Hardware

Mac Studio M3 Ultra Specifications

M3 Ultra

Apple Silicon

96GB

Unified Memory

8TB

SSD Storage

60

Core GPU

Software Bundle

Complete AI Software Inventory

Every tool, model, and framework installed on this Mac Studio. Filter by category to explore the full stack.

Full Runbook

From unboxing to always-on AI station

1 Pre-Setup Checklist Before powering on

Unbox & Inspect

Unbox the Mac Studio, check for physical damage. Connect to a display via HDMI or Thunderbolt.

Peripherals & Network

Connect keyboard and mouse (wired recommended for first setup). Connect Ethernet cable — preferred over Wi-Fi for stability during large model downloads.

Credentials Ready

Have these ready before starting:

Apple ID credentials
Anthropic API key (sk-ant-...)
OpenAI API key (sk-...)
GitHub Personal Access Token (ghp_...)
Discord Bot Token(s) (MTI...)
Discord Server ID

2 macOS Initial Setup ~10 min

First Boot Wizard

Power on, select language & region (Hong Kong). Skip Migration Assistant. Sign in with Apple ID. Create admin account with a strong password (16+ chars).

Privacy-first: Disable Analytics, Siri, and Screen Time. Enable Location Services for timezone only. Choose Dark appearance for always-on station.

Post-Wizard Configuration

Show commands

# Check macOS version
sw_vers

# Update to latest macOS
softwareupdate --list
softwareupdate --install -a

# Set hostname
sudo scutil --set HostName mac-studio-ai
sudo scutil --set LocalHostName mac-studio-ai
sudo scutil --set ComputerName "Mac Studio AI"

# Set timezone
sudo systemsetup -settimezone Asia/Hong_Kong

# Show hidden files and extensions
defaults write NSGlobalDomain AppleShowAllExtensions -bool true
defaults write com.apple.finder AppleShowAllFiles -bool true
defaults write com.apple.desktopservices DSDontWriteNetworkStores -bool true
killall Finder

3 System Security ~10 min

FileVault Disk Encryption

Encrypt the entire disk at rest. Record the recovery key offline in a safe location.

Show command

sudo fdesetup enable
# SAVE the recovery key that is displayed!

Firewall & Stealth Mode

Show commands

# Enable firewall
sudo /usr/libexec/ApplicationFirewall/socketfilterfw --setglobalstate on

# Enable stealth mode (don't respond to pings)
sudo /usr/libexec/ApplicationFirewall/socketfilterfw --setstealthmode on

# Verify
sudo /usr/libexec/ApplicationFirewall/socketfilterfw --getglobalstate

Energy Settings (24/7 Operation)

Prevent the Mac from sleeping — critical for always-on AI workloads.

Show commands

sudo pmset -a displaysleep 10       # Display sleeps after 10 min
sudo pmset -a sleep 0              # Computer never sleeps
sudo pmset -a disksleep 0          # Disk never sleeps
sudo pmset -a womp 1              # Wake on LAN
sudo pmset -a autorestart 1       # Auto-restart after power failure
sudo pmset -a powernap 1          # Background tasks during nap

# Verify
pmset -g

SSH Remote Access

Enable secure remote management with key-only authentication.

Show commands

# Enable SSH
sudo systemsetup -setremotelogin on

# Generate SSH key
ssh-keygen -t ed25519 -C "[email protected]"

# Harden SSH — disable password auth
sudo tee -a /etc/ssh/sshd_config.d/hardened.conf << 'EOF'
PasswordAuthentication no
ChallengeResponseAuthentication no
PermitRootLogin no
EOF

sudo launchctl stop com.openssh.sshd
sudo launchctl start com.openssh.sshd

4 Developer Foundations ~15 min

Xcode Command Line Tools

Show command

xcode-select --install

Homebrew

The essential package manager for macOS.

Show commands

# Install Homebrew
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"

# Add to PATH (Apple Silicon)
echo 'eval "$(/opt/homebrew/bin/brew shellenv)"' >> ~/.zprofile
eval "$(/opt/homebrew/bin/brew shellenv)"

# Verify
brew --version

Core CLI Tools

Install all essential command-line tools in one batch.

Show command

brew install \ git node [email protected] wget curl \ htop jq tree tmux \ ripgrep fd bat fzf gh \ cmake pkg-config

Git & GitHub Configuration

Show commands

git config --global user.name "MacAI HK"
git config --global user.email "[email protected]"
git config --global init.defaultBranch main
git config --global pull.rebase true

# Authenticate GitHub CLI
gh auth login

Node.js & Global npm Packages

Show commands

npm install -g \
  pnpm yarn typescript ts-node \
  nodemon pm2 openclaw

Python Setup

Show commands

pip3 install --upgrade pip setuptools wheel
pip3 install virtualenv pipx httpx requests python-dotenv pyyaml
pipx ensurepath

5 AI Runtime Stack ~15 min (excl. downloads)

Ollama (Local LLM Engine)

The core runtime that powers all local AI models. Optimised for Apple Silicon.

Show commands

# Install & start as background service
brew install ollama
brew services start ollama

# Verify
curl http://localhost:11434/api/tags

Ollama Optimisation for M3 Ultra

Tune Ollama for maximum performance on 96GB unified memory.

Show configuration

# Add to ~/.zshrc

export OLLAMA_HOST=0.0.0.0:11434           # Listen on all interfaces
export OLLAMA_NUM_PARALLEL=4               # 4 parallel requests
export OLLAMA_MAX_LOADED_MODELS=3          # Keep 3 models in RAM
export OLLAMA_KEEP_ALIVE=24h               # Models stay loaded 24h
export OLLAMA_FLASH_ATTENTION=1            # Flash attention for speed

# Apply & restart
source ~/.zshrc
brew services restart ollama

LM Studio (GUI)

Visual interface for running and testing LLMs. Provides an OpenAI-compatible API at localhost:1234.

Show command

brew install --cask lm-studio

# After launch: Settings → Server → Start server on launch
# OpenAI-compatible API at http://localhost:1234/v1

Whisper (Speech-to-Text)

Show commands

# Pre-built via Homebrew
brew install whisper-cpp

# Or build from source with CoreML (Neural Engine acceleration)
cd ~/Developer
git clone https://github.com/ggerganov/whisper.cpp.git
cd whisper.cpp
make clean
WHISPER_COREML=1 make -j

# Download models
bash ./models/download-ggml-model.sh base.en
bash ./models/download-ggml-model.sh large-v3

Stable Diffusion (Image Generation)

Show command

# GUI option (easiest)
brew install --cask diffusionbee

# Or Draw Things from Mac App Store (free, Apple Silicon optimised)

# CLI option with Metal GPU acceleration
cd ~/Developer
git clone https://github.com/leejet/stable-diffusion.cpp.git
cd stable-diffusion.cpp
mkdir build && cd build
cmake .. -DGGML_METAL=ON
cmake --build . --config Release -j

6 LLM Models ~2–4 hours (download)

Primary Model: Qwen3.5-122B-A10B (MoE)

The flagship local model for this build. 122 billion total parameters, but only activates 10 billion per request thanks to Mixture-of-Experts architecture — meaning you get 122B-class intelligence at 10B-class speed. Supports 256K context natively. Outperforms GPT-5 mini on tool-use benchmarks (72.2 vs 55.5 on BFCL-V4).

Show commands

# Primary — Qwen3.5 122B MoE (Q4_K_M ~77GB)
ollama pull qwen3.5:122b-a10b    # ~77GB — our main model

# Fast fallback (for simple tasks when 122B is busy)
ollama pull qwen3.5:9b           # ~5GB — lightweight, same family

# Embedding model (required for RAG / document search)
ollama pull nomic-embed-text      # ~274MB

Specialist Models (swap in as needed)

These cannot run alongside the 122B model — Ollama will unload the 122B to free RAM. Keep them downloaded for on-demand use.

Show commands

# Coding-focused
ollama pull qwen2.5-coder:32b    # ~18GB — excellent for code

# Chinese + English bilingual
ollama pull qwen3.5:27b          # ~16GB — dense, fast

# Reasoning
ollama pull deepseek-r1:32b      # ~18GB — strong reasoning

# Vision / Multimodal
ollama pull llama3.2-vision:11b   # ~7GB — image understanding

# Additional embedding
ollama pull mxbai-embed-large     # ~670MB

# Verify all
ollama list

96GB RAM Budget — Interactive Model Explorer

Select any Qwen 3.5 model below to see how it fits in the 96GB unified memory. RAM allocation, context boundaries, and architecture diagrams update in real time.

Compare other RAM tiers (128–512 GB)

Architecture

MoE

Total Params

122B

Active / Token

10B

VRAM (Q4_K_M)

77 GB

Expert Config

60E / 4A

Fits 96GB?

Tight fit

Does not fit.

Context Window Boundary

Context Length	KV Cache RAM	Free for Agents	Verdict

Architecture Diagram

When to Use Which Context Size

Use Case	Context	Config

For 128K+ content: Don't increase context — use RAG (Retrieval-Augmented Generation) via PrivateGPT or AnythingLLM instead. RAG chunks documents into embeddings stored on disk, then retrieves only relevant snippets into a small context window. This keeps RAM stable while handling unlimited document sizes.

Ollama Context Configuration

Create persistent Modelfiles with fixed context sizes for different use cases.

Show recommended Ollama context configuration

# Set default context to 16K (balanced for agents + model)
ollama run qwen3.5:122b-a10b --ctx-size 16384

# Or create a persistent Modelfile with fixed context
cat > ~/Modelfile-qwen122b << 'EOF'
FROM qwen3.5:122b-a10b
PARAMETER num_ctx 16384
PARAMETER num_gpu 999
EOF

ollama create qwen122b-16k -f ~/Modelfile-qwen122b

# For long-document mode (stop agents first!)
cat > ~/Modelfile-qwen122b-long << 'EOF'
FROM qwen3.5:122b-a10b
PARAMETER num_ctx 65536
PARAMETER num_gpu 999
EOF

ollama create qwen122b-64k -f ~/Modelfile-qwen122b-long

# Enable KV cache quantisation to squeeze more context
# (halves KV cache RAM — 32K context uses only ~1.6GB)
cat > ~/Modelfile-qwen122b-q8kv << 'EOF'
FROM qwen3.5:122b-a10b
PARAMETER num_ctx 32768
PARAMETER cache_type q8_0
EOF

ollama create qwen122b-32k-q8kv -f ~/Modelfile-qwen122b-q8kv

7 AI Applications & GUIs ~15 min

Docker Desktop

Required for running Open WebUI, n8n, and other containerised services.

Show command

brew install --cask docker
# Launch Docker Desktop, then enable "Start on login" in settings

Open WebUI (ChatGPT-like Interface)

A beautiful web interface for chatting with your local LLMs — like ChatGPT, but 100% private.

Show command

docker run -d \
  --name open-webui \
  -p 3000:8080 \
  -v open-webui:/app/backend/data \
  --add-host=host.docker.internal:host-gateway \
  -e OLLAMA_BASE_URL=http://host.docker.internal:11434 \
  --restart always \
  ghcr.io/open-webui/open-webui:main

# Access at http://localhost:3000
# Create admin account on first visit

PrivateGPT (Document AI / RAG)

Upload documents and ask questions — entirely local, no cloud.

Show commands

cd ~/Developer
git clone https://github.com/zylon-ai/private-gpt.git
cd private-gpt

# Install with Ollama support
poetry install --extras "ui llms-ollama embeddings-ollama vector-stores-qdrant"

# Start
PGPT_PROFILES=local make run

# Access at http://localhost:8001

AnythingLLM (Simple RAG Alternative)

Show command

brew install --cask anythingllm

# After launch:
# 1. Set LLM provider → Ollama (localhost:11434)
# 2. Set Embedding → Ollama (nomic-embed-text)
# 3. Create workspace → upload documents

8 Agent Framework (OpenClaw) ~20 min

Initialise OpenClaw Workspace

Show commands

openclaw init --workspace ~/Documents/GitHub/MacAI

# Creates:
# ├── SOUL.md           — Agent identity
# ├── HEARTBEAT.md      — Periodic tasks
# ├── .openclaw/
# │   ├── config.yml    — Main configuration
# │   ├── routing.yml   — Bot-to-bot communication
# │   └── secrets.enc   — Encrypted API keys
# └── agents/
#     ├── biz-manager/SOUL.md
#     ├── dev-agent/SOUL.md
#     └── alert-bot/SOUL.md

Configure API Keys

Show commands

openclaw config set ANTHROPIC_API_KEY "sk-ant-api03-..."
openclaw config set OPENAI_API_KEY "sk-..."
openclaw config set GITHUB_TOKEN "ghp_..."
openclaw config set DISCORD_BOT_TOKEN "MTI..."
openclaw config set DISCORD_SERVER_ID "123456789012345678"

# Verify
openclaw config list

SOUL.md — Agent Identity

Show example

# SOUL.md — MacAI Assistant

Name: MacAI Assistant
Role: Business operations AI for Hong Kong SME
Language: Bilingual (English / 繁體中文)
Tone: Professional, concise, helpful

Boundaries:
  - NEVER share confidential client data
  - NEVER execute financial transactions without human approval
  - ALWAYS escalate legal questions to human
  - ALWAYS log all actions for audit trail

HEARTBEAT.md — Periodic Tasks

Show example

# HEARTBEAT.md — Automated Check-ins

interval: 15m
tasks:
  - Check Discord #alerts for new messages
  - Review pending GitHub PRs
  - Monitor system health (RAM, CPU, disk)

daily (09:00 HKT):
  - Generate morning briefing → #daily-briefing
  - Summarise overnight Discord activity

weekly (Monday 09:00):
  - Generate weekly summary report
  - Check for model updates (ollama)

Bot Routing Configuration

Show configuration

# .openclaw/routing.yml

peers:
  BizManager:
    can_send_to: [DevAgent, AlertBot, ContentBot]
    channels: [ai-assistant, daily-briefing]
    tools: [discord, github-issues]
  DevAgent:
    can_send_to: [BizManager, AlertBot]
    channels: [macai-dev]
    tools: [github, claude-code, terminal]
  AlertBot:
    can_send_to: [BizManager]
    channels: [alerts]
  ContentBot:
    can_send_to: [BizManager]
    channels: [content-drafts]

Deploy All Agents

Show commands

openclaw agent start --name "BizManager" --soul agents/biz-manager/SOUL.md
openclaw agent start --name "DevAgent" --soul agents/dev-agent/SOUL.md
openclaw agent start --name "AlertBot" --soul agents/alert-bot/SOUL.md
openclaw agent start --name "ContentBot" --soul agents/content-bot/SOUL.md

# Check running agents
openclaw agent list

9 Discord Bot Setup ~15 min

Create Discord Application

For each bot (BizManager, DevAgent, AlertBot, ContentBot):

Go to discord.com/developers/applications
Click New Application → name it
Bot tab → Add Bot → copy Token
Enable: Message Content Intent, Server Members Intent, Presence Intent
OAuth2 → Scopes: bot, applications.commands
Permissions: Send/Read/Manage Messages, Embed Links, Attach Files, Read History, Add Reactions
Copy invite URL → open in browser → authorise to server

Discord Server Channel Structure

📋 OPERATIONS
├── #daily-briefing     — Morning reports (BizManager)
├── #ai-assistant       — General client queries
├── #task-board         — Active task tracking
└── #alerts             — System alerts (AlertBot)

💻 DEVELOPMENT
├── #macai-dev          — Dev tasks (DevAgent)
├── #code-review        — Automated PR reviews
└── #deployments        — Deployment notifications

📝 CONTENT
├── #content-drafts     — Draft content (ContentBot)
└── #content-approved   — Approved content

⚙️ ADMIN
├── #bot-logs           — Bot activity logs
└── #system-health      — RAM/CPU/disk monitoring

10 Coding & Development Tools ~10 min

Claude Code CLI

Show commands

npm install -g @anthropic-ai/claude-code

# Enable team agent mode
echo 'export CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1' >> ~/.zshrc
source ~/.zshrc

# Verify
claude --version

VS Code + Extensions

Show commands

brew install --cask visual-studio-code

# Essential extensions
code --install-extension GitHub.copilot
code --install-extension GitHub.copilot-chat
code --install-extension ms-python.python
code --install-extension dbaeumer.vscode-eslint
code --install-extension esbenp.prettier-vscode
code --install-extension eamodio.gitlens
code --install-extension continue.continue

Continue.dev (Local AI Code Assistant)

Open-source code assistant that uses your local Ollama models.

Show configuration

# ~/.continue/config.json
{
  "models": [
    {
      "title": "Qwen Coder 32B",
      "provider": "ollama",
      "model": "qwen2.5-coder:32b",
      "apiBase": "http://localhost:11434"
    },
    {
      "title": "Qwen3.5 122B MoE",
      "provider": "ollama",
      "model": "qwen3.5:122b-a10b",
      "apiBase": "http://localhost:11434"
    }
  ],
  "tabAutocompleteModel": {
    "title": "Qwen Coder 7B",
    "provider": "ollama",
    "model": "qwen2.5-coder:7b"
  }
}

Terminal Enhancements

Show commands

brew install --cask iterm2

# Oh My Zsh
sh -c "$(curl -fsSL https://raw.githubusercontent.com/ohmyzsh/ohmyzsh/master/tools/install.sh)"

# Powerlevel10k theme
git clone --depth=1 https://github.com/romkatv/powerlevel10k.git \
  ${ZSH_CUSTOM:-$HOME/.oh-my-zsh/custom}/themes/powerlevel10k

# Plugins
git clone https://github.com/zsh-users/zsh-autosuggestions \
  ${ZSH_CUSTOM:-~/.oh-my-zsh/custom}/plugins/zsh-autosuggestions
git clone https://github.com/zsh-users/zsh-syntax-highlighting.git \
  ${ZSH_CUSTOM:-~/.oh-my-zsh/custom}/plugins/zsh-syntax-highlighting

11 Automation & Workflows ~10 min

n8n (Workflow Automation)

Visual workflow builder — connects Discord, GitHub, Ollama, email, and more.

Show command

docker run -d \
  --name n8n \
  -p 5678:5678 \
  -v n8n_data:/home/node/.n8n \
  --restart always \
  n8nio/n8n

# Access at http://localhost:5678

PM2 Process Manager

Keeps all agents running 24/7 with auto-restart on crash or reboot.

Show ecosystem config

// ecosystem.config.js
module.exports = {
  apps: [
    {
      name: 'openclaw-biz',
      script: 'openclaw',
      args: 'agent start --name BizManager --soul agents/biz-manager/SOUL.md',
      restart_delay: 5000,
      max_restarts: 10,
      autorestart: true
    },
    {
      name: 'openclaw-dev',
      script: 'openclaw',
      args: 'agent start --name DevAgent --soul agents/dev-agent/SOUL.md',
      restart_delay: 5000,
      autorestart: true
    },
    {
      name: 'openclaw-alert',
      script: 'openclaw',
      args: 'agent start --name AlertBot --soul agents/alert-bot/SOUL.md',
      restart_delay: 5000,
      autorestart: true
    }
  ]
};

# Start all, save, and enable boot startup
pm2 start ecosystem.config.js
pm2 save
pm2 startup

12 Always-On Configuration ~10 min

Auto-Start Verification

After a reboot, all services should come back automatically:

Show verification commands

# Verify after reboot
brew services list         # Ollama → started
docker ps                   # open-webui, n8n → running
pm2 list                    # All agents → online
ollama list                 # Should respond
curl localhost:3000          # Open WebUI → 200 OK
curl localhost:5678          # n8n → 200 OK

Health Check Script (launchd)

Runs every 5 minutes to monitor and auto-restart failed services.

Show script

#!/bin/bash — healthcheck.sh
# Runs every 5 min via launchd

TIMESTAMP=$(date '+%Y-%m-%d %H:%M:%S')

# Check Ollama
if curl -s http://localhost:11434/api/tags > /dev/null 2>&1; then
  echo "[$TIMESTAMP] ✓ Ollama: Running"
else
  echo "[$TIMESTAMP] ✗ Ollama: DOWN — restarting"
  brew services restart ollama
fi

# Check Docker containers
if curl -s http://localhost:3000 > /dev/null 2>&1; then
  echo "[$TIMESTAMP] ✓ Open WebUI: Running"
else
  docker restart open-webui
fi

# Check PM2
if pm2 jlist > /dev/null 2>&1; then
  echo "[$TIMESTAMP] ✓ PM2: Running"
else
  pm2 resurrect
fi

Show launchd plist

<!-- ~/Library/LaunchAgents/com.macai.healthcheck.plist -->

<?xml version="1.0" encoding="UTF-8"?>
<plist version="1.0">
<dict>
    <key>Label</key>
    <string>com.macai.healthcheck</string>
    <key>ProgramArguments</key>
    <array>
        <string>/bin/bash</string>
        <string>/Users/admin/scripts/healthcheck.sh</string>
    </array>
    <key>StartInterval</key>
    <integer>300</integer>
    <key>RunAtLoad</key>
    <true/>
</dict>
</plist>

# Load:
launchctl load ~/Library/LaunchAgents/com.macai.healthcheck.plist

13 Monitoring & Maintenance Ongoing

Monitoring Commands

Show commands

htop                         # Real-time process monitor
memory_pressure               # RAM pressure level
df -h                         # Disk space
du -sh ~/.ollama/models/*      # Model sizes
ollama list                    # Loaded models
docker stats                   # Container resources
pm2 monit                      # Agent dashboard

Regular Maintenance Schedule

Frequency	Task	Command
Weekly	Update Homebrew packages	`brew update && brew upgrade && brew cleanup`
Weekly	Update Ollama models	`ollama pull qwen3.5:122b-a10b`
Weekly	Update npm globals	`npm update -g`
Monthly	Clean up Docker	`docker system prune -f`
Monthly	Clear old logs	`rm -f /tmp/macai-healthcheck*.log`
Monthly	macOS updates	`softwareupdate --list`

Complete Software List

Everything installed on this machine

CLI Tools (via Homebrew)

Software	Command	Purpose
Git	`brew install git`	Version control
Node.js	`brew install node`	JavaScript runtime
Python 3.12	`brew install [email protected]`	Python runtime
wget / curl	`brew install wget curl`	Download tools
htop	`brew install htop`	Process monitor
jq	`brew install jq`	JSON processor
tmux	`brew install tmux`	Terminal multiplexer
ripgrep	`brew install ripgrep`	Fast search
fd	`brew install fd`	Fast find
bat	`brew install bat`	Better cat
fzf	`brew install fzf`	Fuzzy finder
gh	`brew install gh`	GitHub CLI
tree	`brew install tree`	Directory viewer
cmake / pkg-config	`brew install cmake pkg-config`	Build tools

AI Runtimes & Applications

Software	Install Method	Purpose
Ollama	`brew install ollama`	Local LLM engine
LM Studio	`brew install --cask lm-studio`	LLM GUI + API
whisper.cpp	`brew install whisper-cpp`	Speech-to-text
DiffusionBee	`brew install --cask diffusionbee`	Image generation
Open WebUI	Docker container	ChatGPT-like UI
PrivateGPT	Git clone + Poetry	Document Q&A (RAG)
AnythingLLM	`brew install --cask anythingllm`	Simple RAG
n8n	Docker container	Workflow automation

LLM Models (via Ollama)

Model	Size	Purpose
Qwen3.5-122B-A10B	~77 GB	Primary model — 122B MoE, 10B active
Qwen3.5 9B	~5 GB	Fast fallback, same family
Qwen3.5 27B	~16 GB	Dense bilingual (swap-in)
Qwen 2.5 Coder 32B	~18 GB	Code generation (swap-in)
DeepSeek R1 32B	~18 GB	Reasoning (swap-in)
Llama 3.2 Vision 11B	~7 GB	Image understanding (swap-in)
Nomic Embed Text	~274 MB	Document embeddings (RAG)
MxBai Embed Large	~670 MB	High-quality embeddings (RAG)

Development Tools

Software	Install Method	Purpose
VS Code	`brew install --cask visual-studio-code`	Code editor
Claude Code	`npm install -g @anthropic-ai/claude-code`	AI coding CLI
OpenClaw	`npm install -g openclaw`	Agent orchestration
PM2	`npm install -g pm2`	Process manager
Docker Desktop	`brew install --cask docker`	Container runtime
iTerm2	`brew install --cask iterm2`	Better terminal
Rectangle	`brew install --cask rectangle`	Window management
Stats	`brew install --cask stats`	Menu bar monitor

Quick Start

One-liner install (after Homebrew)

Copy and paste this to install the entire stack in one batch.

# CLI tools
brew install git node [email protected] wget htop jq tree tmux ripgrep fd bat fzf gh cmake pkg-config

# AI stack + Apps
brew install ollama
brew install --cask docker lm-studio diffusionbee anythingllm visual-studio-code iterm2 rectangle stats

# Dev tools
npm install -g pnpm typescript ts-node nodemon pm2 openclaw @anthropic-ai/claude-code n8n

# Start Ollama
brew services start ollama

# Download primary model (Qwen3.5 122B MoE — ~77GB, takes ~1-2 hours)
ollama pull qwen3.5:122b-a10b && ollama pull qwen3.5:9b && ollama pull nomic-embed-text

Mac Studio M3 Ultra 96GBComplete Setup Guide

Mac Studio M3 Ultra Specifications

Complete AI Software Inventory

From unboxing to always-on AI station

Unbox & Inspect

Peripherals & Network

Credentials Ready

First Boot Wizard

Post-Wizard Configuration

FileVault Disk Encryption

Firewall & Stealth Mode

Energy Settings (24/7 Operation)

SSH Remote Access

Xcode Command Line Tools

Homebrew

Core CLI Tools

Git & GitHub Configuration

Node.js & Global npm Packages

Python Setup

Ollama (Local LLM Engine)

Ollama Optimisation for M3 Ultra

LM Studio (GUI)

Whisper (Speech-to-Text)

Stable Diffusion (Image Generation)

Primary Model: Qwen3.5-122B-A10B (MoE)

Specialist Models (swap in as needed)

96GB RAM Budget — Interactive Model Explorer

Context Window Boundary

Architecture Diagram

When to Use Which Context Size

Ollama Context Configuration

Docker Desktop

Open WebUI (ChatGPT-like Interface)

PrivateGPT (Document AI / RAG)

AnythingLLM (Simple RAG Alternative)

Initialise OpenClaw Workspace

Configure API Keys

SOUL.md — Agent Identity

HEARTBEAT.md — Periodic Tasks

Bot Routing Configuration

Deploy All Agents

Create Discord Application

Discord Server Channel Structure

Claude Code CLI

VS Code + Extensions

Continue.dev (Local AI Code Assistant)

Terminal Enhancements

n8n (Workflow Automation)

PM2 Process Manager

Auto-Start Verification

Health Check Script (launchd)

Monitoring Commands

Regular Maintenance Schedule

Everything installed on this machine

CLI Tools (via Homebrew)

AI Runtimes & Applications

LLM Models (via Ollama)

Development Tools

One-liner install (after Homebrew)

Want us to set this up for you?

Mac Studio M3 Ultra 96GB
Complete Setup Guide