Home Lab Setup

Home Lab Setup

Want to share your gaming PC’s LLM with your laptop, tablet, or other devices? This guide covers setting up Gambi as a permanent service in your home lab.

The Scenario

You have:

• Server: Desktop/gaming PC with GPU running Ollama 24/7
• Clients: Laptop, tablet, phone, or other devices that want to use the LLM

With Gambi, you can access your server’s LLM from anywhere on your home network.

Architecture

┌─────────────────────────────────────────────────────┐
│                   Home Network                       │
│                                                     │
│  ┌─────────────┐         ┌─────────────────────┐   │
│  │   Server    │         │      Clients        │   │
│  │  (GPU Box)  │◄───────►│  Laptop, Tablet...  │   │
│  │             │         │                     │   │
│  │ • Ollama    │         │ • SDK apps          │   │
│  │ • Hub       │         │ • Scripts           │   │
│  │ • Room      │         │ • Notebooks         │   │
│  └─────────────┘         └─────────────────────┘   │
│                                                     │
└─────────────────────────────────────────────────────┘

Server Setup

1. Install Gambi

curl -fsSL https://raw.githubusercontent.com/arthurbm/gambi/main/scripts/install.sh | bash

2. Create a Systemd Service

Create /etc/systemd/system/gambi-hub.service:

[Unit]
Description=Gambi Hub
After=network.target

[Service]
Type=simple
User=your-username
ExecStart=/usr/local/bin/gambi serve --port 3000 --mdns
Restart=always
RestartSec=10

[Install]
WantedBy=multi-user.target

Enable and start:

sudo systemctl enable gambi-hub
sudo systemctl start gambi-hub

3. Create a Persistent Room

You’ll want the same room code every time. Create a startup script:

/home/your-username/gambi-setup.sh

#!/bin/bash
# Wait for hub to be ready
sleep 5

# Create room (or use existing)
ROOM_CODE=$(gambi create 2>/dev/null | grep -oP 'Code: \K\w+')
echo "Room code: $ROOM_CODE"

# Join with local Ollama
gambi join $ROOM_CODE \
  --endpoint http://localhost:11434 \
  --model llama3 \
  --nickname homelab-gpu

Or add another systemd service for the participant:

[Unit]
Description=Gambi Participant
After=gambi-hub.service ollama.service
Requires=gambi-hub.service

[Service]
Type=simple
User=your-username
ExecStartPre=/bin/sleep 5
ExecStart=/usr/local/bin/gambi join YOURCODE --endpoint http://localhost:11434 --model llama3 --nickname homelab
Restart=always
RestartSec=10

[Install]
WantedBy=multi-user.target

4. Optional: Static Room Code

For a truly permanent setup, you can hardcode a room code. Check the hub logs for the generated code, or look at implementing room persistence (see Roadmap).

Client Setup

From Another Machine

import { createGambi } from "gambi-sdk";
import { generateText } from "ai";

const gambi = createGambi({
  roomCode: "YOUR_ROOM_CODE",
  hubUrl: "http://192.168.1.100:3000", // Your server's IP
});

const result = await generateText({
  model: gambi.any(),
  prompt: "Hello from my laptop!",
});

Using mDNS

If mDNS works on your network (most home networks), you can skip the IP:

const gambi = createGambi({
  roomCode: "YOUR_ROOM_CODE",
  // Hub discovered automatically via mDNS
});

From Scripts

Simple shell script to test:

#!/bin/bash
curl -X POST http://192.168.1.100:3000/rooms/YOURCODE/v1/responses \
  -H "Content-Type: application/json" \
  -d '{
    "model": "*",
    "input": "Hello!"
  }'

Security Considerations

Home Network Only

Gambi is designed for trusted local networks. Don’t expose it to the internet without additional security measures.

Firewall Rules

Allow traffic only from your local network:

# UFW example
sudo ufw allow from 192.168.1.0/24 to any port 3000

Future: Password Protection

Room password protection is on the roadmap. For now, rely on network isolation.

Multiple LLMs

You can run multiple models on the same server:

# Terminal 1: Join with llama3
gambi join YOURCODE \
  --endpoint http://localhost:11434 \
  --model llama3 \
  --nickname homelab-llama

# Terminal 2: Join with mistral (same Ollama, different model)
gambi join YOURCODE \
  --endpoint http://localhost:11434 \
  --model mistral \
  --nickname homelab-mistral

Then target specific models from clients:

// Use llama for code
const code = await generateText({
  model: gambi.model("llama3"),
  prompt: "Write a function...",
});

// Use mistral for text
const text = await generateText({
  model: gambi.model("mistral"),
  prompt: "Write an email...",
});

Monitoring

Check Hub Status

sudo systemctl status gambi-hub

View Logs

sudo journalctl -u gambi-hub -f

List Participants

gambi list

Troubleshooting

Service Won’t Start

1. Check logs: sudo journalctl -u gambi-hub -n 50
2. Verify gambi is installed: which gambi
3. Test manually: gambi serve --port 3000

Can’t Connect from Client

1. Check server firewall
2. Verify IP address: ip addr on server
3. Test connectivity: ping 192.168.1.100 from client
4. Test port: nc -zv 192.168.1.100 3000

Ollama Not Responding

1. Check Ollama is running: curl http://localhost:11434/v1/models
2. Verify model is pulled: ollama list
3. Check Ollama logs

What’s Next?

• Add more models as participants
• Set up monitoring with TUI
• Check Troubleshooting for common issues
• See the Architecture to understand how it all works