Goldspot AI Computer

Hardware Specs

8 GB

RAM (DDR4)

28 GB

USB Storage

ARM64

Architecture

~5W

Power Draw

$0.50

Monthly Cost

24/7

Always On

Connect to Your Goldspot

Local Network (SSH)

When you're on the same WiFi/LAN as the Goldspot:

ssh pi@goldspot.local

SSH key auth is configured — no password needed from your Mac.

By IP Address

If mDNS doesn't work, check your router's admin page for the Goldspot's IP, then:

ssh pi@<your-goldspot-ip>

Via Tailscale (from anywhere)

After setting up Tailscale:

ssh pi@goldspot

Using Claude Code

Authentication

This Goldspot uses OAuth authentication — no API key needed. Credentials are stored locally and tokens auto-refresh.

# Check auth status
claude auth status

# If auth fails, re-sync credentials
# from your authenticated Mac.
# See private setup notes for details.

Start Claude Code

# Launch interactive mode
claude

# Ask a one-off question
claude "explain this error" < error.log

# Work on a project
cd ~/my-project
claude

Common Claude Code Commands

/help          # Show all commands
/clear         # Clear conversation
/compact       # Compress context
/cost          # Show token usage
/init          # Create CLAUDE.md for a project

Local AI: Qwen 3.5

A fully local, private AI model running on the Goldspot via Ollama. No internet or API keys needed for inference — everything runs on-device.

4B

Parameters

3.4 GB

Model Size

~4.5 GB

RAM Usage

CPU

Inference

Run It

From your Mac, one command to start chatting:

# Interactive chat
ssh pi@goldspot.local -t "ollama run qwen3.5:4b"

# Or SSH in first, then run
ssh pi@goldspot.local
ollama run qwen3.5:4b

Type your prompt and press Enter. Type /bye to exit.

What to Expect

!

Speed: Expect ~2-5 tokens/sec on the Pi 4 CPU. Responses will stream in slowly but steadily. Good for code review, writing, and analysis — not for real-time chat.

!

First run: The first prompt takes ~15-30 seconds to load the model into RAM. Subsequent prompts in the same session are faster.

!

Quality: Qwen 3.5 4B punches above its weight — tool calling, reasoning, and vision built in. Comparable to much larger models from a year ago.

Use as an API

Ollama exposes an OpenAI-compatible API locally. From the Goldspot itself:

# Local API call
curl http://localhost:11434/api/generate \
  -d '{"model":"qwen3.5:4b","prompt":"hello"}'

Works with any OpenAI-compatible client. Access from other devices via SSH tunnel or Tailscale.

Manage Models

ollama list                    # Installed models
ollama pull qwen3.5:0.8b       # Faster, lighter
ollama pull qwen3.5:9b         # Smarter, uses ~7GB RAM
ollama rm qwen3.5:4b           # Remove a model
ollama ps                      # Running models

Remote Access with Tailscale

Tailscale creates a private VPN so you can access your Goldspot from anywhere — phone, laptop, coffee shop, anywhere.

Activate Tailscale

1

On the Goldspot, run:

sudo tailscale up

2

It will print a URL — open it in your browser and log in with Google/GitHub/etc.

3

Install Tailscale on your Mac too:

brew install --cask tailscale

4

Now connect from anywhere:

ssh pi@goldspot

Enable SSH via Tailscale (optional)

Let Tailscale handle SSH auth (no passwords needed):

sudo tailscale up --ssh

Quick Reference

System

htop          # System monitor
df -h         # Disk space
free -h       # Memory usage
uptime        # How long running
sudo reboot   # Restart

Development

node -v       # Node.js version
npm init -y   # New project
python3       # Python REPL
git clone     # Clone a repo
tmux          # Terminal mux

Networking

ip addr             # Show IPs
tailscale status    # VPN status
curl ifconfig.me    # Public IP
ping google.com     # Test internet

tmux (Terminal Multiplexer)

tmux                # New session
tmux ls             # List sessions
tmux a              # Reattach
Ctrl+B, D           # Detach
Ctrl+B, C           # New window

Project Ideas

Beginner

Personal API Server

Build a REST API that runs 24/7. Use Claude Code to scaffold it.

mkdir ~/api && cd ~/api
claude "create an express API
 with /health and /time
 endpoints"

Intermediate

Git Auto-Deployer

Webhook listener that auto-pulls and restarts your apps on push.

claude "build a webhook
 server that does git pull
 and pm2 restart on push"

Advanced

AI Agent Farm

Run multiple Claude Code agents on different repos simultaneously.

tmux new -s agent1
claude  # work on project A
# Ctrl+B, C
claude  # work on project B

Beginner

Cron Job Runner

Schedule scripts to run automatically — backups, scrapers, reports.

# Edit crontab
crontab -e

# Run script daily at 6am
0 6 * * * /home/pi/backup.sh

Architecture

Your Device (Anywhere) Goldspot (Home) ┌──────────────────────┐ ┌──────────────────────┐ │ Laptop / Phone │ │ Raspberry Pi 4 8GB │ │ │ │ │ │ Terminal ──────────────────────────── Claude Code │ │ VS Code ───────────────────────────── SSH Server │ │ Browser ───────────────────────────── Web Apps │ │ │ │ │ └──────────┬───────────┘ └──────────┬───────────┘ │ │ │ Tailscale VPN │ └───────── encrypted ───────────────┘ WiFi / LAN: ssh pi@goldspot.local Tailscale: ssh pi@goldspot

Maintenance

Keep Updated

# Update system packages
sudo apt update && sudo apt upgrade -y

# Update Claude Code
sudo npm update -g @anthropic-ai/claude-code

# Update Node.js
sudo apt update && sudo apt install -y nodejs

If Something Goes Wrong

# Can't SSH? Check if it's on the network
ping goldspot.local

# Lost the IP? Scan your network
arp -a | grep -v incomplete

# Need to restart?
# Unplug power, wait 10s, plug back in

# Nuclear option: re-flash the USB drive
# and start over (your data will be lost)

Device Info

Hostnamegoldspot

Login credentials and network details are kept private. Change your password with: passwd