Just want AI running right now? Three commands. Then come back and do the full setup.
-
✓
Install Ollama
curl -fsSL https://ollama.com/install.sh | sh
Quick
-
✓
Pull a capable free model
ollama pull phi3.5:3.8b-mini-instruct-q4_K_M
Quick
-
✓
Talk to your AI
ollama run phi3.5... "Hello. Are you mine now?"
Quick
-
✓
Cancel a subscription
You just saved $20/month. Do the full setup when you're ready.
Win
Get your node. This is your foundation — every other step runs on this hardware.
-
✓
Order Raspberry Pi 5 (8GB)
~$80. The 8GB model only — 4GB won't run models comfortably.
Order First
-
✓
Order NVMe SSD (512GB) + M.2 HAT+
SD cards are too slow for inference workloads. NVMe is mandatory.
Order First
-
✓
Order active cooler + 27W USB-C power supply
Official Pi 5 active cooler. Standard USB-C is not enough — must be 27W.
Order First
-
✓
Flash Raspberry Pi OS Lite (64-bit) to NVMe
Use Raspberry Pi Imager. Enable SSH. Set hostname to something memorable.
Hands-on
-
✓
Verify NVMe boot (not SD card)
Run: df -h and confirm /dev/nvme0 not /dev/mmcblk0
Hands-on
-
✓
Run full system update
sudo apt update && sudo apt full-upgrade -y && sudo reboot
Quick
Get AI running on your hardware. This is the core capability everything else builds
on.
-
✓
Install core dependencies
git curl wget build-essential cmake python3 python3-pip python3-venv libopenblas-dev nodejs
docker.io
Hands-on
-
✓
Install Ollama
curl -fsSL https://ollama.com/install.sh | sh
Quick
-
✓
Pull primary model (Phi-3.5 Mini)
ollama pull phi3.5:3.8b-mini-instruct-q4_K_M — 2.4GB, best reasoning for its size
Quick
-
✓
Pull fast model (Llama 3.2 1B)
ollama pull llama3.2:1b-instruct-q8_0 — for background tasks and heartbeats
Quick
-
✓
Verify inference works
ollama run phi3.5:3.8b-mini-instruct-q4_K_M "What is 2+2? Answer in one word."
Quick
-
✓
Compile llama.cpp (start it, walk away)
cmake -B build -DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=OpenBLAS && cmake --build build -j4
Hands-on
-
✓
Install Petals for distributed inference
python3 -m venv ~/petals-env && pip install petals
Hands-on
Connect your node to your community's private encrypted network. Do this before
sharing any services.
-
✓
Create Tailscale account (one per person)
tailscale.com — free tier supports 100 devices. Each person creates their own.
Quick
-
✓
Install Tailscale on your Pi
curl -fsSL https://tailscale.com/install.sh | sh && sudo tailscale up
Quick
-
✓
Verify your Tailscale IP and hostname
tailscale ip — you should see a 100.x.x.x address and a *.ts.net hostname
Quick
-
✓
Ping test between all nodes in your group
ping alice-node.tail12345.ts.net from bob's node. If it works, the mesh is live.
Hands-on
-
✓
Set up Ansible on your machine (not the Pi)
pip install ansible — this lets you push commands to all nodes at once
Hands-on
-
✓
Create inventory.ini with all node hostnames
List every node's Tailscale hostname. Test with: ansible all -i inventory.ini -m ping
Hands-on
Layer community services on top of your mesh. Each one replaces a subscription. Do
them in order — CasaOS first.
-
✓
Install CasaOS (your home OS dashboard)
curl -fsSL https://get.casaos.io | sudo bash — then open http://your-tailscale-ip in a
browser
Quick
-
✓
Install NextCloud AIO from CasaOS app store
Files, calendar, contacts, documents. One click in CasaOS.
Quick
-
✓
Install NextCloud client on your computer and phone
nextcloud.com/clients — configure to point at your Pi's Tailscale address
Quick
-
✓
Install Navidrome for music
Point it at your music library. Install Symfonium (Android) or Amperfy (iOS).
Quick
-
✓
Install Matrix/Synapse for encrypted chat
From CasaOS, or via Docker. Everyone gets Element app, points at your homeserver.
Hands-on
-
✓
Install SearXNG for anonymous search
docker run -d -p 8888:8080 searxng/searxng — 30 minutes including Docker setup
Quick
-
✓
Install Pi-hole + Unbound for DNS
curl -sSL https://install.pi-hole.net | bash — blocks ads network-wide
Hands-on
-
✓
Install Vaultwarden for password management
docker run -d -p 8082:80 vaultwarden/server — use any Bitwarden client
Quick
-
✓
Install Jellyfin for your video library
From CasaOS app store. Add your media folder. Stream from any device.
Quick
-
✓
Install Home Assistant for smart home
From CasaOS. 3,000+ device integrations, all running locally with zero cloud.
Hands-on
Pool your nodes into a distributed AI pod. This is where the community capability
exceeds what any individual can do alone.
-
✓
Start Petals server on each node
python3 -m petals.cli.run_server meta-llama/Meta-Llama-3.1-8B-Instruct --num_blocks 8
Hands-on
-
✓
Verify collective inference pool is working
Nodes should find each other via DHT. Check with: python3 -c 'from petals import
AutoDistributedModelForCausalLM; ...'
Hands-on
-
✓
Set up OpenClaw gateway (one per person)
npm install -g openclaw && openclaw setup — point at local Ollama
Hands-on
-
✓
Configure model routing: fast / heavy
heartbeat + subagent → 1B model. heavy tasks → Petals collective pool.
Hands-on
-
✓
Run openclaw doctor — fix all warnings
Never connect channels until doctor shows clean. Gateway bound to Tailscale IP only.
Hands-on
-
✓
Write your SOUL.md
Your AI's identity, values, and explicit 'never do this' boundaries. Stored in
~/.openclaw/SOUL.md
Quick
-
✓
Fine-tune a model on your community's knowledge (optional)
Collect 500+ instruction-response pairs from your notes, docs, Slack exports. Run Unsloth.
Hands-on