Raspberry Pi

Raspberry Pi is AI Butler’s favorite deployment target. A Pi 4 (4GB) handles a dozen messaging channels, smart home control, memory search, and the web chat simultaneously — all running offline if you pair it with Ollama on a beefier machine.

What You’ll Need

• Raspberry Pi 4 (4GB or 8GB) or Pi 5
• 32GB+ microSD card (or better, an SSD via USB 3)
• 64-bit Raspberry Pi OS (Bookworm or newer)
• Optional: a separate machine for Ollama if you want fully-local AI

Note

A Pi 3 technically works for messaging + memory + IoT, but it’s painfully slow if you try to run Ollama on it. For Pi 3, stick to cloud model APIs (Claude, GPT, Gemini).

1. Flash the OS

Use Raspberry Pi Imager and pick Raspberry Pi OS (64-bit). In the advanced options, set a hostname, enable SSH, and configure WiFi.

2. Install AI Butler

SSH into the Pi and download the ARM64 release:

cd /tmp
curl -sSL https://github.com/LumabyteCo/aibutler/releases/latest/download/aibutler_Linux_arm64.tar.gz | tar xz
sudo mv aibutler /usr/local/bin/

Verify:

aibutler version

3. Set Up a systemd Service

sudo useradd --system --home /var/lib/aibutler --create-home aibutler
sudo mkdir -p /etc/aibutler
sudo chown aibutler:aibutler /var/lib/aibutler /etc/aibutler

Create /etc/systemd/system/aibutler.service:

[Unit]
Description=AI Butler
After=network-online.target
Wants=network-online.target

[Service]
Type=simple
User=aibutler
Group=aibutler
WorkingDirectory=/var/lib/aibutler
ExecStart=/usr/local/bin/aibutler run --config /etc/aibutler/config.yaml
Restart=on-failure
RestartSec=5s

# Hardening
NoNewPrivileges=true
ProtectSystem=strict
ProtectHome=true
PrivateTmp=true
ReadWritePaths=/var/lib/aibutler
CapabilityBoundingSet=

[Install]
WantedBy=multi-user.target

Enable and start:

sudo systemctl daemon-reload
sudo systemctl enable --now aibutler
sudo systemctl status aibutler

4. Configure

sudo -u aibutler aibutler vault set anthropic_api_key sk-ant-...

Or use a fully-local setup pointing at Ollama on a separate machine:

/etc/aibutler/config.yaml

settings:
  model_provider: ollama

configurations:
  model:
    ollama:
      base_url: http://ollama-server.local:11434
      model: llama3.3

5. Access

The web chat is at http://aibutler.local:3377 (mDNS) or the Pi’s IP. Connect messaging channels via the guides in Channels.

Performance Tips

Use an SSD

MicroSD cards are the #1 bottleneck. A cheap USB 3 SSD enclosure with a 250GB SATA SSD transforms a Pi 4 — boot times, database writes, and backup performance all improve dramatically.

Tune SQLite

AI Butler already sets WAL mode, but on a Pi you can push it further:

configurations:
  database:
    cache_size_mb: 64       # default 32
    mmap_size_mb: 256       # default 64
    wal_autocheckpoint: 1000

Offload Ollama

Running an LLM on the Pi itself is slow. Instead, run Ollama on a more powerful machine (NAS, old desktop, Mac mini) and point AI Butler at it over your LAN. The Pi handles channels, memory, and tools; the other machine handles inference.

Schedule heavy tasks

If you run batch jobs (summaries, backups), schedule them overnight:

aibutler schedule create --name nightly-digest --cron "0 3 * * *" --prompt "..."

Power & Cooling

• Use the official Pi 4 USB-C power supply (3A) — undervoltage causes silent corruption
• Pi 5 needs the official 5A supply
• A passive heatsink is enough for messaging-only workloads; add a fan if you push the CPU with Ollama or media processing

Backup

Back up /var/lib/aibutler/aibutler.db nightly to another machine:

# On the Pi, via rsync over SSH
rsync -av /var/lib/aibutler/aibutler.db backup-host:/srv/backups/pi-aibutler/

Or use the built-in remote backup target pointed at a MinIO instance on your NAS.