GitHub Docs
Deployment Guides

Raspberry Pi

Deploy VibeCody on Raspberry Pi

Run VibeCody on a Raspberry Pi 3, 4, or 5 as a privacy-first, always-on AI coding assistant.

Setup time: 5–10 minutes Cost: $35 hardware Models: TinyLlama to Mistral 7B

Quick Start

curl -fsSL https://raw.githubusercontent.com/TuringWorks/vibecody/main/deploy/raspberry-pi/setup.sh | sh

Or with remote access:

cd vibecody/deploy/raspberry-pi
./setup.sh --tailscale

Model Recommendations

Pi Model RAM Recommended Model Performance
Pi 3 1 GB tinyllama:1.1b (or cloud provider) ~2 tok/s
Pi 4 (4 GB) 4 GB phi:2.7b ~5 tok/s
Pi 4 (8 GB) 8 GB mistral:7b ~3 tok/s
Pi 5 (8 GB) 8 GB mistral:7b or codellama:7b ~5 tok/s

Tip: For Pi 3, use a cloud AI provider (Claude, OpenAI, Gemini) instead of local models. The Pi runs VibeCLI itself; the AI runs in the cloud.

Step-by-Step

1. Run Setup Script

cd vibecody/deploy/raspberry-pi
./setup.sh

The script will:

  • Detect your Pi model and RAM
  • Install the aarch64 VibeCLI binary
  • Install Ollama (ARM build)
  • Create swap space if RAM < 4 GB
  • Install a systemd service
  • Copy an optimized config for your Pi model

2. Access VibeCody

# Local
curl http://localhost:7878/health

# From the REPL
vibecli

3. Remote Access (Optional)

Tailscale (recommended):

./setup.sh --tailscale
# Then from any device on your Tailscale network:
curl http://raspberrypi.your-tailnet.ts.net:7878/health

Cloudflare Tunnel:

./setup.sh --cloudflare
cloudflared tunnel login
cloudflared tunnel create vibecody
cloudflared tunnel route dns vibecody vibecody.example.com
cloudflared tunnel run vibecody

Optimized Configs

The setup script installs a config optimized for your Pi:

  • Pi 3 (config-pi3.toml): TinyLlama 1.1B, 1024 context, 10 max agent steps
  • Pi 4 (config-pi4.toml): Phi 2.7B, 2048 context, 25 max agent steps
  • Pi 5 (config-pi5.toml): Mistral 7B, 4096 context, 50 max agent steps

Service Management

sudo systemctl status vibecody    # Check status
sudo systemctl restart vibecody   # Restart
sudo journalctl -u vibecody -f    # View logs

Troubleshooting

Problem Solution
“Killed” during model load Not enough RAM — use a smaller model or add swap
Ollama slow to start First model load takes time; subsequent runs are faster
SD card fills up Models use 1–4 GB; use a 32 GB+ SD card or USB storage
Can’t reach from network Check firewall: sudo ufw allow 7878
Pi 3 too slow Use cloud AI providers — set --provider claude in config

What’s Next