A one-click local AI assistant for Windows 11 — built for noobs, by a noob, with the help of an AI.
Turns your gaming PC into a private, offline AI workstation with a multi-agent crew (Foreman + specialist workers) that can read/write files, run commands, update Windows, sort folders, write code, search the web, remember things, and more.
💰 100% free. No accounts. No API keys. No subscriptions. Everything runs locally on your hardware.
The easiest way to use this toolkit is the GUI:
Double-click
LAUNCHER.bat— that's it.
You get a friendly window with big buttons, live status lights, and an activity log. No PowerShell. No command line. No decisions to make beyond clicking the next button down the list.
Prefer the manual route? Every *-Run-Me.bat file still works standalone.
- 🧠 Local AI — Qwen3-Coder runs on your GPU via Ollama. Nothing leaves your PC.
- 👷 Multi-agent crew — A Foreman delegates to specialists (Sysadmin, Coder, Librarian, Researcher, Inspector).
- 🛡️ Safety-first — Every command is shown before running. Read-only Inspector mode for diagnosis. Librarian always dry-runs first.
- 🚚 Solves the "moving models breaks everything" headache — One script fixes
OLLAMA_MODELSproperly so models stay where you put them. - 🔌 8 free MCP superpowers — Web search, web fetching, long-term memory, sequential reasoning, Git, SQLite, time, fast filesystem.
- 🧹 Clean uninstaller — Finds and removes every orphan model folder across all your drives.
- 🩺 Health Check — One click shows you exactly what's installed, what's broken, and where your models actually live.
- 🎛️ GUI launcher — Big buttons, live status, activity log. No command line needed.
- 🎨 Friendly desktop icon — Procedurally generated, no external image needed.
Run MAKE-ICON-Run-Me.bat and this little robot lands on your desktop.
Built and tested on:
| Component | Spec |
|---|---|
| GPU | NVIDIA RTX 3070 Ti (8 GB VRAM) |
| CPU | Intel i7-11700K |
| RAM | 64 GB |
| OS | Windows 11 Pro 25H2 |
Should work on anything with ≥ 6 GB VRAM and ≥ 16 GB RAM. Less VRAM? Swap qwen3-coder:14b for qwen3:8b in the scripts.
-
Download the repo — click the green Code button → Download ZIP → extract somewhere easy like your Desktop.
-
Double-click
MAKE-ICON-Run-Me.bat— generates a friendly robot shortcut on your desktop. -
Double-click the new 🤖 NoobAI icon on your desktop.
-
Click Yes on the UAC (Admin) prompt.
-
In the launcher window, click the buttons top to bottom:
- 📦 Install everything (base) — ~15–30 min, big download
- 🚚 Move AI models to another drive — optional but recommended
- 👷 Add the AI Crew (specialists)
- ⚡ Add MCP Superpowers (free)
-
Click 🚀 Launch the AI Crew to start chatting. 🎉
The status panel updates after every step so you can always see what's installed and what's not.
Prefer to skip the launcher? Double-click the .bat files in order:
| # | Double-click | What it does |
|---|---|---|
| 1 | START-HERE-Run-Me.bat |
Installs Ollama, VS Code, Git, Cline, and the AI model |
| 2 | MOVE-MODELS-Run-Me.bat |
(Optional) Move models off C: to another drive properly |
| 3 | SETUP-CREW-Run-Me.bat |
Installs Roo Code + the 6-agent crew |
| 4 | SETUP-SUPERPOWERS-Run-Me.bat |
Adds 8 free MCP superpowers |
| 5 | HEALTH-CHECK-Run-Me.bat |
Confirms everything is green ✅ |
Once installed, open VS Code via the desktop shortcut, click the Roo Code (kangaroo) icon in the sidebar, and pick a mode:
| Mode | Specialist | Use for |
|---|---|---|
| 👷 Foreman | Manager | Always start here. Reads your request, delegates to the right worker. |
| 🛠️ Sysadmin | Windows expert | Updates, installs, PATH, env vars, services |
| 💻 Coder | Programmer | Writes, builds, debugs code |
| 📚 Librarian | File organiser | Sorts, renames, dedupes — always dry-runs first |
| 🔍 Researcher | Planner | Reads docs, compares options (read + markdown only) |
| 🔎 Inspector | Auditor | Diagnoses problems — hardcoded read-only, can't break anything |
"My PC feels slow. Sort it out." → Inspector audits → Foreman shows plan → Sysadmin disables startup junk, runs updates, cleans disk.
"My Downloads folder is a mess." → Librarian dry-runs sorting it by file type → you approve → it moves them and logs every action.
"Build me a Python script that backs up my Documents to D: every night." → Researcher plans → Coder writes & tests → Sysadmin registers a scheduled task.
"Why won't
noderun from PowerShell?" → Inspector reads your PATH → reports findings → Sysadmin fixes it.
After running the Superpowers step, every specialist gains these abilities:
| Superpower | What it does |
|---|---|
| filesystem | Fast, safe file ops across only the folders you whitelist |
| fetch | Grab any web page as clean text |
| duckduckgo | Web search — no API key, no account |
| git | Full git operations (status, diff, log, commit, etc.) |
| memory | Long-term memory across chats (knowledge graph in AI-Crew/memory.json) |
| sequential-thinking | Forces step-by-step reasoning — makes the small model behave much smarter |
| time | Current time, date math, timezone conversion |
| sqlite | Build/query local databases |
All are open-source, no keys, no accounts. The cheat sheet on your desktop lists what each does.
| File | Type | Purpose |
|---|---|---|
LAUNCHER.bat |
GUI front door | Start here. Opens the friendly window. |
NoobAI-Launcher.ps1 |
GUI engine | The WPF window logic (called by LAUNCHER.bat) |
MAKE-ICON-Run-Me.bat |
Launcher | Calls Make-Icon.ps1 |
Make-Icon.ps1 |
Worker | Generates NoobAI.ico + creates desktop shortcut |
START-HERE-Run-Me.bat |
Launcher | Calls Setup-LocalAI.ps1 |
Setup-LocalAI.ps1 |
Worker | Base install: Ollama, VS Code, Git, Cline, model |
MOVE-MODELS-Run-Me.bat |
Launcher | Calls Move-AI-Models.ps1 |
Move-AI-Models.ps1 |
Worker | Moves models to another drive + sets OLLAMA_MODELS |
SETUP-CREW-Run-Me.bat |
Launcher | Calls Setup-AI-Team.ps1 |
Setup-AI-Team.ps1 |
Worker | Installs Roo Code + defines the 6-agent crew |
SETUP-SUPERPOWERS-Run-Me.bat |
Launcher | Calls Setup-MCP-Superpowers.ps1 |
Setup-MCP-Superpowers.ps1 |
Worker | Installs Node.js, uv, and 8 MCP servers |
HEALTH-CHECK-Run-Me.bat |
Launcher | Calls Health-Check.ps1 |
Health-Check.ps1 |
Worker | Read-only diagnosis: installs, env vars, model locations, GPU |
UNINSTALL-Run-Me.bat |
Launcher | Calls Uninstall-LocalAI.ps1 |
Uninstall-LocalAI.ps1 |
Worker | Step-by-step removal (every step opt-in) |
CHANGELOG.md |
Doc | Version history |
LICENSE |
Doc | MIT licence |
Every .ps1 script has a header with version + repo URL. The .bat files are friendly launchers — all the real work is in the matching .ps1 files. Open any of them in Notepad to see exactly what they do.
If you've ever moved your Ollama models from C: to another drive and then everything broke or got duplicated, here's why:
Ollama only checks one environment variable (
OLLAMA_MODELS) to find your models. If you move the folder but don't update that variable, Ollama silently falls back to the default location on C: and re-downloads everything, leaving orphan duplicates eating your disk.
The 🚚 button (or MOVE-MODELS-Run-Me.bat) fixes this:
- Scans every drive for stray model folders (finds your duplicates!)
- Lets you pick the real one and the destination drive
- Uses
robocopyto safely copy with verification - Sets
OLLAMA_MODELSsystem-wide (Machine scope, permanent) - Restarts Ollama and runs
ollama listso you see proof it works
The 🩺 Health Check shows you orphan folders any time.
| Problem | Try this |
|---|---|
winget not found |
Install App Installer from the Microsoft Store, reboot, retry. |
| Scripts won't run / "execution policy" error | Use LAUNCHER.bat or any *-Run-Me.bat — never run .ps1 directly. |
| VS Code 'code' command not found | Open VS Code once manually, then re-run the script. |
| Model download stuck | Press Ctrl+C, click 🔄 Update AI models in the launcher. |
| Roo Code doesn't see MCP servers | Close & reopen VS Code. First-time server fetch takes 5–30 seconds. |
| "Out of VRAM" errors | Edit the scripts to use qwen3:8b instead of qwen3-coder:14b. |
| GUI launcher won't open | Make sure all the .ps1 files are in the same folder as LAUNCHER.bat. |
| Something else broken | Click 🩺 Health Check in the launcher — it shows you exactly what's wrong. |
| Tool | Purpose | License |
|---|---|---|
| Ollama | Runs LLMs locally | MIT |
| VS Code | The editor/chat host | MIT |
| Git | Version control | GPL v2 |
| Cline | AI agent extension | Apache 2.0 |
| Roo Code | Multi-agent extension | Apache 2.0 |
| Node.js | Runs MCP servers | MIT |
| uv | Runs Python MCP servers | MIT/Apache 2.0 |
| Qwen3-Coder | The AI brain | Apache 2.0 |
| 8 MCP servers | Superpowers | MIT |
| Everything (optional) | Instant file search | Freeware |
- A 14B local model is roughly as smart as GPT-3.5 / early GPT-4 for code. The MCP superpowers (especially sequential-thinking + memory) close a lot of that gap, but don't expect Claude/GPT-5-level performance.
- Speed is ~15–25 tokens/sec on an RTX 3070 Ti. Snappy enough for real work but not instant.
- The AI will occasionally suggest wrong commands. Always read what it shows you before approving. The Inspector mode is hardcoded read-only — use it when in doubt.
- This is built for Windows 11. Probably works on Windows 10 but untested.
- Every PowerShell script self-elevates to Administrator (you'll see a UAC prompt).
- Every script is idempotent — safe to re-run; skips anything already installed.
- The Uninstaller asks Y/N for every destructive step; default is No.
- The Inspector specialist cannot run write/install/delete commands — enforced by the extension, not just trust.
- The Librarian specialist always dry-runs first and logs every action for undo.
- Nothing phones home. Read the scripts yourself — they're plain text.
- AI scaffolding & scripts: built with the help of Arena.ai's Agent Mode
- Multi-agent pattern inspired by Cline and Roo Code
- All the open-source projects listed in the "What gets installed" table — the real heroes
- Colour palette: Catppuccin Mocha
MIT — do whatever you want with it. No warranty. If it breaks your PC, you get to keep both pieces. 😅
Drop a star on the repo. It's free and tells me to keep maintaining it.