Skip to main content

Raspberry Pi Deployment

Deploy CyberMem to Raspberry Pi for edge AI memory.

Overview​

RPi deployment features:

  • Tailscale Funnel for zero-config public HTTPS
  • SQLite storage (SD card friendly)
  • Ollama local embeddings (ARM optimized)
  • Low resource footprint (~1GB RAM)

Requirements​

  • Raspberry Pi 4/5 (4GB+ RAM recommended)
  • Raspberry Pi OS (64-bit preferred)
  • Docker installed
  • Tailscale account (free tier works)

Quick Deploy​

# SSH into your Pi
ssh pi@raspberrypi.local

# Run on the Pi
npx @cybermem/cli init --rpi

# Or with HTTPS remote access
npx @cybermem/cli init --rpi --remote-access

Tailscale Setup​

1. Install Tailscale​

curl -fsSL https://tailscale.com/install.sh | sh
sudo tailscale up

2. Enable Funnel​

# Enable HTTPS funnel on port 8626
sudo tailscale funnel 8626

3. Get Your URL​

tailscale funnel status
# Shows: https://your-machine-name.ts.net

Architecture​

Internet ──► Tailscale Funnel ──► Traefik:8626 ──► Core API
                                      │
                                      └──► Dashboard:3000

Configuration​

RPi configuration in ~/.cybermem/.env:

# Target environment
CYBERMEM_TARGET=rpi

# Database
OM_DB_PATH=/data/openmemory.sqlite

# Embeddings
EMBEDDING_PROVIDER=ollama
OLLAMA_URL=http://ollama:11434

# Generated Security Token (required for remote access)
OM_TOKEN=sk-...

Remote MCP Configuration​

For AI clients connecting to your RPi, use mcp-remote (standard stdio-to-HTTP bridge):

{
  "mcpServers": {
    "cybermem-rpi": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://your-rpi.ts.net:8626/mcp",
        "--header",
        "X-API-Key:sk-your-token"
      ]
    }
  }
}

Platform Compatibility​

⚠️ 64-bit OS Required: Traefik and Prometheus require 64-bit Raspberry Pi OS. If using 32-bit userspace, use traefik:v2.11 image.

Check your platform:

uname -m          # Should show: aarch64
getconf LONG_BIT  # Should show: 64

Access dashboard via Tailscale:

Troubleshooting​

Container Crashes (Exit 159)​

Platform mismatch — 32-bit Docker cannot run arm64 images.

32-bit Userspace with 64-bit Kernel

If your kernel is aarch64 but Docker shows linux/arm, install 64-bit Docker static binary:

# Stop 32-bit Docker
sudo systemctl stop docker docker.socket

# Download arm64 Docker
curl -fsSL https://download.docker.com/linux/static/stable/aarch64/docker-27.5.1.tgz -o /tmp/docker.tgz

# Install to /usr/local/bin
sudo tar -xzf /tmp/docker.tgz -C /usr/local/bin --strip-components=1

# Start with new binary
sudo /usr/local/bin/dockerd &

# Verify
docker version  # Should show OS/Arch: linux/arm64

Note: After reboot, Docker will use 32-bit version again. Make permanent by updating systemd service.

Slow Embeddings​

Ollama on Pi is slower. Consider:

# Use smaller model
docker exec cybermem-ollama ollama pull all-minilm

SD Card Wear​

Move data to USB drive:

volumes:
  openmemory-data:
    driver_opts:
      device: /mnt/usb/cybermem

Next Steps​