Skip to content

Troubleshooting & FAQ

This page covers real, observed issues — not a generic “check your connection” list. If something else goes wrong, the exact behavior matters more than a guess, so start with what you actually saw.

On the machine that runs your hub, one command checks the whole hub: your memories, search, connected AI tools, the configuration and the storage. It says what it found and what to do about each finding:

Terminal window
palaia doctor

With --fix it also performs the repairs that cannot lose anything, such as clearing a lock left behind by a crash or rebuilding search from your notes, and then reports what it did. In the container, run it as docker exec palaia palaia doctor.

“Sign-in fails immediately, every time, for every AI tool”

Section titled ““Sign-in fails immediately, every time, for every AI tool””

If your hub runs somewhere that routes its own outbound internet traffic through a company or organizational proxy, the built-in safety check that fetches a signing-in tool’s published details can mistake that proxy for a risk and refuse to fetch them — so every sign-in fails at the very first step, for every tool, not just one.

If that describes your setup, set this in your hub’s environment before starting it:

Terminal window
FASTMCP_SSRF_TRUST_PROXY=1

Only set this if you actually run behind a proxy you trust — it tells the safety check to trust the one already configured in your environment, rather than turning the check off generally.

“ChatGPT can look things up but won’t save anything”

Section titled ““ChatGPT can look things up but won’t save anything””

This is ChatGPT’s own rule, not something palaia’s setup controls: a Plus/Pro account gets a read-only connection — lookups work exactly the same either way — while write access needs a Business, Enterprise or Edu workspace. The ChatGPT connect page says this inline so it isn’t a surprise partway through setup.

“palaia.local doesn’t resolve on my network”

Section titled ““palaia.local doesn’t resolve on my network””

First check the address: for a Docker install it is http://palaia.local:8420/, with the port — without :8420 the browser knocks on a door of the machine that palaia does not answer at. If that still doesn’t open, it is expected in a couple of common setups: Docker Desktop on macOS/Windows runs containers inside a hidden VM that this kind of network announcement can’t reach past; a container started with -p 8420:8420 on Linux (rather than the Linux command on Install it, which shares the machine’s own network) can’t reach past Docker’s own network either; and some routers keep the name from crossing between separate networks, such as a guest Wi-Fi. See Install it for exactly which setups it works on. It’s always a convenience, never a requirement — your hub’s regular address (shown in its startup log, and in the dashboard) always works. On a server set up with the get.palaia.ai command, use the address the setup printed instead (usually http://palaia/); that server is reached over your tailnet, not by this name.

“How do I know a save actually happened?”

Section titled ““How do I know a save actually happened?””

Ask the AI tool directly — a well-connected one tells you in one short line what it saved, right after saving it. To check independently, open your memory in the dashboard under Memories: brand-new saves show up in its inbox folder first (listed at the top of the tree), before a quiet background step files them into a proper note — see Your memory for what that step does. If something you expected to be saved isn’t showing up anywhere, the tool likely didn’t call the save action at all — Connect your AI’s “teach it to look things up and save things on its own” section for that tool is worth revisiting.

“Two AI tools I connected don’t seem to know the same things”

Section titled ““Two AI tools I connected don’t seem to know the same things””

Check that both are actually pointed at the same memory — if you kept more than one (work and personal, say), it’s easy to connect one tool to each by mistake. The dashboard’s AI apps page shows exactly which memory each connection reaches.

The project’s issue tracker is the right place for anything not covered here: github.com/byte5ai/palaia.