can import an existing Hermes or OpenClaw agent from its home folder.
It brings across the persona, instructions, memory and model, and you re-enter API keys yourself.
A few viewers asked me for more tutorials, so this walkthrough is my answer to that.
I also run StarNet as a tab inside my own Agent OS, so it sits next to my other agents rather than replacing them.
If you want a serious always-on agent first, start with how to set up Hermes Agent in one click.
🔥 Want to build StarNet into your own agent operating system? The AI Profit Boardroom has the agentic operating system where you can build StarNet in, plus the training and weekly coaching calls to help you set it up. → Join the AI Profit Boardroom
My Verdict On The StarNet AI Agent For Operators
StarNet is the most fun way I've found to see what a team of AI agents is doing.
The setup is fast, the Grok sign-in is painless, and the permissions panel is clearer than most harnesses I've used.
It isn't my main production system, and I wouldn't pretend it is.
But as a free, open-source way to learn agent teams, it's well worth an afternoon.
If you'd like help planning where agents fit in your business, you can book a free strategy session with my team.
Related Reading
- My StarNet GitHub setup guide covers the repo, the requirements, the licence and how to update.
- How to set up Hermes Agent in one click is the fastest way to get a serious always-on agent running.
- My Claude Obsidian setup shows you how to give agents a memory full of your real experience.
Also On Our Network
- 🌐 How businesses and creators actually use StarNet agents
- 🌐 StarNet reviewed feature by feature, with scores
- 🌐 StarNet quick start inside an Agent OS workflow
- 🌐 The Grok, Claude and Kimi brains behind StarNet, benchmarked
- 🌐 Hermes agents in their own pixel office
FAQ: StarNet AI Agent
What is the StarNet AI agent?
StarNet is a free, open-source, local-first desktop harness where you create AI agents and watch them work inside a pixel-art space station.
The agents make real model calls and use real tools, and the station shows their live state.
Is StarNet free?
Yes, the StarNet code is open source under the MIT License.
The model you connect may cost money, unless you use a subscription you already pay for or a free local model through Ollama.
Which models can a StarNet agent use?
You can sign in with Grok, Kimi, ChatGPT through Codex or the Claude CLI.
You can also paste API keys for OpenRouter, Anthropic, OpenAI, Gemini, xAI, DeepSeek and more, or run local models through Ollama.
Can I use StarNet with Hermes Agent?
StarNet doesn't run Hermes as its brain.
Its README says it can import an existing Hermes or OpenClaw agent, bringing across the persona, instructions, memory and model, while API keys have to be entered again.
What is the difference between Deploy and Summon in StarNet?
Deploy overwrites an existing agent's setup with the crew role you picked.
Summon adds a brand new agent with that role, standing next to your current agents in the station.
📺 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
Create one agent, keep approval on, recruit a crew when you're ready, and you'll be running your first StarNet AI agent the right way.
