Skip to main content
Version: Next (nightly)

Raspberry Pi

A Pi 4 or 5 with 4GB or more runs Cobblr comfortably for a household or a single person. The setup is the same two files and docker compose up -d as everywhere else. The Pi-specific parts are the OS, the storage, and knowing the first pull takes a little longer.

Use a 64-bit OS​

Run a 64-bit Raspberry Pi OS (or another arm64 Linux), and confirm uname -m reports aarch64 before you begin.

  • The published images are multi-arch, so on an arm64 host Docker pulls the arm64 build automatically. Nothing compiles on the Pi. The first run just pulls and starts.
  • A 32-bit OS will not pull the right image.

Put the database on an SSD, not the microSD​

This is the one that matters for a Pi. Put your install folder on the SSD's mount point and run docker compose up -d from there, so every bind mount lands on the SSD.

  • Keep the whole ./data/ tree, and especially ./data/db, on a USB SSD. A microSD card is slow and wears out under a database's write pattern.
  • A cheap SATA SSD in a USB 3.0 UASP enclosure is a solid setup.

Bring it up​

The flow is identical to Install. Generate your files in the setup builder, copy them onto the Pi, and:

docker compose up -d
  • The first pull is slower over a Pi's network and storage than on a desktop, so give it a few minutes.
  • Updates are the usual docker compose pull && docker compose up -d, or the updater if you would rather it happened by itself.
  • For HTTPS on a Pi, Tailscale is the least-effort path and is what lets the phone camera work.

Resource notes​

  • Memory is the thing to watch, not CPU. Postgres, the api, and the proxy sit within a 4GB Pi with room to spare, since the AI and embedding work runs on the provider you connect rather than on the Pi.
  • For a busy group rather than a household, a small x86 box has more headroom for the money. See the sizing notes.