Skip to content

Install it

palaia runs as a single container. If you already have Docker (Docker Engine on Linux, or Docker Desktop on macOS/Windows), this is the whole install.

On a Linux machine — a home server, a NAS, a Raspberry Pi:

Terminal window
docker run -d --name palaia \
--network host -e PALAIA_CLOUD_BIND=127.0.0.1 \
-v palaia_home:/data \
--restart unless-stopped \
--security-opt no-new-privileges:true --cap-drop ALL \
--read-only --tmpfs /tmp --tmpfs /run \
ghcr.io/byte5ai/palaia:stable

With Docker Desktop on macOS or Windows:

Terminal window
docker run -d --name palaia \
-p 8420:8420 \
-v palaia_home:/data \
--restart unless-stopped \
--security-opt no-new-privileges:true --cap-drop ALL \
--read-only --tmpfs /tmp --tmpfs /run \
ghcr.io/byte5ai/palaia:stable

The two differ in one line only. On Linux the container shares the machine’s own network, so its name, palaia.local, reaches your other devices (see below), and PALAIA_CLOUD_BIND keeps the entrance for cloud AI apps on the machine itself. Docker Desktop runs containers inside a hidden virtual machine, so there the port is handed out instead.

Release candidate: until 3.0.0 is final there is no stable image yet. Where a command or file on this page says ghcr.io/byte5ai/palaia:stable, use ghcr.io/byte5ai/palaia:beta for now.

The five extra flags close off what a non-root container process could otherwise still reach — nothing about the install changes if you leave them off, they simply make the container harder to escape from if something inside it were ever compromised.

That pulls the image, starts it, and keeps it running (and restarting after a reboot). Everything palaia saves lives in the palaia_home volume, so the container itself is disposable — you can remove and recreate it without losing anything.

Open http://palaia.local:8420/ in your browser from any device on the same network — a laptop, a phone — or http://localhost:8420/ on the machine itself (with Docker Desktop, always http://localhost:8420/). palaia’s setup opens by itself and walks you through, one step at a time:

  1. How you’ll use palaia — with ChatGPT and Claude (they reach palaia over a secure web address and sign in; the dashboard stays on your private network), or only on your own devices.
  2. Your first memory — a name such as work. What it’s for is optional; AI apps read it to pick the right memory.
  3. Your account — a username and a password of at least 12 characters. Needed for ChatGPT and Claude; optional when palaia stays private.
  4. Cloud access — only for ChatGPT and Claude: palaia’s public https:// address. palaia restarts once, asks you to sign in, then checks that address the way ChatGPT and Claude reach it.
  5. Connect an AI app — the steps for the app you use, ending when palaia receives its first request.

Setup remembers where you are: after a reload or a restart it continues at the first step that is still open. See Connect your AI for every app.

Installing on a Linux server instead? Get palaia running has the one command that also sets up Docker and Tailscale.

Prefer a config file, or already run other containers?

Section titled “Prefer a config file, or already run other containers?”

A ready-to-use file is at v3/deploy/docker-compose.yml in the palaia repository. It is set up like the Linux command above; its first lines say the three edits for Docker Desktop.

Terminal window
git clone https://github.com/byte5ai/palaia.git
cd palaia/v3/deploy && docker compose up -d

Or, a script that runs the right one of the two commands with a couple of sanity checks (Docker present, Docker actually running) and prints the address to open at the end:

Terminal window
curl -fsSL https://get.palaia.ai/docker | bash

None of these three paths is more “correct” than another — pick whichever fits how you already manage containers.

The container announces itself as palaia.local on your local network, so you don’t have to remember an IP address: http://palaia.local:8420/ opens it from your other devices, the way a Home Assistant box opens at its name and :8123. The :8420 stays part of the address — the container runs without any special rights, and the plain web port (80) needs them. Whether the name reaches your devices depends on how your container engine handles network traffic:

  • Docker on Linux — the Linux command and the compose file run the container on the machine’s own network, so palaia.local resolves for other devices on the same network that support it (most desktops and phones do). Some routers keep the name from crossing between separate networks, such as a guest Wi-Fi.
  • Docker Desktop (macOS/Windows) runs containers inside a hidden VM, so the announcement never reaches your actual network. palaia.local will not resolve here — use http://localhost:8420/ on that computer.

Either way, nothing about setup depends on this working: every startup prints the address that does work to the container’s log (docker logs palaia), and the wizard is reachable at that address regardless.

Phase one keeps this manual and honest about it — there is no silent in-place update:

Terminal window
docker pull ghcr.io/byte5ai/palaia:stable
docker compose up -d # or the equivalent docker run, after docker rm

Your data lives in the named volume, not the container, so this is safe — pulling a newer image and recreating the container keeps everything. The dashboard footer shows the version you’re running; when a newer one is published, a notice appears at the top of the dashboard with the steps for your kind of install.

If you install through a self-hosting app store (Umbrel, CasaOS, Runtipi, TrueNAS SCALE, and similar), check there first — those platforms often handle updates and networking for you, and a listing may not be live for every platform yet.