.__         .__        
___________  |  | _____  |__|____   
\____ \__  \ |  | \__  \ |  \__  \  
|  |_> > __ \|  |__/ __ \|  |/ __ \_
|   __(____  /____(____  /__(____  /
|__|       \/          \/        \/ 

palaia is a hub you run yourself: on a server in your office, in your own cloud account, or on a machine at home. Every AI tool you use connects to it once: Claude, ChatGPT, Codex, Gemini, and any other tool that speaks MCP, the open standard AI tools use to reach outside data and tools. From then on they share one memory and one set of tools, and they can hand work to each other. Your data stays in plain files on hardware you control.

Think Home Assistant, for your AI tools.

• Your AI tools remember, together. Tell one assistant about a decision and the others know it too. Your memory is a folder of plain Markdown notes you can open in any editor (Obsidian opens it as-is), search, and back up like any other files.
• Set up a tool once, use it everywhere. Install an add-on in palaia and every connected AI can use it. No more pasting the same configuration into five apps.
• Your agents can work as a team. An AI session on your laptop can hand a task to one on your server, with the context attached, and find out who is already working on what.
• Runs where you decide. A server in your office, a machine in your cloud account, a NAS, or a Raspberry Pi. No account with us, no subscription, no data leaving your infrastructure unless you say so.
• Safe by default. Your dashboard stays on your private network. ChatGPT and Claude reach only palaia's AI connection, over HTTPS, and sign in with your account. Back everything up with one click.

You need Docker. That is the whole prerequisite list.

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:beta

Then open http://palaia.local:8420 in a browser on any device on the same network (or http://<your-machine>:8420 if your network blocks the name). That is the command for a Linux machine; with Docker Desktop on a Mac or Windows PC, replace the --network host … line with -p 8420:8420 and open http://localhost:8420. palaia's setup opens by itself and takes it from there: your first memory, your account, cloud access if you want it, your first AI app. No config files, no second command.

Once 3.0.0 is final, use :stable instead of :beta. Updating later is one click in the dashboard.

Other ways to install

Release candidate: the files and packages in this table pin ghcr.io/byte5ai/palaia:stable, which does not exist until 3.0.0 is final. Until then, change :stable to :beta in whatever you use.

Stuck? See Your first shared memory for a walkthrough with screenshots, or troubleshooting.

Claude Code · Claude Desktop · claude.ai · ChatGPT · Codex · Gemini CLI · Grok · LM Studio · any other MCP-compatible tool

Each one has a short connect guide. Claude Desktop gets a one-click bundle: download, click, connected.

Everything is managed from a dashboard in your browser:

Overview	Memories
AI apps	Add-ons

• Documentation: getting started, connecting each tool, backup and restore, troubleshooting.
• How it works: the full feature list, the architecture, the test evidence behind the claims, and when palaia is not the right fit.
• Security and the threat model.
• What is left before 3.0.0 and what shipped.
• For contributors: CONTRIBUTING.md and v3/README.md for the dev setup.

palaia v2 is the Python package at the root of this repository. It is retired: it gets no new features, v3 replaces it, and only a critical fix (security, data loss, a broken release) can still land. It stays installable, and its documentation is unchanged. When you are ready, one command imports your v2 knowledge into v3, and the migration guide covers what changes and how to roll back.

• Questions, bugs, ideas: open an issue.
• Security vulnerabilities: please do not open a public issue. Use GitHub's private vulnerability reporting instead. Details in v3/SECURITY.md.

MIT · © 2026 byte5 GmbH