Your first shared memory
This is the moment palaia is for: two different AI tools, given the same question, give the same answer — because they share one memory instead of starting from nothing every time. Fifteen minutes end to end, most of it waiting for downloads.
1. Install (about five minutes)
Section titled “1. Install (about five minutes)”Follow Install it if you haven’t already: one command, then the browser-based setup creates your first memory. This walkthrough assumes you named it something you’ll recognize, like “work” or “home”.
2. Connect your first AI tool
Section titled “2. Connect your first AI tool”Pick one from Connect your AI — Claude Code CLI is a good first choice if you already have it installed. Open a terminal where it’s set up and paste this to it directly, letting it configure itself:
Please connect yourself to my palaia hub as an MCP server:http://palaia.local:8420/mcp/defaultSend the header "Authorization: Bearer <paste-your-token>" with every request.Then run a test recall and tell me what you found.(Replace <paste-your-token> with the token the dashboard shows when you
click Issue token for this tool on its connect page — it is shown once,
and every request needs it. Replace the address with your own hub’s, too:
the dashboard’s connect page shows the exact one, starting with the address
you opened it at — for a server set up with the get.palaia.ai command
usually http://palaia/.) It should report
back that the connection worked and that your memory is currently empty —
that’s expected, you haven’t saved anything yet.
Now give it something worth keeping:
We’re calling this project “Fieldnotes.” Commit messages start with the ticket number in brackets, like
[FN-42] fix the export bug. We settled on that after a PR got merged without one and nobody could tell which release it belonged to.
A well-connected tool saves that on its own, in the same turn, without you asking it to. If it doesn’t, you can ask directly: “please save that.”
3. Connect a second, different AI tool
Section titled “3. Connect a second, different AI tool”This is the part that matters — a memory only one tool uses is just that tool’s notes. Pick a different tool from the same list — Codex is a natural second if the first one was Claude Code, or use anything else on the list. Same pattern: paste the same kind of self-configuring message (each tool’s page has its own exact wording), swapping in your address.
Then ask it something the first tool was never told directly:
What do you know about the Fieldnotes project’s commit message format?
If the connection worked, it answers correctly — [FN-42] fix the export bug — even though this tool never saw that conversation. It found it
because the first tool saved it, and both tools were pointed at the same
memory.
4. What just happened
Section titled “4. What just happened”Nothing was copied between the two tools. There is one memory; both tools read from and wrote to it directly, over their own connection, at whatever moment each one needed to. Add a third tool tomorrow and it starts already knowing everything the first two taught it.
From here:
- Your memory explains what’s actually happening when something gets saved, and how to look through it yourself.
- Teach it to look things up and save things on its own (each tool’s connect page has this section) makes the behavior above automatic rather than something you have to ask for every time.
- Troubleshooting & FAQ covers what to check if step 2 or 3 didn’t go the way this page describes.