Docs

Troubleshooting

Your assistant will not connect, cannot see your wiki, or a page you expected never arrived. Nine fixes, most of them a setting rather than a fault, and how to reach a person if none of them match.

Your assistant will not connect

Check the address first. ChatGPT, Claude and the terminal assistants each have their own:

ChatGPT:

https://my.eigma.app/api/mcp-chatgpt

Claude:

https://my.eigma.app/api/mcp-claude

Claude Code and Codex:

https://my.eigma.app/api/mcp

In ChatGPT, Developer mode has to be on under Settings then Plugins before you can add a plugin of your own at all.

If the connection was added but keeps dropping, or the assistant reports that it was interrupted, try the same request once more. These interruptions happen between the assistant and our server and they clear on their own. If it keeps failing for more than a few minutes, tell us and include the assistant you were using.

The assistant never used Eigma

Say Eigma by name: “save this conversation to my Eigma wiki” works where “save this” often does not.

In Claude there is a second switch that is easy to miss. A connector you added is not automatically on for a given chat. Open the chat, click the + at the bottom left, choose Connectors, and turn Eigma on there. Eigma looking connected while nothing happens is almost always this.

What it produced is not good enough

If a reorganise stops halfway, pages land in odd chapters, or what comes back is thinner than the conversation was, the cause is usually the model rather than Eigma. The fast, lightweight modes answer in a single pass, and most of what you ask Eigma to do takes several.

Switch your assistant to its reasoning model rather than its quickest one. The middle tier is usually enough. The lighter models do work, the results are just weaker.

For a long job, surveying a whole field or restructuring across many chapters, use Claude’s Cowork or ChatGPT’s Work mode. Both are built for work that runs over several steps.

Nothing is lost while you experiment. Whatever the assistant sent is still waiting in your inbox, so you can reject it and ask again on a stronger model without touching your wiki. The FAQ says the same in one line.

It connects, but your wiki looks empty

Your assistant reads your cloud backup, not the copy in your browser. So a wiki that only ever existed on one device, without an account, is invisible to it.

Open Settings then Account, sign in, and let it sync once. Pages you add after that are included the next time it syncs.

A page you expected never arrived

Nothing joins your wiki on its own, by design. What your assistant sends waits in Ingest, shown next to the page as it stands today, and it joins your wiki only when you press Add to wiki. If you cannot find something your assistant said it saved, look there first.

If the inbox is empty too, check that you were signed in to the same account the assistant is connected to. Pages go to whichever account answered the sign in prompt during setup.

A change says a page “is locked, so its text was not changed”

A locked page is never overwritten, including by something you accepted. When a change would have touched one, Eigma tells you which page it was and leaves the whole change waiting in your inbox rather than applying half of it.

You have two ways out: unlock the page yourself and accept the change again, or use Unlock once and accept, which unlocks it, applies the change, and asks whether to leave it unlocked afterwards.

You signed in with Google or GitHub and now want a password

Use forgot password on the sign in screen and enter the same address. You will get a link that lets you set one. Both ways of signing in then work on the same account.

Disconnecting an assistant

Open Settings then Connected assistants, and use Disconnect next to the one you want to remove. Its access ends immediately, and it can no longer read your wiki or send anything to your inbox. Nothing already in your wiki is affected.

Deleting your data

Open Settings then Account and use the delete option at the bottom. It asks you to type a confirmation word first, because it cannot be undone. That removes your backup, anything queued for your review, any pending change drafts, and every assistant’s authorisation.

The data deletion page explains what is removed and what a deletion cannot reach.

Still stuck

The quickest way to reach us is inside Eigma itself. Open Settings, then Feedback, say what happened, and press Send feedback. Leave your address in the optional email field if you would like an answer. Ideas and requests are as welcome there as bug reports, and we read all of it.

You can also write to support@eigma.app. Either way it reaches the person who builds Eigma directly, and you will normally hear back within a few days. English or German is fine.

It helps a great deal if you include:

  • what you asked your assistant to do, in the words you used
  • what happened instead, including any message you saw
  • which assistant, and whether you were on a phone or a computer
  • whether you were signed in to Eigma at the time

Security concerns go to the email address rather than the feedback box. Say so in the subject line and we will look at it first.